Self-adaptive virtual synchronous generator control method of energy storage converter

Abstract

The invention discloses a self-adaptive virtual synchronous generator control method of an energy storage converter, and relates to the technical field of self-adaptive adjustment. The method comprises the following steps: S1, generating an electrical comprehensive state set based on three-phase voltage and three-phase current of a grid-connected point; s2, constructing a bearing trigger function and evaluating to generate a bearing trigger identifier; s3, obtaining a bearing fault diagnosis result based on the constructed diagnosis sample sequence; s4, according to the bearing fault diagnosis result and the diagnosis sample sequence, an adjustment parameter set is obtained, and an adjustment instruction is issued to the virtual synchronous generator; s5, updating the virtual power angle and the frequency reference; extracting a current vector and remodeling an original voltage reference according to an admittance parameter group; adjusting a power droop target and power angle evolution of a virtual synchronous generator of the energy storage converter according to a projection active instruction and a projection reactive instruction obtained through projection operation; the self-adaptive virtual synchronous generator control method realizes self-adaptive virtual synchronous generator control which keeps voltage shaping and virtual power angle continuity under current bearing constraint.

Description

Technical Field

[0001] This invention relates to the field of adaptive regulation technology, and more specifically, to an adaptive virtual synchronous generator control method for an energy storage converter. Background Technology

[0002] In existing energy storage grid-connected technologies, virtual synchronous generator (VSR) control methods have been widely adopted to simulate the inertia and damping characteristics of traditional synchronous machines, thereby improving grid stability after the integration of new energy sources. Under normal operating conditions, this method achieves stable regulation of grid voltage and frequency through the evolution of virtual power angle and power droop characteristics, and can enhance the support capability of weak grids to a certain extent. However, with the increase in the capacity of energy storage devices and the promotion of multi-machine parallel systems, energy storage converters are more prone to current over-limit problems when approaching their current carrying capacity. At this time, the forming capability and phase continuity of the VSR face greater challenges. Especially under conditions of severe grid voltage fluctuations, rapid changes in line impedance, or parallel operation, the converter not only needs to avoid overcurrent protection triggering, but also needs to ensure continuous output of voltage amplitude and phase to maintain system stability.

[0003] Existing technologies still have significant shortcomings in handling current-carrying constraints. Specifically, current-limiting measures often directly and rigidly reduce the voltage reference or power command, easily leading to loss of voltage shaping and a sharp drop in the voltage amplitude at the grid connection point, thus compromising the voltage support function. Furthermore, the virtual power angle lacks a continuous adjustment mechanism during current-limiting, making it prone to abrupt changes, resulting in frequency instability or phase mismatch between multiple units. These problems are particularly prominent in weak grid, fault ride-through, and multi-unit parallel operation scenarios, directly impacting the support capacity of energy storage converters and the safety of grid operation.

[0004] Therefore, there is an urgent need for an adaptive control method that can maintain stable voltage shaping and achieve continuous virtual power angle adjustment under current carrying constraints to ensure the reliable operation of energy storage systems under complex grid conditions. In view of this, this invention proposes an adaptive virtual synchronous generator control method for energy storage converters to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: an adaptive virtual synchronous generator control method for an energy storage converter, comprising:

[0006] S1: Based on the three-phase voltage and three-phase current at the grid connection point, a comprehensive electrical state set is obtained; the comprehensive electrical state set includes voltage phase, current phase, current amplitude, instantaneous active power, and instantaneous reactive power;

[0007] S2: Construct a load-bearing trigger function based on the electrical integrated state set and calculate the load-bearing trigger function value; when the load-bearing trigger function value is greater than the preset load-bearing trigger threshold, generate a load-bearing trigger identifier;

[0008] S3: When a load triggering flag is detected, the electrical comprehensive status set within a preset time window is constructed into a diagnostic sample sequence and input into the load fault diagnosis model, and the load fault diagnosis result is output; the load fault diagnosis result includes voltage shaping loss flag and virtual power angle jump flag;

[0009] S4: When only voltage shaping loss flag or only virtual power angle jump flag exists, the load fault diagnosis result value obtained by quantifying the load fault diagnosis result and the diagnosis sample sequence are input into the parameter intelligent setting model to obtain a set of adjustment parameters including static admittance share, dynamic admittance share, time constant, command projection intensity, power angle change rate and phase angle mitigation term; the set of adjustment parameters is converted into adjustment commands to adjust the virtual synchronous generator of the energy storage converter;

[0010] S5: When both voltage shaping loss flag and virtual power angle jump flag exist, update the virtual power angle and frequency reference based on the diagnostic sample sequence; perform synchronous calculation and line equivalent impedance estimation on the three-phase voltage and three-phase current at the grid connection point to construct the current constraint domain; extract the current vector and reshape the original voltage reference according to the admittance parameter group to obtain the constrained voltage reference; adjust the power droop target and power angle evolution rate of the virtual synchronous generator of the energy storage converter based on the projected active power command and projected reactive power command obtained by the projection calculation.

[0011] Furthermore, the implementation methods of S5 include:

[0012] The virtual power angle and frequency reference are updated based on the diagnostic sample sequence;

[0013] The three-phase voltage and three-phase current at the grid connection point are obtained by using recursive least squares or extended Kalman filtering algorithm to obtain the estimated value of the line equivalent impedance.

[0014] A current constraint domain is constructed based on the estimated values ​​of the three-phase voltage at the grid connection point, the equivalent impedance of the line, and the preset current carrying capacity threshold.

[0015] The positive sequence components of the three-phase current at the grid connection point are extracted and the coordinates are transformed to obtain the current vector;

[0016] Based on the current vector and the admittance parameter set, the original voltage reference is constrained and reshaped to obtain a constrained voltage reference; the admittance parameter set includes the static admittance fraction, the dynamic admittance fraction, and the time constant.

[0017] Projection calculations are performed on the active and reactive power commands based on the current constraint domain and the constrained voltage reference to obtain the projected active and reactive power commands.

[0018] The power droop target is adjusted based on the projected active power command and the projected reactive power command, and the power angle evolution rate is adjusted according to the power droop target.

[0019] Furthermore, methods for updating the virtual power angle and frequency reference based on diagnostic sample sequences include:

[0020] A small window smoothing process is applied to the instantaneous active and reactive power in the diagnostic sample sequence to obtain the active power measurement value and the reactive power measurement value; the instantaneous active power deviation and the instantaneous reactive power deviation are calculated based on the preset instantaneous active power target value and instantaneous reactive power target value.

[0021] Zero-sensitivity zone, amplitude limiting, and rate of change constraints are applied to the instantaneous active power deviation and instantaneous reactive power deviation to obtain the effective amount of deviation; the effective amount of deviation is proportionally converted based on the frequency correction caused by the unit active power deviation to obtain the frequency increment; the frequency increment is subjected to bandwidth limiting and noise suppression processing to obtain the target frequency reference.

[0022] Discrete integration is performed on the target frequency reference over the sampling period to obtain the target virtual power angle;

[0023] Update the current frequency reference to the target frequency reference; update the current virtual power angle to the target virtual power angle.

[0024] Furthermore, the method for constructing the current constraint domain is as follows:

[0025] Based on the estimated values ​​of the three-phase voltage at the grid connection point, the equivalent impedance of the line, and the preset current carrying capacity threshold, a mapping relationship from instantaneous active and reactive power to current amplitude is established. For any set of instantaneous active and reactive power, the corresponding current amplitude is calculated, and the solution set whose current amplitude does not exceed the preset current carrying capacity threshold is defined as the current constraint domain.

[0026] Furthermore, methods for constraining and reshaping the original voltage reference to obtain a constrained voltage reference include:

[0027] Multiplying the current vector by the static admittance component yields the steady-state compensation voltage component.

[0028] A current proximity factor is constructed based on the current amplitude and a preset current carrying threshold; a piecewise linear barrier function is constructed based on the current proximity factor and a preset current proximity threshold.

[0029] The piecewise linear barrier function output by the piecewise linear barrier function is multiplied by the dynamic component of the admittance to obtain the dynamic compensation voltage component, and then filtered by the time constant to smooth the response.

[0030] The constrained voltage reference is obtained by superimposing the steady-state compensation voltage component and the dynamic compensation voltage component and then subtracting them from the original voltage reference.

[0031] Furthermore, the method of adjusting the power droop target based on the projected active power command and the projected reactive power command, and adjusting the power angle evolution rate according to the power droop target, includes:

[0032] Based on the projected active power command and the projected reactive power command, combined with the frequency-active power droop characteristic and the voltage-reactive power droop characteristic, the corresponding active power target point and reactive power target point are calculated respectively, and the active power target point and reactive power target point are merged to form the first power droop target.

[0033] The current actual active power is compared with the active power in the power droop target to obtain the projected active power deviation; the projected active power deviation is converted into a frequency correction amount according to the preset active power-frequency droop coefficient, and the first power angle evolution rate is formed under the action of the preset inertia coefficient and damping coefficient.

[0034] Adjust the current power droop target to the first power droop target, and adjust the current power angle evolution rate to the first power angle evolution rate.

[0035] Furthermore, the method for obtaining the electrical integrated state set includes:

[0036] The three-phase voltage and three-phase current at the grid connection point are sampled with a unified clock and timestamp aligned. Sensor drift and noise are eliminated based on anti-aliasing filtering and DC bias removal. The measurement delay of the voltage and current channels is compensated to obtain synchronized voltage and current data sequences.

[0037] The three-phase voltage and three-phase current at the grid connection point are transformed by two-phase orthogonal transformation to construct voltage orthogonal components and current orthogonal components; the voltage orthogonal components enter the phase synchronization unit to obtain the voltage phase; the current phase is synchronously obtained according to the angle relationship between the current orthogonal components and the voltage phase.

[0038] The current amplitude is estimated based on the voltage orthogonal component and the current orthogonal component, and instantaneous fluctuations are suppressed by using a sliding window root mean square or small window smoothing to obtain the current amplitude.

[0039] Instantaneous active power and instantaneous reactive power are determined according to instantaneous power theory;

[0040] The voltage phase, current phase, current amplitude, instantaneous active power, and instantaneous reactive power are constructed into a comprehensive electrical state set.

[0041] Furthermore, the training method for the bearing fault diagnosis model includes:

[0042] A pre-constructed load cell fault diagnosis dataset is constructed, which includes ZD group load cell fault diagnosis data and corresponding load cell fault diagnosis results, where ZD is a positive integer; the load cell fault diagnosis data includes a sequence of diagnosis samples; the load cell fault diagnosis dataset is divided into a training set and a validation set, the training set is used for learning the parameters of the load cell fault diagnosis model, and the validation set is used for real-time monitoring of the generalization performance and overfitting of the load cell fault diagnosis model;

[0043] A deep neural network with a multilayer perceptron as its core is used as the load cell fault diagnosis model. The load cell fault diagnosis data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer uses a nonlinear activation function to extract features, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each load cell fault diagnosis result. Finally, the load cell fault diagnosis result corresponding to the highest probability is taken as the prediction result of the load cell fault diagnosis model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is set: when the prediction accuracy on the validation set reaches or exceeds the prediction accuracy threshold, the load cell fault diagnosis model is determined to have converged and training is terminated.

[0044] Furthermore, the training method for the intelligent parameter setting model includes:

[0045] A pre-constructed intelligent parameter setting dataset is constructed, which includes CS group intelligent parameter setting data and the set of adjustment parameters corresponding to the CS group intelligent parameter setting data, where CS is a positive integer; the intelligent parameter setting data includes numerical values ​​carrying fault diagnosis results and diagnostic sample sequences; the intelligent parameter setting dataset is divided into a training set and a validation set, the training set is used for parameter learning of the intelligent parameter setting model, and the validation set is used to monitor the generalization performance and overfitting degree of the intelligent parameter setting model in real time.

[0046] A deep neural network with a multilayer perceptron as its core is used as the intelligent parameter setting model. The intelligent parameter setting data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer uses a nonlinear activation function to extract features, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each set of adjustment parameters. Finally, the set of adjustment parameters corresponding to the highest probability is taken as the prediction result of the intelligent parameter setting model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is set: when the prediction accuracy on the validation set reaches or exceeds the prediction accuracy threshold, the intelligent parameter setting model is determined to have converged and training is terminated.

[0047] Compared with existing technologies, the adaptive virtual synchronous generator control method for energy storage converters proposed in this invention has the following technical effects and advantages:

[0048] In this invention, a complete adaptive virtual synchronous generator control link is formed through the organic integration of operational data monitoring, load trigger determination, load fault diagnosis, and intelligent adjustment modules. Specifically, an electrical comprehensive state set is constructed based on the three-phase voltage and three-phase current at the grid connection point, which can reflect the phase relationship and power exchange of the grid connection interface in real time, providing accurate data support for subsequent constraint control. The load trigger function is used to determine the current load state, ensuring that constraints and diagnoses are only initiated when the operation approaches the capacity boundary, avoiding unnecessary intervention under normal operation. Furthermore, under load conditions, the load fault diagnosis model can quickly identify two key failure modes: voltage shaping loss and virtual power angle jump, enabling targeted diagnosis of potential risks.

[0049] Driven by the diagnostic results, this application designs an intelligent adjustment mechanism. When a single failure mode occurs, the set of adjustment parameters is output based on the parameter intelligent setting model, and the static share of admittance, dynamic share of admittance, time constant, command projection intensity, power angle change rate, and phase angle mitigation term are adaptively adjusted to achieve voltage shaping and power angle continuity under current limiting conditions. When the risk of bearing continues or concurrent failure modes occur, the command value is ensured to fall within the feasible region through constrained voltage reference generation and power command projection, and the power droop target and power angle evolution rate are dynamically corrected to ensure that the response of the virtual synchronous generator is consistent with the actual power demand of the power grid.

[0050] In summary, this invention not only solves the problems of voltage shaping loss, virtual power angle jump, and parallel coordination imbalance that are prone to occur under current-limited conditions in the prior art, but also significantly improves the stability of the system under weak grid, large disturbance, and fault ride-through conditions, realizing the safe operation and stable support of the energy storage converter within the current carrying capacity constraint, and has outstanding engineering application value and promotion significance. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of an adaptive virtual synchronous generator control system for an energy storage converter according to Embodiment 1 of the present invention;

[0052] Figure 2 This is a flowchart of an adaptive virtual synchronous generator control method for an energy storage converter according to Embodiment 2 of the present invention;

[0053] Figure 3 This is a flowchart illustrating the implementation method of the second intelligent adjustment module in Embodiment 1 of the present invention;

[0054] Figure 4This is a flowchart of the method for updating virtual power angle and frequency reference based on diagnostic sample sequence according to Embodiment 1 of the present invention;

[0055] Figure 5 This is a flowchart of the method for constraining and reshaping the original voltage reference to obtain a constrained voltage reference according to Embodiment 1 of the present invention;

[0056] Figure 6 This is a flowchart of the method for obtaining the electrical integrated state set according to Embodiment 1 of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0058] Example 1

[0059] Please see Figure 1 As shown in the figure, this embodiment discloses an adaptive virtual synchronous generator control system for an energy storage converter, including an operation data monitoring module, a load triggering determination module, a load intelligent diagnosis module, a first intelligent adjustment module, and a second intelligent adjustment module. Each module is connected by wires and / or wirelessly to realize data transmission.

[0060] The running data monitoring module performs synchronous calculations based on the three-phase voltage and three-phase current at the grid connection point to obtain the electrical comprehensive state set; the electrical comprehensive state set includes voltage phase, current phase, current amplitude, instantaneous active power and instantaneous reactive power.

[0061] like Figure 6 As shown, the method for obtaining the electrical integrated state set includes:

[0062] The three-phase voltage and three-phase current at the grid connection point are sampled with a unified clock and timestamp aligned. Sensor drift and noise are eliminated based on anti-aliasing filtering and DC bias removal. The measurement delay of the voltage and current channels is compensated to obtain synchronized voltage and current data sequences.

[0063] The three-phase voltage and three-phase current at the grid connection point are transformed into orthogonal voltage and current components through two-phase orthogonal transformation. The orthogonal voltage components are then entered into the phase synchronization unit to obtain the voltage phase. Based on the angle relationship between the orthogonal current components and the voltage phase, the current phase is synchronously obtained. For scenarios with unbalanced three-phase loads or harmonics in the distribution network, positive sequence component extraction and narrowband filtering are first performed on the orthogonal voltage and current components before phase and amplitude estimation is performed to ensure the consistency of the fundamental frequency in subsequent measurements.

[0064] It should be noted that the phase synchronization unit is a core unit in the electrical measurement and control system, used to extract and synchronize the voltage signal phase. Its main function is to analyze the quadrature components of the three-phase voltage at the grid connection point, calculate the instantaneous phase of the system's fundamental voltage in real time, and ensure that it remains consistent with the grid reference phase.

[0065] The current amplitude is estimated based on the voltage orthogonal component and the current orthogonal component, and instantaneous fluctuations are suppressed by using a sliding window root mean square or small window smoothing to obtain the current amplitude.

[0066] Instantaneous active power and instantaneous reactive power are determined according to instantaneous power theory;

[0067] The voltage phase, current phase, current amplitude, instantaneous active power, and instantaneous reactive power are constructed into a comprehensive electrical state set.

[0068] It should be noted that, in this invention, when constructing the electrical integrated state set, it is required that the three-phase voltage and three-phase current be measured at the same point, using the same time reference. The electrical integrated state set obtained based on the grid connection point measurements can accurately reflect the phase relationship and power exchange of the grid connection interface, thereby ensuring that the constrained voltage reference, power command projection, and power angle evolution rate adjustment correspond one-to-one with the grid connection physical constraints, avoiding mismatch between the target and the constraints.

[0069] If sampling is not performed at the grid connection point, systematic biases will be introduced. Specifically, the converter side before the LCL filter contains switching harmonics and dead-zone distortion, and the LCL filter and line impedance generate additional phase shifts and voltage drops, causing the active power, reactive power, and current amplitudes inferred from the measurements at this location to be inconsistent with the actual values ​​at the grid connection point. Measurements at the midpoint of the LCL filter or on the capacitor side are affected by resonance and negative sequence / higher harmonic amplification, making impedance estimation and phase angle synchronization prone to distortion. In addition, when the measurement location is downstream of the energy storage converter or far from the feeder, the obtained voltage and current will be superimposed with multiple biases. First, the reactive voltage drop generated by the transformer and line causes the voltage amplitude and phase to change relative to the grid connection point. Second, the transformation ratio and phase errors of the current transformer and voltage transformer cause the measured values ​​to deviate from the true values. Third, the time delay and asynchrony introduced by the communication link and sampling process cause the voltage and current to no longer correspond to the same moment. The combined effect of the above three factors will cause the calculated active power, reactive power and current amplitude to be inconsistent with the actual grid connection point, thereby affecting the current constraint determination, power angle synchronization and parallel coordination.

[0070] In practical engineering applications, if measurements cannot be directly taken at the grid connection point due to limitations, the measurements taken at the non-grid connection point can be inversely calculated to the grid connection point based on known filter parameters and the equivalent impedance of the line. Compensation and reconstruction should then be performed on sampling delay, positive / negative sequence components, and harmonic components. Only after this reconstruction can the comprehensive electrical state set be generated. Otherwise, it will be difficult to ensure consistency between key aspects of this invention, such as constrained voltage reference and power command projection, and the physical boundaries of the grid connection, thereby weakening the stability and feasibility of the solution.

[0071] The load trigger determination module constructs a load trigger function based on the electrical integrated state set and calculates the load trigger function value; when the load trigger function value is greater than the preset load trigger threshold, a load trigger identifier is generated as the start signal for subsequent constraint control.

[0072] For example, in the technical solution of the present invention, for a weak grid sensitive energy storage converter, the load triggering threshold can be set to 1.00; for a parallel sensitive energy storage converter, the load triggering threshold can be set to 1.08.

[0073] The method for obtaining the value of the bearer trigger function includes:

[0074] ;

[0075] in, To carry the trigger function, This is to carry the trigger function value. The current carrying ratio is the ratio of the current amplitude to a preset current carrying threshold. For example, in this application, the current carrying threshold is set to 350 amps. For instantaneous reactive power in the electrical comprehensive state set, For instantaneous active power in the electrical comprehensive state set, It is a very small positive number, that is, it approaches 0. The voltage phase in the electrical integrated state set, The current phase in the electrical integrated state set, It is a constant. This is the maximum apparent power of the energy storage converter device itself. , and For the corresponding weighting coefficients, , and The sum of the three is 1. For example, in this application, it can be... The value is set to 0.3, The value is set to 0.3, The value is set to 0.4.

[0076] To avoid calculation deviations or confusion of physical meanings caused by inconsistent units of physical quantities in the formula for calculating the load trigger function value, the parameters involved in the calculation of the load trigger function value in this application have been dimensionless and normalized. For example, the numerical range of all parameters in the formula for calculating the load trigger function value is limited to [0, 1] to ensure the scientific nature of the formula structure and the comparability of the results.

[0077] It should be noted that the direct physical constraint of grid connection over-limit comes from the current carrying capacity. Therefore, the current carrying capacity ratio is constructed and used as the main term. The current carrying capacity ratio takes a value of no more than 1 within the safe zone and a value of more than 1 when it exceeds the carrying capacity threshold. Indicates the proportion of reactive power. This represents phase angle normalization. The main term is linearly amplified using two monotonic, bounded indicators: reactive power ratio and phase angle normalization. The weighting coefficients are... and Non-negative, thus taking into account the need for increased risk assessment for the same current when power is mainly reactive or the phase angle is large; and the extremely small positive number when instantaneous active and instantaneous reactive power approach 0. To prevent the denominator from being zero and the function from continuously returning to zero, thus avoiding false triggers. Additionally, the view is overlaid on the occupancy rate item. The apparent occupancy rate is used to characterize the saturation proximity of energy storage converter devices in the apparent capacity dimension. The apparent occupancy rate term is asynchronous with the main term but complementary. Specifically, when the voltage is high or the topology changes cause the current to approach the current carrying threshold while the apparent power is close to the capacity limit, such as in the case of high voltage and high power factor, relying solely on the current may underestimate the risk; adding the apparent occupancy rate term can reflect the capacity saturation trend in advance and avoid double-counting with the main term.

[0078] The intelligent diagnostic module for bearings, when detecting a bearing trigger flag, constructs a diagnostic sample sequence from the set of comprehensive electrical states within a preset time window, inputs it into the bearing fault diagnosis model, and outputs the bearing fault diagnosis result; the bearing fault diagnosis result includes voltage shaping loss flag and virtual power angle jump flag.

[0079] In an exemplary embodiment of this application, the preset time window is constructed centered on the moment when the trigger flag is detected. Continuous measurement data from one unit of time before and one unit of time after that moment are selected and combined as a diagnostic sample sequence. The unit of time can be preset to 3 minutes, or it can be configured and adjusted according to the characteristics of power grid fluctuations and the dynamic response capability of the device.

[0080] The training method for the load-bearing fault diagnosis model includes:

[0081] A pre-constructed load cell fault diagnosis dataset is constructed, which includes ZD group load cell fault diagnosis data and corresponding load cell fault diagnosis results, where ZD is a positive integer; the load cell fault diagnosis data includes a sequence of diagnosis samples; the load cell fault diagnosis dataset is divided into a training set and a validation set, the training set is used for learning the parameters of the load cell fault diagnosis model, and the validation set is used for real-time monitoring of the generalization performance and overfitting of the load cell fault diagnosis model;

[0082] A deep neural network with a multilayer perceptron as its core is used as the load cell fault diagnosis model. Load cell fault diagnosis data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer extracts features using a nonlinear activation function, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each load cell fault diagnosis result. Finally, the load cell fault diagnosis result corresponding to the highest probability is taken as the prediction result of the load cell fault diagnosis model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is implemented: when the prediction accuracy on the validation set reaches or exceeds the prediction accuracy threshold, the load cell fault diagnosis model is considered to have converged and training is terminated. For example, the prediction accuracy threshold can be set to 95% in this application.

[0083] It should be noted that when the load trigger determination module generates a load trigger identifier, it indicates that the energy storage converter device has entered the current load boundary, thus activating the intelligent diagnostic module for load fault diagnosis. Specifically, generating the load trigger identifier before load fault diagnosis allows for the gating of diagnostics for voltage shaping loss and virtual power angle jumps during high-risk phases, reducing false alarm rates and computational burden in the steady-state phase, and making adaptive adjustment more targeted. The load status, as a trigger precondition consistent with the equipment's physical boundary, ensures clear causal evidence and sufficient evidence for control intervention, reducing jitter and secondary oscillations caused by random fluctuations, and avoiding unnecessary performance degradation due to frequent interventions within normal margins.

[0084] Using the load-bearing trigger flag as a prerequisite, the electrical comprehensive state set within a preset time window is used as input to the load-bearing fault diagnosis model. Data shaping and window smoothing are performed under the same time reference, and the load-bearing fault diagnosis results are output. The load-bearing fault diagnosis results include voltage shaping loss flags and virtual power angle jump flags. By triggering diagnosis under load conditions, false alarms and invalid calculations in the daily steady-state stage can be reduced, and the two most relevant failure modes can be focused on during the risk accumulation stage, enabling rapid identification of shaping consistency problems and phase continuity problems caused by current limiting. If either the voltage shaping loss flag or the virtual power angle jump flag is present in the load-bearing fault diagnosis results, an adaptive adjustment flag is generated and subsequent parameter adjustments are linked; if neither the voltage shaping loss flag nor the virtual power angle jump flag is present, the existing control parameters are maintained and monitoring continues, thereby ensuring safety while avoiding unnecessary disturbances.

[0085] The first intelligent adjustment module, when only voltage shaping loss indicator or only virtual power angle jump indicator exists, inputs the load fault diagnosis result value obtained by quantifying the load fault diagnosis result and the diagnosis sample sequence into the parameter intelligent setting model to obtain a set of adjustment parameters including static admittance share, dynamic admittance share, time constant, command projection intensity, power angle change rate and phase angle mitigation term; and converts the set of adjustment parameters into adjustment commands to adjust the virtual synchronous generator of the energy storage converter.

[0086] The training method for the intelligent parameter setting model includes:

[0087] A pre-constructed intelligent parameter setting dataset is constructed, which includes CS group intelligent parameter setting data and the set of adjustment parameters corresponding to the CS group intelligent parameter setting data, where CS is a positive integer; the intelligent parameter setting data includes numerical values ​​carrying fault diagnosis results and diagnostic sample sequences; the intelligent parameter setting dataset is divided into a training set and a validation set, the training set is used for parameter learning of the intelligent parameter setting model, and the validation set is used to monitor the generalization performance and overfitting degree of the intelligent parameter setting model in real time.

[0088] A deep neural network with a multilayer perceptron as its core is used as the intelligent parameter setting model. The intelligent parameter setting data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer extracts features using a nonlinear activation function, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each set of adjustment parameters. Finally, the set of adjustment parameters corresponding to the highest probability is taken as the prediction result of the intelligent parameter setting model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is implemented: when the prediction accuracy on the validation set reaches or exceeds a prediction accuracy threshold, the intelligent parameter setting model is considered to have converged and training is terminated. For example, the prediction accuracy threshold can be set to 95% in this application.

[0089] It should be noted that the static admittance share characterizes the fixed voltage drop compensation ratio applied to the voltage reference by the virtual admittance under steady-state operation or slow disturbances. The static admittance share mainly reflects the reference impedance characteristics of the system. By injecting a compensation amount proportional to the current vector into the voltage reference, the energy storage converter exhibits damping and power distribution capabilities over long time scales. The dynamic admittance share characterizes the dynamic response component added to the virtual admittance when there are rapid disturbances or when the current approaches the current carrying threshold. The dynamic admittance share dynamically adjusts with changes in the current amplitude. Its main function is to rapidly increase the compensation amplitude to the voltage reference when the current rises sharply, thereby suppressing further current increases. It can quickly limit and dynamically suppress the current rise caused by short-term large disturbances without disrupting the voltage shaping characteristics. The time constant characterizes the rate of change and filtering characteristics of the dynamic admittance response. When the current changes abruptly, the time constant determines the rise and decay rate of the dynamic admittance share, avoiding voltage reference spikes or oscillations caused by excessively rapid changes in the compensation amount.

[0090] The command projection intensity is used to characterize the strength and speed at which candidate active / reactive targets are projected into the current constraint domain. When the load risk increases, the command projection intensity is increased based on the load trigger function value, making the boundary projection of active and reactive targets more complete and convergent faster, thereby ensuring that the current obtained by target mapping does not exceed the current load threshold.

[0091] The rate of change of power angle is used to characterize the rate at which the virtual power angle evolves over time. Its adjustment essentially constrains the upper limits of the amplitude and rate of change of phase evolution to ensure phase continuity and dynamic controllability. When virtual power angle jumps occur, the upper limit of the rate of change of power angle is tightened and a rate-limiting factor is applied. Combined with a phase angle mitigation term, this allows the virtual power angle to converge smoothly at a predictable slope, avoiding phase jumps or secondary oscillations caused by excessively rapid frequency correction. The phase angle mitigation term is used to correct the virtual power angle in a controlled and gradual manner, thereby avoiding phase jumps caused by excessively rapid frequency correction or mode switching. Essentially, it constructs a bounded, rate-limited, and adjustable weighted additional correction based on the phase deviation between the grid connection point voltage phase and the virtual power angle, and superimposes it in parallel with the rate of change of power angle. This allows the virtual power angle to slowly align with the grid reference or parallel-connected phase along a predictable slope, ensuring the continuity and controllability of voltage shaping and power angle evolution of the virtual synchronous generator in the energy storage converter.

[0092] The second intelligent adjustment module updates the virtual power angle and frequency reference based on the diagnostic sample sequence when both the voltage shaping loss flag and the virtual power angle jump flag are present; it performs synchronous calculations and line equivalent impedance estimation on the three-phase voltage and three-phase current at the grid connection point to construct a current constraint domain; it extracts the current vector and reshapes the original voltage reference according to the admittance parameter group to obtain the constrained voltage reference; and it adjusts the power droop target and power angle evolution rate of the virtual synchronous generator of the energy storage converter based on the projected active power command and projected reactive power command obtained by the projection calculation.

[0093] like Figure 3 As shown, the implementation method of the second intelligent adjustment module is as follows:

[0094] The virtual power angle and frequency reference are updated based on the diagnostic sample sequence;

[0095] The three-phase voltage and three-phase current at the grid connection point are obtained by using recursive least squares or extended Kalman filtering algorithm to obtain the estimated value of the line equivalent impedance.

[0096] A current constraint domain is constructed based on the estimated values ​​of the three-phase voltage at the grid connection point, the equivalent impedance of the line, and a preset current carrying capacity threshold. The current constraint domain is used to describe the set of instantaneous active and reactive power combinations that ensure the grid-connected current amplitude does not exceed the preset current carrying capacity threshold under the conditions of the estimated values ​​of the three-phase voltage at the grid connection point, the equivalent impedance of the line, and the preset current carrying capacity threshold.

[0097] The positive sequence components of the three-phase current at the grid connection point are extracted and the coordinates are transformed to obtain the current vector; the current vector contains the d-axis component and q-axis component of the current in the orthogonal coordinate system.

[0098] Based on the current vector and the admittance parameter set, the original voltage reference is constrained and reshaped to obtain a constrained voltage reference; the admittance parameter set includes the static admittance fraction, the dynamic admittance fraction, and the time constant.

[0099] It should be noted that the original voltage reference refers to the three-phase voltage target signal generated by the controller of the energy storage converter according to the principle of virtual synchronous generator (VSG) before any constraints or corrections are applied.

[0100] Projection operations are performed on the active and reactive power commands based on the current constraint domain and the constrained voltage reference to obtain projected active and reactive power commands. The active power command reflects the active power that the energy storage converter should output or absorb at the current moment, used to maintain grid frequency stability and achieve energy exchange. The reactive power command reflects the reactive power that the energy storage converter should provide or absorb at the current moment, used to maintain the grid connection point voltage level and improve power quality. The projection operation refers to mapping the original active and reactive power commands to the boundaries of the current constraint domain and the constrained voltage reference. The results obtained after the projection operation are the projected active and projected reactive power commands.

[0101] The power droop target is adjusted based on the projected active power command and the projected reactive power command, and the power angle evolution rate is adjusted according to the power droop target.

[0102] like Figure 4 As shown, the method for updating the virtual power angle and frequency reference based on the diagnostic sample sequence includes:

[0103] A small window smoothing process is applied to the instantaneous active and reactive power in the diagnostic sample sequence to obtain the active power measurement value and the reactive power measurement value. The instantaneous active power deviation and the instantaneous reactive power deviation are calculated based on the preset instantaneous active power target value and instantaneous reactive power target value. The absolute value of the difference between the instantaneous active power target value and the active power measurement value is the instantaneous active power deviation, and the absolute value of the difference between the instantaneous reactive power target value and the reactive power measurement value is the instantaneous reactive power deviation.

[0104] Zero-sensitivity zone, amplitude limiting, and rate of change constraints are applied to the instantaneous active power deviation and instantaneous reactive power deviation to obtain the effective amount of deviation; the effective amount of deviation is proportionally converted based on the frequency correction caused by the unit active power deviation to obtain the frequency increment; the frequency increment is subjected to bandwidth limiting and noise suppression processing to obtain the target frequency reference.

[0105] It should be noted that the zero-sensitivity zone is used to define the range within which the control system does not respond or responds weakly to minute deviations. When the absolute value of the deviation is less than a preset width, the controller does not generate effective adjustments to suppress measurement noise and signal jitter, and to prevent frequent small-amplitude adjustments. For example, in this application, the width of the zero-sensitivity zone can be set to ±0.02pu, which corresponds to an actual voltage deviation of approximately ±1.2V or a frequency deviation of ±0.02Hz.

[0106] Discrete integration is performed on the target frequency reference over the sampling period to obtain the target virtual power angle;

[0107] Update the current frequency reference to the target frequency reference; update the current virtual power angle to the target virtual power angle.

[0108] For example, in a preferred embodiment of this application, assuming a rated power of 500kW, the setting data for instantaneous active power target value and instantaneous reactive power target value are shown in Table 1.

[0109] Table 1. Example of instantaneous active power target value and instantaneous reactive power target value setting data.

[0110]

[0111] The method for constructing the current constraint domain is as follows:

[0112] Based on the estimated values ​​of the three-phase voltage at the grid connection point, the equivalent impedance of the line, and the preset current carrying capacity threshold, a mapping relationship from instantaneous active and reactive power to current amplitude is established. For any set of instantaneous active and reactive power, the corresponding current amplitude is calculated, and the solution set whose current amplitude does not exceed the preset current carrying capacity threshold is defined as the current constraint domain.

[0113] The method for calculating the current amplitude includes:

[0114] ;

[0115] in, YG represents the instantaneous active power, and WG represents the instantaneous reactive power. This represents the effective value of the three-phase voltage at the corresponding grid connection point.

[0116] like Figure 5 As shown, the methods for constraining and reshaping the original voltage reference to obtain a constrained voltage reference include:

[0117] Multiplying the current vector by the static admittance component yields the steady-state compensation voltage component.

[0118] A current proximity factor is constructed based on the current amplitude and a preset current carrying threshold; a piecewise linear barrier function is constructed based on the current proximity factor and a preset current proximity threshold.

[0119] The piecewise linear barrier function output by the piecewise linear barrier function is multiplied by the dynamic component of the admittance to obtain the dynamic compensation voltage component, and then filtered by the time constant to smooth the response.

[0120] The constrained voltage reference is obtained by superimposing the steady-state compensation voltage component and the dynamic compensation voltage component and then subtracting them from the original voltage reference.

[0121] The method for calculating the current proximity factor includes:

[0122] ;

[0123] in, For current proximity factor, The current amplitude, This is a preset current carrying capacity threshold. For example, in this application, the current proximity threshold is... Set it to 0.7.

[0124] The piecewise linear barrier function is:

[0125] ;

[0126] in, For piecewise linear barrier function values, For example, in this application, the corresponding adjustment coefficient can be... Set to 2.

[0127] The method for adjusting the power droop target based on projected active power command and projected reactive power command, and adjusting the power angle evolution rate according to the power droop target, includes:

[0128] Based on the projected active power command and the projected reactive power command, and combined with the frequency-active power droop characteristic and the voltage-reactive power droop characteristic, the corresponding active power target point and reactive power target point are calculated respectively, and the active power target point and reactive power target point are merged to form the first power droop target. This power droop target not only reflects the power regulation task of the energy storage converter under the conditions of grid frequency, grid voltage and energy constraints, but also serves as a benchmark for judging the dynamics of the virtual power angle.

[0129] The current actual active power is compared with the active power in the power droop target to obtain the projected active power deviation; the projected active power deviation is converted into a frequency correction amount according to the preset active power-frequency droop coefficient, that is, the active power-frequency droop coefficient is multiplied by the projected active power deviation to obtain the frequency correction amount, and the first power angle evolution rate is formed under the action of the preset inertia coefficient and damping coefficient.

[0130] It should be noted that the active power-frequency droop factor is a proportional coefficient set in the virtual synchronous generator control to describe the linear relationship between frequency deviation and active power output. For example, if the active power-frequency droop factor is specified as 0.05pu, it means that when the system frequency decreases from 50Hz to 49.5Hz, the converter should increase the active power from 0 to the maximum active power output. In this way, the magnitude of the power angle evolution rate is always driven and corrected by the power droop target, ensuring that the evolution of the virtual power angle reflects the actual power demand of the power grid.

[0131] Adjust the current power droop target to the first power droop target, and adjust the current power angle evolution rate to the first power angle evolution rate.

[0132] It should be noted that the inertia coefficient is used to characterize the inertial resistance of a virtual synchronous generator to frequency changes. This means that by introducing an equivalent inertia term into the power angle evolution equation, the virtual power angle will not immediately fluctuate drastically when subjected to active power deviations or disturbances, but will respond in a smoother manner. A larger inertia coefficient results in a slower rate of frequency change, meaning stronger disturbance rejection and frequency stability, but also reduced sensitivity to dynamic adjustment; a smaller inertia coefficient allows for more flexible frequency adjustment, but weakens disturbance rejection.

[0133] The damping coefficient characterizes the energy dissipation capability of a virtual synchronous generator during power angle oscillation. It means that by introducing a damping term into the power angle evolution equation, the power angle oscillation caused by disturbances can gradually decay, avoiding continuous oscillation or overshoot. A larger damping coefficient results in faster oscillation decay and enhanced stability, but at the cost of some dynamic sensitivity; a damping coefficient that is too small may lead to prolonged power angle oscillation time or even instability.

[0134] For example, in a preferred embodiment of this application, when dealing with small energy storage converters, the inertia coefficient can be set to 1.0 to 2.5s, including 1s and 2.5s; and the damping coefficient can be set to 50kW / Hz. When dealing with medium and large grid-connected devices, the inertia coefficient can be set to 3.0 to 5.0s, including 3s and 5s; and the damping coefficient can be set to 100 to 150kW / Hz.

[0135] Through the above adjustment process, a continuous mapping of "projected power command - power droop target - power angle evolution rate" is achieved, so that the technical solution of this application still has the same dynamic characteristics as the synchronous machine and stable grid support capability under current limiting constraints.

[0136] Furthermore, by synchronously calculating the three-phase voltage and current at the grid connection point and establishing a diagnostic sample sequence, the phase, amplitude, and power information of the current system can be accurately captured, providing real-time support for subsequent control. Based on recursive least squares or extended Kalman filtering to obtain the line equivalent impedance estimate, and combined with the current carrying capacity threshold to construct a current constraint domain, the safe current boundary can be identified in advance before power command allocation, avoiding operation beyond the equipment's capacity.

[0137] Based on this, the current vector is obtained through positive sequence component extraction and coordinate transformation. Then, the original voltage reference is constrained and reshaped using the admittance parameter set to ensure that the generated constrained voltage reference can maintain voltage shaping while limiting the current to within the carrying threshold. Active and reactive power commands are projected based on the current constraint domain and the constrained voltage reference, ensuring that the command values ​​always fall within the feasible range, thereby avoiding the risk of sudden overcurrent. The power droop target is dynamically corrected based on the projected active and reactive power commands, and the power angle evolution rate is adjusted accordingly to achieve continuous evolution of the virtual power angle and consistency with the frequency reference. Through this series of interconnected steps, this invention can maintain grid stability and power distribution coordination under weak grid, fault, or large disturbance conditions, thereby solving the problems of power angle jumps, voltage shaping loss, and multi-machine coordination imbalance under current-limited conditions in existing technologies.

[0138] Example 2

[0139] Please see Figure 2 As shown, this embodiment provides an adaptive virtual synchronous generator control method for an energy storage converter, including:

[0140] S1: Based on the three-phase voltage and three-phase current at the grid connection point, a comprehensive electrical state set is obtained; the comprehensive electrical state set includes voltage phase, current phase, current amplitude, instantaneous active power, and instantaneous reactive power;

[0141] S2: Construct a load-bearing trigger function based on the electrical integrated state set and calculate the load-bearing trigger function value; when the load-bearing trigger function value is greater than the preset load-bearing trigger threshold, generate a load-bearing trigger identifier;

[0142] S3: When a load triggering flag is detected, the electrical comprehensive status set within a preset time window is constructed into a diagnostic sample sequence and input into the load fault diagnosis model, and the load fault diagnosis result is output; the load fault diagnosis result includes voltage shaping loss flag and virtual power angle jump flag;

[0143] S4: When only voltage shaping loss flag or only virtual power angle jump flag exists, the load fault diagnosis result value obtained by quantifying the load fault diagnosis result and the diagnosis sample sequence are input into the parameter intelligent setting model to obtain the set of adjustment parameters; the set of adjustment parameters is converted into adjustment commands to adjust the virtual synchronous generator of the energy storage converter;

[0144] S5: Update the virtual power angle and frequency reference based on the diagnostic sample sequence; perform synchronous calculation and line equivalent impedance estimation on the three-phase voltage and three-phase current at the grid connection point to construct the current constraint domain; extract the current vector and reshape the original voltage reference according to the admittance parameter group to obtain the constrained voltage reference; adjust the power droop target and power angle evolution rate of the virtual synchronous generator of the energy storage converter according to the projected active power command and projected reactive power command obtained by the projection calculation.

[0145] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the claimed scope.

[0146] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive virtual synchronous generator control method for an energy storage converter, characterized in that, include: S1: Based on the three-phase voltage and three-phase current at the grid connection point, a comprehensive electrical state set is obtained; the comprehensive electrical state set includes voltage phase, current phase, current amplitude, instantaneous active power, and instantaneous reactive power; S2: Construct a load-bearing trigger function based on the electrical integrated state set and calculate the load-bearing trigger function value; when the load-bearing trigger function value is greater than the preset load-bearing trigger threshold, generate a load-bearing trigger identifier; S3: When a load triggering flag is detected, the electrical comprehensive status set within the preset time window is constructed into a diagnostic sample sequence and input into the load fault diagnosis model, and the load fault diagnosis result is output. The load fault diagnosis results include voltage shaping loss indicator and virtual power angle jump indicator; S4: When only voltage shaping loss flag or only virtual power angle jump flag exists, the load fault diagnosis result value obtained by quantifying the load fault diagnosis result and the diagnosis sample sequence are input into the parameter intelligent setting model to obtain a set of adjustment parameters including static admittance share, dynamic admittance share, time constant, command projection intensity, power angle change rate and phase angle mitigation term; the set of adjustment parameters is converted into adjustment commands to adjust the virtual synchronous generator of the energy storage converter; S5: When both voltage shaping loss flag and virtual power angle jump flag exist, update the virtual power angle and frequency reference based on the diagnostic sample sequence; perform synchronous calculation of three-phase voltage and three-phase current at the grid connection point and estimate the line equivalent impedance to construct a current constraint domain. Extract the current vector and reshape the original voltage reference according to the admittance parameter group to obtain the constrained voltage reference; adjust the power droop target and power angle evolution rate of the virtual synchronous generator of the energy storage converter according to the projected active power command and projected reactive power command obtained by the projection calculation.

2. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 1, characterized in that, The implementation methods of S5 include: The virtual power angle and frequency reference are updated based on the diagnostic sample sequence; The three-phase voltage and three-phase current at the grid connection point are obtained by using recursive least squares or extended Kalman filtering algorithm to obtain the estimated value of the line equivalent impedance. A current constraint domain is constructed based on the estimated values ​​of the three-phase voltage at the grid connection point, the equivalent impedance of the line, and the preset current carrying capacity threshold. The positive sequence components of the three-phase current at the grid connection point are extracted and the coordinates are transformed to obtain the current vector; Based on the current vector and the admittance parameter set, the original voltage reference is constrained and reshaped to obtain a constrained voltage reference; the admittance parameter set includes the static admittance fraction, the dynamic admittance fraction, and the time constant. Projection calculations are performed on the active and reactive power commands based on the current constraint domain and the constrained voltage reference to obtain the projected active and reactive power commands. The power droop target is adjusted based on the projected active power command and the projected reactive power command, and the power angle evolution rate is adjusted according to the power droop target.

3. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 2, characterized in that, Methods for updating virtual power angle and frequency references based on diagnostic sample sequences include: A small window smoothing process is applied to the instantaneous active and reactive power in the diagnostic sample sequence to obtain the active power measurement value and the reactive power measurement value; the instantaneous active power deviation and the instantaneous reactive power deviation are calculated based on the preset instantaneous active power target value and instantaneous reactive power target value. Zero-sensitivity zone, amplitude limiting, and rate of change constraints are applied to the instantaneous active power deviation and instantaneous reactive power deviation to obtain the effective amount of deviation; the effective amount of deviation is proportionally converted based on the frequency correction caused by the unit active power deviation to obtain the frequency increment; the frequency increment is subjected to bandwidth limiting and noise suppression processing to obtain the target frequency reference. Discrete integration is performed on the target frequency reference over the sampling period to obtain the target virtual power angle; Update the current frequency reference to the target frequency reference; update the current virtual power angle to the target virtual power angle.

4. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 2, characterized in that, The method for constructing the current constraint domain is as follows: Based on the estimated values ​​of the three-phase voltage at the grid connection point, the equivalent impedance of the line, and the preset current carrying capacity threshold, a mapping relationship from instantaneous active and reactive power to current amplitude is established. For any set of instantaneous active and reactive power, the corresponding current amplitude is calculated, and the solution set whose current amplitude does not exceed the preset current carrying capacity threshold is defined as the current constraint domain.

5. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 2, characterized in that, Methods for constraining and reshaping the original voltage reference to obtain a constrained voltage reference include: Multiplying the current vector by the static admittance component yields the steady-state compensation voltage component. A current proximity factor is constructed based on the current amplitude and a preset current carrying threshold; a piecewise linear barrier function is constructed based on the current proximity factor and a preset current proximity threshold. The piecewise linear barrier function output by the piecewise linear barrier function is multiplied by the dynamic component of the admittance to obtain the dynamic compensation voltage component, and then filtered by the time constant to smooth the response. The constrained voltage reference is obtained by superimposing the steady-state compensation voltage component and the dynamic compensation voltage component and then subtracting them from the original voltage reference.

6. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 2, characterized in that, The method for adjusting the power droop target based on projected active power command and projected reactive power command, and adjusting the power angle evolution rate according to the power droop target, includes: Based on the projected active power command and the projected reactive power command, combined with the frequency-active power droop characteristic and the voltage-reactive power droop characteristic, the corresponding active power target point and reactive power target point are calculated respectively, and the active power target point and reactive power target point are merged to form the first power droop target. The current actual active power is compared with the active power in the power droop target to obtain the projected active power deviation; the projected active power deviation is converted into a frequency correction amount according to the preset active power-frequency droop coefficient, and the first power angle evolution rate is formed under the action of the preset inertia coefficient and damping coefficient. Adjust the current power droop target to the first power droop target, and adjust the current power angle evolution rate to the first power angle evolution rate.

7. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 1, characterized in that, The method for obtaining the electrical integrated state set includes: The three-phase voltage and three-phase current at the grid connection point are sampled with a unified clock and timestamp aligned. Sensor drift and noise are eliminated based on anti-aliasing filtering and DC bias removal. The measurement delay of the voltage and current channels is compensated to obtain synchronized voltage and current data sequences. The three-phase voltage and three-phase current at the grid connection point are transformed by two-phase orthogonal transformation to construct voltage orthogonal components and current orthogonal components; the voltage orthogonal components enter the phase synchronization unit to obtain the voltage phase; the current phase is synchronously obtained according to the angle relationship between the current orthogonal components and the voltage phase. The current amplitude is estimated based on the voltage orthogonal component and the current orthogonal component, and instantaneous fluctuations are suppressed by using a sliding window root mean square or small window smoothing to obtain the current amplitude. Instantaneous active power and instantaneous reactive power are determined according to instantaneous power theory; The voltage phase, current phase, current amplitude, instantaneous active power, and instantaneous reactive power are constructed into a comprehensive electrical state set.

8. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 1, characterized in that, The training method for the load-bearing fault diagnosis model includes: A pre-constructed load cell fault diagnosis dataset is constructed, which includes ZD group load cell fault diagnosis data and corresponding load cell fault diagnosis results, where ZD is a positive integer; the load cell fault diagnosis data includes a sequence of diagnosis samples; the load cell fault diagnosis dataset is divided into a training set and a validation set, the training set is used for learning the parameters of the load cell fault diagnosis model, and the validation set is used for real-time monitoring of the generalization performance and overfitting of the load cell fault diagnosis model; A deep neural network with a multilayer perceptron as its core is used as the load cell fault diagnosis model. The load cell fault diagnosis data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer uses a nonlinear activation function to extract features, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each load cell fault diagnosis result. Finally, the load cell fault diagnosis result corresponding to the highest probability is taken as the prediction result of the load cell fault diagnosis model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is set: when the prediction accuracy on the validation set reaches or exceeds the preset prediction accuracy threshold, the load cell fault diagnosis model is determined to have converged and training is terminated.

9. The adaptive virtual synchronous generator control method for an energy storage converter according to claim 1, characterized in that, The training method for the intelligent parameter setting model includes: A pre-constructed intelligent parameter setting dataset is constructed, which includes CS group intelligent parameter setting data and the set of adjustment parameters corresponding to the CS group intelligent parameter setting data, where CS is a positive integer; the intelligent parameter setting data includes numerical values ​​carrying fault diagnosis results and diagnostic sample sequences; the intelligent parameter setting dataset is divided into a training set and a validation set, the training set is used for parameter learning of the intelligent parameter setting model, and the validation set is used for real-time monitoring of the generalization performance and overfitting degree of the intelligent parameter setting model; A deep neural network with a multilayer perceptron as its core is used as the intelligent parameter setting model. The intelligent parameter setting data is standardized and vectorized before being input into the deep neural network, which consists of an input layer, hidden layers, and an output layer. Each hidden layer uses a nonlinear activation function to extract features, and the output layer uses a softmax activation function to obtain the probability distribution corresponding to each set of adjustment parameters. Finally, the set of adjustment parameters corresponding to the highest probability is taken as the prediction result of the intelligent parameter setting model. During training, the cross-entropy loss function is used as the optimization objective, and a gradient descent-type optimization algorithm is used to update the network weights. An early stopping strategy is set: when the prediction accuracy on the validation set reaches or exceeds the preset prediction accuracy threshold, the intelligent parameter setting model is determined to have converged and training is terminated.

Images