Energy storage converter grid voltage adaptive control method based on data acquisition

Abstract

This invention relates to the field of energy storage converter control technology, specifically a data acquisition-based adaptive control method for grid voltage of energy storage converters. The method includes: acquiring the original multi-phase voltage and current sequences at the grid's point of common coupling (PCC) using a multi-channel high-speed synchronous acquisition device, and preprocessing them to obtain real-time electrical quantity data. An improved Hilbert-Huang transform algorithm, dynamically adjusting the intrinsic mode function (EMF) selection criterion based on the grid background harmonic content, is used to separate the dynamic characteristic components of voltage disturbances from the steady-state fundamental component and construct a disturbance feature vector. An attention-based classifier is used to identify the disturbance category and severity level, and the corresponding control objective and dynamic control parameters are obtained through a preset control strategy mapping table. This method can adapt to complex grid harmonic environments, accurately identify voltage disturbance states, and achieve adaptive matching of energy storage converter voltage control parameters.

Description

Technical Field

[0001] This invention relates to the field of energy storage converter control technology, and in particular to a method for adaptive grid voltage control of energy storage converters based on data acquisition. Background Technology

[0002] Existing energy storage converters primarily employ fixed-parameter closed-loop regulation for grid voltage control. They extract grid electrical characteristics through conventional filtering and Fourier transform, relying on threshold values ​​to identify voltage disturbances. Control parameters are pre-set based on typical operating conditions, and control strategies are switched manually or through simple table lookups. Traditional Hilbert-Huang transforms use fixed criteria for mode selection, failing to adapt to real-time changes in grid background harmonics. The accuracy of separating the fundamental and disturbance components is limited, and harmonic interference and noise components are easily introduced. Voltage disturbance identification often uses conventional classification networks, with fixed distributions of attention to different amplitudes and types of disturbances. This leads to deviations in the determination of disturbance categories and severity levels, and the control strategy cannot accurately match the real-time disturbance status.

[0003] Voltage fluctuations in the power grid are accompanied by complex harmonics and transient disturbances. Conventional component extraction methods struggle to maintain feature purity in a variable harmonic environment, and disturbance features are easily masked by background components. Traditional classification methods fail to focus on key disturbance features, resulting in discrepancies between classification results and actual operating conditions. Control objectives and parameters cannot adapt to disturbance states in real time, leading to deviations between converter response characteristics and grid demands. Therefore, it is necessary to adaptively adjust mode selection rules based on the background harmonic content of the power grid to accurately separate voltage disturbance-related features. Simultaneously, a feature focusing mechanism should be introduced to optimize disturbance classification performance. Based on the classification results, corresponding control parameters should be matched to achieve adaptive regulation of the grid voltage by the energy storage converter. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a data acquisition-based adaptive control method for grid voltage of energy storage converters.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a data acquisition-based adaptive control method for grid voltage of an energy storage converter, comprising: By deploying a multi-channel high-speed synchronous acquisition device on the AC side of the energy storage converter, the raw sampling data sequence of multiphase voltage and current at the grid common connection point is acquired in real time. The original sampled data sequence is preprocessed to obtain the real-time electrical quantity data set; An improved Hilbert-Huang transform algorithm is applied to the real-time electrical quantity data set. The improved Hilbert-Huang transform algorithm dynamically adjusts the intrinsic mode function screening criteria based on the background harmonic content of the power grid to separate the dynamic characteristic components and steady-state fundamental components that characterize voltage disturbances from the data. Based on the dynamic feature components, a disturbance feature vector describing the transient process of grid voltage is constructed; The disturbance feature vector is input into a voltage disturbance pattern classifier based on an attention mechanism, and the voltage disturbance pattern classifier outputs the specific disturbance category and severity level of the current grid voltage. Based on the disturbance category and severity level, a preset control strategy mapping table is queried to obtain the corresponding voltage control target set and dynamic control parameter set.

[0006] As a further aspect of the present invention, the improved Hilbert-Huang transform algorithm dynamically adjusts the intrinsic mode function screening criteria based on the background harmonic content of the power grid, including: Empirical mode decomposition is performed on the standardized real-time electrical quantity data set to generate a series of candidate intrinsic mode function components; Perform spectrum analysis on the real-time electrical quantity data set to calculate the background harmonic content index of a specific frequency band; The background harmonic content index is compared with a preset threshold to determine whether the current power grid is in a high harmonic background or a low harmonic background state. When the background state is determined to be high harmonic, the screening criteria for intrinsic mode functions are tightened. That is, the number of extreme points and zero crossings of each candidate component must be strictly equal, and the local mean curve must be closer to the zero line, so as to suppress the mixing of high-frequency harmonic interference into the effective dynamic characteristic components. When the background is determined to be low harmonic, the intrinsic mode function screening criteria are relaxed, that is, the local mean values ​​of the upper and lower envelopes of the candidate components are allowed to have slight shifts, so as to retain more dynamic characteristic components that reflect subtle voltage fluctuations. Based on the adjusted screening criteria, valid intrinsic mode functions that meet the conditions are selected from the candidate components of the intrinsic mode functions and reorganized into the dynamic characteristic components. The remaining parts are regarded as the steady-state fundamental wave components.

[0007] As a further aspect of the present invention, based on the dynamic feature components, a disturbance feature vector describing the transient process of the power grid voltage is constructed, including: Time-frequency analysis is performed on the dynamic feature components to extract their energy distribution characteristics, instantaneous frequency change rate characteristics, and envelope shape characteristics at different time scales; Calculate the start and end times, duration, peak amplitude, and cumulative energy value of the dynamic characteristic components; The energy distribution characteristics, instantaneous frequency change rate characteristics, envelope shape characteristics, start and end times, duration, peak amplitude, and cumulative energy value are arranged and normalized according to a predetermined dimensional order, and combined to form a multidimensional numerical vector, which is the disturbance feature vector.

[0008] As a further aspect of the present invention, time-frequency analysis is performed on the dynamic feature components to extract their energy distribution characteristics, instantaneous frequency change rate characteristics, and envelope shape characteristics at different time scales, including: The dynamic feature components are analyzed using continuous wavelet transform to generate their time-frequency distribution spectrum; In the time-frequency distribution map, several representative time scale windows are selected; For each selected time scale window, the signal energy of the dynamic feature component within the time scale window is calculated. The signal energy is obtained by summing the squares of the magnitudes of all time-frequency coefficients within the time scale window. The signal energy distribution under all time scale windows constitutes the energy distribution feature. The instantaneous frequency is calculated based on the analytical signal of the dynamic feature components. The first derivative of the curve of instantaneous frequency changing with time is obtained to obtain the rate of change curve of instantaneous frequency changing with time. The maximum value, mean value and trend characteristics are extracted from the rate of change curve to form the instantaneous frequency rate of change characteristics. The envelope shape is obtained by calculating the upper and lower envelopes of the dynamic characteristic component waveform; The geometric features of the envelope shape are extracted, including the average width of the envelope line, the variance of the width over time, and the slope features of the envelope line at the start and end of the disturbance, which together constitute the envelope shape features.

[0009] As a further aspect of the present invention, the construction and operation process of the voltage disturbance mode classifier includes: A training database containing various typical voltage disturbance samples is pre-built, and each disturbance sample is labeled with a disturbance category label and a severity level label; The voltage perturbation pattern classifier based on the attention mechanism is trained using labeled perturbation samples, so that it learns the mapping relationship from the perturbation feature vector to the perturbation category and severity level. In real-time operation, the voltage perturbation pattern classifier receives the currently constructed perturbation feature vector; The multi-head attention layer inside the classifier calculates the correlation weights between features of each dimension within the perturbation feature vector, highlighting the key features most relevant to the current perturbation. The classifier's output layer simultaneously generates a joint probability distribution of the perturbation category and severity level, and the combination with the highest probability is taken as the classification output.

[0010] As a further aspect of the present invention, the process of constructing and querying the control strategy mapping table includes: The control strategy mapping table is indexed by a combination of all disturbance categories and severity levels identified by the voltage disturbance pattern classifier. Under each index entry, a corresponding set of voltage control targets is stored. The set of voltage control targets includes the voltage amplitude target, phase target, unbalance limit target, and harmonic content limit target that need to be maintained. Under each index entry, a corresponding set of dynamic control parameters is also stored. The set of dynamic control parameters includes the proportional-integral parameters of the inner loop current controller, the control bandwidth of the outer loop voltage, and the adjustment rate limits of active power and reactive power. Based on the disturbance category and severity level output by the voltage disturbance mode classifier, a matching query is performed in the control strategy mapping table to obtain the voltage control target set and the dynamic control parameter set.

[0011] As a further aspect of the present invention, the method further includes: combining the steady-state fundamental component, the voltage control target set, and the dynamic control parameter set to calculate in real time the compensation current reference command that the energy storage converter should output; The compensation current reference command is input to the pulse width modulation controller of the energy storage converter to generate a corresponding switching drive signal. The switching drive signal is applied to the power switching device of the energy storage converter, causing it to output the compensation current corresponding to the compensation current reference command, so as to dynamically support and correct the grid voltage. The real-time calculation of the compensation current reference command that the energy storage converter should output includes: The amplitude and phase of the positive sequence component of the current grid voltage are calculated using the steady-state fundamental component. Calculate the desired grid voltage vector based on the voltage amplitude target and phase target in the voltage control target set; Compare the current grid voltage vector with the expected grid voltage vector, and calculate the voltage deviation. Based on the voltage deviation, and combined with the control bandwidth and adjustment rate limits in the dynamic control parameter set, the d-axis and q-axis reference values ​​of the compensation current are generated using the voltage control outer loop algorithm, which constitutes the compensation current reference command. During the calculation process, the limitations on unbalance and harmonics in the voltage control target set need to be considered simultaneously, and the corresponding negative sequence current compensation command and harmonic current compensation command components need to be added to the compensation current reference command.

[0012] As a further aspect of the present invention, the specific implementation process of the voltage control outer loop algorithm includes: The voltage deviation is decomposed into a d-axis voltage deviation component and a q-axis voltage deviation component in a rotating coordinate system; The d-axis voltage deviation component and the q-axis voltage deviation component are respectively subjected to proportional-integral regulation, and the parameters of the proportional-integral regulator are determined by the proportional-integral parameters in the dynamic control parameter set. The output of the proportional-integral controller is the initial d-axis current reference value and q-axis current reference value; The initial d-axis current reference value and q-axis current reference value are input into the limiting circuit constructed by the adjustment rate limit value in the set of dynamic control parameters to obtain the final d-axis and q-axis reference values.

[0013] As a further aspect of the present invention, the generation of the corresponding switch drive signal includes: The compensation current reference command is transformed from the rotating coordinate system back to the stationary three-phase coordinate system to obtain the three-phase instantaneous compensation current reference value; The instantaneous value of the three-phase current actually output by the energy storage converter is obtained through a current sensor; The three-phase instantaneous compensation current reference value is compared with the actual output instantaneous three-phase current value to obtain the three-phase current error value; The three-phase current error value is input into the current inner loop controller, and the parameters of the current inner loop controller are also set by the dynamic control parameter set. The current inner loop controller outputs a three-phase modulated wave signal. The three-phase modulated wave signal is compared with a triangular carrier wave to generate the switching drive signal that controls the power switching device to turn on and off.

[0014] As a further aspect of the present invention, it also includes an online self-updating process for the mapping table between the voltage disturbance mode classifier and the control strategy: The system continuously records the disturbance feature vectors, disturbance categories and severity levels output by the classifier, and the actual recovery of the grid voltage after implementing the corresponding control strategies during historical operation. When the actual recovery deviates continuously from the control expectation, or when disturbance characteristics not defined in the mapping table frequently occur, the self-update process is triggered. The self-updating process uses newly accumulated data to incrementally train the voltage perturbation pattern classifier, optimize its classification boundary, or add new perturbation categories. At the same time, the voltage control target set and dynamic control parameter set under the corresponding item in the control strategy mapping table are optimized or supplemented based on the newly accumulated data.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Based on the dynamic adjustment of intrinsic mode function screening criteria according to the background harmonic content of the power grid, the improved Hilbert-Huang transform can adaptively adjust the mode screening conditions according to the real-time harmonic level of the power grid when processing electrical quantity data. This suppresses the influence of background harmonics on the signal decomposition process, clearly distinguishes the steady-state fundamental component and dynamic characteristic component in the voltage signal, reduces the mixing of harmonic interference and random fluctuations into the disturbance characteristics, and enables the extracted dynamic characteristic components to fully reflect the transient change information of the power grid voltage. The purity and recognizability of the disturbance characteristics are improved, and the constructed disturbance feature vector can truly correspond to the actual disturbance pattern of the power grid voltage, avoiding interference from irrelevant components in the subsequent disturbance identification process.

[0016] By inputting the disturbance feature vector into an attention-based voltage disturbance pattern classifier, differentiated weights can be assigned to different dimensions of the disturbance features. This highlights the features that play a dominant role in disturbance determination and weakens the recognition interference caused by non-critical features, making the disturbance category and severity level output by the classifier more closely match the real-time operating status of the power grid. Based on the obtained disturbance category and severity level, a preset control strategy mapping table can be queried to directly obtain the voltage control target set and dynamic control parameter set adapted to the current operating conditions. This establishes a corresponding matching relationship between the control parameters and the power grid disturbance state, allowing the control output to adjust synchronously with changes in power grid voltage disturbances. This ensures that the control behavior of the energy storage converter remains coordinated with the power grid voltage state, enhancing the targeting and adaptability of control commands. Attached Figure Description

[0017] Figure 1 This is a state diagram of the data acquisition-based adaptive control method for grid voltage of energy storage converters according to the present invention. Figure 2 A flowchart for extracting three features from dynamic feature time-frequency analysis; Figure 3 A flowchart illustrating the construction and operation process of a voltage disturbance mode classifier. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 The overall implementation scheme of the adaptive control method for grid voltage of energy storage converter based on data acquisition is as follows: A multi-channel high-speed synchronous acquisition device deployed on the AC side of the energy storage converter acquires the original sampled data sequence of multiphase voltage and current at the grid's point of common coupling in real time. The original sampled data sequence is preprocessed, including filtering, calibration, and timing alignment, to obtain a real-time electrical quantity data set. An improved Hilbert-Huang transform algorithm is applied to the real-time electrical quantity data set. This algorithm dynamically adjusts the intrinsic mode function screening criteria based on the grid background harmonic content to separate the dynamic characteristic components and steady-state fundamental components characterizing voltage disturbances from the data. Based on the dynamic characteristic components, a disturbance feature vector describing the transient process of grid voltage is constructed. The disturbance feature vector is input into a voltage disturbance pattern classifier based on an attention mechanism. This voltage disturbance pattern classifier outputs the specific disturbance category and severity level of the current grid voltage. Based on the disturbance category and severity level, a preset control strategy mapping table is queried to obtain the corresponding voltage control target set and dynamic control parameter set.

[0021] In one embodiment of the present invention, empirical mode decomposition is performed on the standardized real-time electrical quantity data set to generate a series of candidate intrinsic mode functions (IMFs). Spectral analysis is performed on the real-time electrical quantity data set to calculate the background harmonic content index for a specific frequency band. The background harmonic content index is compared with a preset threshold to determine whether the current power grid is in a high-harmonic background or low-harmonic background state. When a high-harmonic background state is determined, the IMF screening criteria are tightened, requiring that the number of extreme points and zero-crossing points of each candidate component must be strictly equal, and the local mean curve must be closer to the zero line, thereby suppressing high-frequency harmonic interference from mixing into effective dynamic characteristic components. When a low-harmonic background state is determined, the IMF screening criteria are relaxed, allowing slight shifts in the local mean values ​​of the upper and lower envelopes of candidate components to retain more dynamic characteristic components reflecting subtle voltage fluctuations. Based on the adjusted screening criteria, effective IMFs that meet the conditions are selected from the candidate IMFs and recombined into dynamic characteristic components; the remaining components are considered as steady-state fundamental components.

[0022] In practical implementation, the improved Hilbert-Huang transform algorithm dynamically adjusts the intrinsic mode function (IMF) selection criteria based on the background harmonic content of the power grid. The implementation process is illustrated through an example involving a power grid background harmonic scenario. Empirical mode decomposition (EMD) is performed on the standardized real-time electrical quantity data set to generate a series of IMF candidate components arranged from high to low frequency, denoted as IMF1, IMF2, ..., IMFn. In practical implementation, spectral analysis is performed on the real-time electrical quantity data set to calculate the background harmonic content index for a specific frequency band. The formula for calculating the background harmonic content index H can be expressed as:

[0023] in: Representing the The amplitude of the subharmonic voltage. Represents the fundamental voltage amplitude, set This includes all considered harmonic orders, such as the 3rd, 5th, 7th, 9th, and 11th. In an example scenario, the threshold for distinguishing between high-harmonic and low-harmonic backgrounds is set as follows: When the calculated background harmonic content index Greater than the preset threshold At this time, it is determined that the current power grid is in a high harmonic background state. In some embodiments, the high harmonic background state may correspond to the background harmonic content index. For scenarios with values ​​greater than 0.05, in practical implementation, when the background is determined to be high harmonic, the intrinsic mode function (EMF) screening criteria are tightened. Specifically, this requires that the number of extreme points and the number of zero-crossing points for each candidate component must be strictly equal, and the local mean curve calculated from the upper and lower envelopes must be closer to the zero line. This "closer" can be achieved by setting a smaller upper limit for the absolute value of the local mean. This is achieved by suppressing the intrusion of high-frequency harmonic interference into the effective dynamic characteristic components.

[0024] In another example scenario, when the calculated background harmonic content index Less than or equal to a preset threshold At this time, it is determined that the current power grid is in a low harmonic background state. In some embodiments, the low harmonic background state may correspond to the background harmonic content index. Scenarios with values ​​less than or equal to 0.02. In practical implementation, when the background is determined to be low harmonic, the intrinsic mode function (EMF) screening criteria are relaxed. Specifically, this means that slight shifts in the local means of the upper and lower envelopes of candidate components are allowed. These "slight shifts" can be achieved by setting a relatively lenient upper limit on the absolute value of the local means. To achieve this, in which Greater than This is to retain more dynamic characteristic components reflecting subtle voltage fluctuations. Optionally, the calculation frequency range of the background harmonic content index can be configured according to the harmonic spectrum characteristics of the actual power grid. It can be understood that the threshold... The specific values ​​can be set according to the typical harmonic levels of the target power grid. Based on the adjusted screening criteria, all valid intrinsic mode functions (IMFs) that meet the conditions are selected from the candidate IMF components and reorganized into dynamic characteristic components. The remaining components that fail the screening, along with low-frequency residual components, are collectively considered as steady-state fundamental components. Optionally, the screening process is iterative, applying the dynamically adjusted criteria to each candidate IMF component. In practical implementation, the comparison between the tightening criteria under high harmonic background conditions and the relaxation criteria under low harmonic background conditions is reflected in the number and characteristics of the selected valid IMFs. Under high harmonic background conditions, the number of selected components is usually smaller and the frequency components are relatively concentrated, while under low harmonic background conditions, the number of selected components may be larger, containing richer transient details.

[0025] In one embodiment of the present invention, the process of constructing a perturbation feature vector based on dynamic feature components includes feature extraction and vector combination. Time-frequency analysis is performed on the dynamic feature components to extract their energy distribution characteristics, instantaneous frequency change rate characteristics, and envelope shape characteristics at different time scales. The start and end times, duration, peak amplitude, and cumulative energy value of the dynamic feature components are calculated. The energy distribution characteristics, instantaneous frequency change rate characteristics, envelope shape characteristics, start and end times, duration, peak amplitude, and cumulative energy value are arranged and normalized according to a predetermined dimensional order, and combined to form a multidimensional numerical vector, which is the perturbation feature vector. Specifically, the time-frequency analysis steps are: continuous wavelet transform is used to analyze the dynamic feature components to generate their time-frequency distribution spectrum. Several representative time scale windows are selected in the time-frequency distribution spectrum.

[0026] For each selected timescale window, the signal energy of the dynamic characteristic component within that window is calculated. This signal energy is obtained by summing the squares of the magnitudes of all time-frequency coefficients within the window. The signal energy distribution across all timescale windows constitutes the energy distribution characteristic. Based on the analytic signal of the dynamic characteristic component, its instantaneous frequency is calculated. The first derivative of the instantaneous frequency versus time curve is obtained, yielding the rate of change curve of the instantaneous frequency over time. The maximum value, mean, and trend characteristics are extracted from this rate of change curve, collectively forming the instantaneous frequency rate of change characteristic. The envelope shape is obtained by calculating the upper and lower envelopes of the dynamic characteristic component waveform. The geometric features of the envelope shape are extracted, including the average width of the envelope, the variance of the width over time, and the slope of the envelope at the start and end of the disturbance, collectively forming the envelope shape characteristic.

[0027] In practical implementation, the process of constructing perturbation feature vectors based on dynamic feature components includes time-frequency analysis of the dynamic feature components, calculation of feature parameters, and vector combination. Specifically, time-frequency analysis is performed on the dynamic feature components to extract their energy distribution characteristics, instantaneous frequency change rate characteristics, and envelope shape characteristics at different time scales. For further details, please refer to [link / reference]. Figure 2 The dynamic characteristic components are analyzed using continuous wavelet transform to generate a time-frequency distribution map of the dynamic characteristic components. Several representative time-scale windows are selected in the time-frequency distribution map. For each selected time-scale window, the signal energy of the dynamic characteristic component within that window is calculated. The signal energy is obtained by summing the squares of the magnitudes of all time-frequency coefficients within the time-scale window. The signal energy distribution under all time-scale windows constitutes the energy distribution feature. Based on the analytic signal of the dynamic characteristic component, its instantaneous frequency is calculated. The first derivative of the instantaneous frequency versus time curve is obtained to obtain the rate of change curve of the instantaneous frequency over time. The maximum value, mean, and trend characteristics are extracted from the rate of change curve to constitute the instantaneous frequency rate of change feature. The envelope shape is obtained by calculating the upper and lower envelopes of the dynamic characteristic component waveform. The geometric features of the envelope shape are extracted, including the average width of the envelope, the variance of the width over time, and the slope characteristics of the envelope at the start and end of the disturbance, which together constitute the envelope shape feature. In some embodiments, the envelope width variance... The calculation formula is expressed as:

[0028] in: Represents at discrete time points The envelope width at a given point is the difference between the upper and lower envelope values. This represents the average envelope width across all time points. This represents the total number of time points. In practice, the start and end times, duration, peak amplitude, and cumulative energy value of the dynamic feature components are calculated. The start and end times of the dynamic feature components are determined by detecting the time point when the amplitude of the dynamic feature component first continuously exceeds a preset threshold and the time point when it last falls back below the preset threshold. The duration is the difference between the start and end times. The peak amplitude is the absolute maximum value of the amplitude of the dynamic feature component within the duration. The cumulative energy value is the sum of the squares of the amplitudes of all sampling points of the dynamic feature component within the duration. In practice, the energy distribution features, instantaneous frequency change rate features, envelope shape features, start and end times, duration, peak amplitude, and cumulative energy value are arranged and normalized according to a predetermined dimensional order, forming a multi-dimensional numerical vector. This multi-dimensional numerical vector is the perturbation feature vector. Optionally, the normalization process can be linearly scaled based on the maximum and minimum values ​​of each feature dimension in the training database. It is understood that the predetermined dimensional order needs to be consistent during model training and online application.

[0029] In some embodiments, the energy distribution characteristics may include energy values ​​at three different time scale windows; the instantaneous frequency change rate characteristics include the maximum value, mean value, and trend coefficient; and the envelope shape characteristics include the average width, width variance, initial slope, and final slope. These, along with the start and end times, duration, peak amplitude, and cumulative energy value, together form a thirteen-dimensional disturbance feature vector. Optionally, the selection of the time scale window can be determined based on the main frequency distribution range of a typical voltage disturbance. It is understood that the dimensions and specific features of the disturbance feature vector can be adjusted according to actual application requirements.

[0030] In one embodiment of the present invention, the construction and operation process of the voltage perturbation pattern classifier includes offline training and online application. See also... Figure 3A training database containing various typical voltage perturbation samples is pre-built, with each perturbation sample labeled with a perturbation category and severity level. The labeled perturbation samples are used to train an attention-based voltage perturbation pattern classifier, enabling it to learn the mapping relationship from perturbation feature vectors to perturbation categories and severity levels. In real-time operation, the voltage perturbation pattern classifier receives the currently constructed perturbation feature vector. The multi-head attention layer within the classifier calculates the correlation weights between features of different dimensions within the perturbation feature vector, highlighting the key features most relevant to the current perturbation. The classifier's output layer simultaneously generates a joint probability distribution of perturbation category and severity level, and the combination with the highest probability is taken as the classification output. The construction and querying of the control policy mapping table are synchronized. The control policy mapping table uses all combinations of perturbation categories and severity levels identified by the voltage perturbation pattern classifier as indexes. Under each index entry, a corresponding set of voltage control targets is stored, including voltage amplitude targets, phase targets, imbalance limit targets, and harmonic content limit targets that need to be maintained. Each index entry also stores a corresponding set of dynamic control parameters. These parameters include the proportional-integral parameters of the inner-loop current controller, the control bandwidth of the outer-loop voltage controller, and the regulation rate limits for active and reactive power. Based on the disturbance category and severity level output by the voltage disturbance mode classifier, a matching query is performed in the control strategy mapping table to obtain the voltage control target set and the dynamic control parameter set.

[0031] In practice, the construction and operation of the voltage disturbance pattern classifier includes two stages: offline training and online application, with the construction and querying of the control policy mapping table corresponding to these stages. A training database containing various typical voltage disturbance samples is pre-built. Each disturbance sample is labeled with a disturbance category and severity level. Disturbance categories include voltage sags, voltage swells, voltage interruptions, and voltage harmonics. Severity levels are divided into three levels based on the amplitude and duration of the voltage deviation: Level 1, Level 2, and Level 3. The labeled disturbance samples are used to train the attention-based voltage disturbance pattern classifier, enabling it to learn the mapping relationship from disturbance feature vectors to disturbance categories and severity levels. The training process optimizes the network parameters by minimizing the classification loss function.

[0032] In real-time operation, the voltage disturbance pattern classifier receives the currently constructed disturbance feature vector. The multi-head attention layer within the classifier calculates the correlation weights between features of each dimension within the disturbance feature vector, highlighting the key features most relevant to the current disturbance. Through the classifier's output layer, a joint probability distribution of the disturbance category and severity level is generated simultaneously. The combination with the highest probability is taken as the classification output, for example, "voltage sag, level two". In specific implementation, the control strategy mapping table uses all combinations of disturbance categories and severity levels identified by the voltage disturbance pattern classifier as indexes. Under each index entry, a corresponding voltage control target set is stored, including the voltage amplitude target, phase target, imbalance limit target, and harmonic content limit target to be maintained. Under each index entry, a corresponding dynamic control parameter set is also stored, including the proportional-integral parameters of the inner loop current controller, the control bandwidth of the outer voltage loop, and the regulation rate limits for active and reactive power. Based on the disturbance category and severity level output by the voltage disturbance pattern classifier, a matching query is performed in the control strategy mapping table to obtain the voltage control target set and the dynamic control parameter set. Optionally, the control strategy mapping table can be stored in the form of a database table or a configuration file. It is understood that the specific values ​​of the voltage control target set and the dynamic control parameter set need to be preset according to the power grid operation specifications and converter performance. In some embodiments, the calculation of the attention weight of the voltage disturbance mode classifier involves an intermediate variable, the calculation formula of which is expressed as:

[0033] in: Indicates the first In the attention head, the first Attention weights for each feature dimension. This represents the raw attention score calculated by the dot product of the query vector and the key vector. As the index variable for the summation operation, it takes integer values ​​of 1, 2, ..., N in each summation operation. Indicates the first In a single attention head, when the index variable When a specific value is taken, the corresponding first... The original attention scores for each feature dimension. This represents the total dimension of the perturbation feature vector. This indicates exponential operations. In some embodiments, a portion of the control policy mapping table can be illustrated using a sample table, see Table 1, which shows some control objectives and parameters under different disturbance categories and severity levels.

[0034] Table 1: Control Strategy Mapping Table

[0035] In one embodiment of the present invention, the reference command for the compensation current to be output by the energy storage converter is calculated in real time by combining the steady-state fundamental component, the voltage control target set, and the dynamic control parameter set. The reference command for the compensation current is input to the pulse width modulation controller of the energy storage converter to generate a corresponding switching drive signal. The switching drive signal is applied to the power switching devices of the energy storage converter, causing them to output the compensation current corresponding to the reference command for compensation current, thereby dynamically supporting and correcting the grid voltage. The process of calculating the reference command for the compensation current to be output by the energy storage converter in real time involves calculating the amplitude and phase of the positive sequence component of the current grid voltage using the steady-state fundamental component.

[0036] Based on the voltage amplitude and phase targets in the voltage control target set, the desired grid voltage vector is calculated. The current grid voltage vector is compared with the desired grid voltage vector to calculate the voltage deviation. Based on the voltage deviation, and combined with the control bandwidth and regulation rate limitations in the dynamic control parameter set, the voltage control outer loop algorithm generates d-axis and q-axis reference values ​​for the compensation current, thus forming the compensation current reference command. During the calculation, the limitations regarding unbalance and harmonics in the voltage control target set must be considered simultaneously, and corresponding negative sequence current compensation and harmonic current compensation command components are added to the compensation current reference command. The specific implementation process of the voltage control outer loop algorithm includes: decomposing the voltage deviation into d-axis and q-axis voltage deviation components in a rotating coordinate system; performing proportional-integral (PI) regulation on the d-axis and q-axis voltage deviation components respectively, with the parameters of the PI regulator determined by the PI parameters in the dynamic control parameter set; and the output of the PI regulator being the preliminary d-axis and q-axis current reference values. The initial d-axis current reference values ​​and q-axis current reference values ​​are input into a limiting circuit constructed from the adjustment rate limit values ​​in the dynamic control parameter set to obtain the final d-axis and q-axis reference values.

[0037] In practical implementation, the reference command for the compensation current to be output by the energy storage converter is calculated in real time by combining the steady-state fundamental component, the voltage control target set, and the dynamic control parameter set. In practical implementation, the positive sequence component amplitude of the current grid voltage is calculated using the steady-state fundamental component. With phase Based on the voltage amplitude target in the voltage control target set. With phase target Calculate the desired grid voltage vector. Compare the current grid voltage vector with the desired grid voltage vector to calculate the voltage deviation. Based on voltage deviation By combining the control bandwidth and regulation rate limitations in the dynamic control parameter set, the d-axis and q-axis reference values ​​of the compensation current are generated using the voltage control outer loop algorithm, thus forming the compensation current reference command. During the calculation process, the limitations on unbalance and harmonics in the voltage control target set must be considered simultaneously, and corresponding negative sequence current compensation command and harmonic current compensation command components are added to the compensation current reference command.

[0038] In practical implementation, the specific implementation process of the voltage control outer loop algorithm includes: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] In a rotating coordinate system, it is decomposed into d-axis voltage deviation components. q-axis voltage deviation component For the d-axis voltage deviation component q-axis voltage deviation component Proportional-integral (PI) regulation is performed separately, and the parameters of the PI controller are determined by the PI parameters in the dynamic control parameter set. The output of the PI controller is the initial d-axis current reference value. q-axis current reference value The initial d-axis current reference value. q-axis current reference value The input is a limiting element constructed from the adjustment rate limit values ​​in the dynamic control parameter set, which yields the final d-axis and q-axis reference values. and In some embodiments, the proportional-integral adjustment process involves the calculation of an integral term, which... The update formula in discrete time is expressed as:

[0039] in: This represents the integral term output value of the current control cycle. This represents the integral term output value of the previous control cycle. This represents the integral coefficient determined by the set of dynamic control parameters. Indicates the sampling period of the control system. The voltage deviation input for the current control cycle (can be...) or In some embodiments, the calculation of the compensation current reference command and the parameter settings of the limiting stage can be referenced to a specific example, see Table 2, which shows two sets of parameters for different disturbance levels.

[0040] Table 2: Voltage Outer Loop Control Parameters

[0041] It is understood that the parameter values ​​in the table are for illustrative purposes; actual applications require adjustments based on the specific converter model and grid requirements. Optionally, for control targets related to voltage imbalance limitations, the calculated values ​​will be adjusted accordingly. and A negative-sequence current command is superimposed on top of this. It can be understood that the generation of the harmonic current compensation command component requires real-time detection of the remaining harmonic components after separation of the steady-state fundamental component. Optionally, different preset values ​​are provided in the control strategy mapping table for different disturbance categories and severity levels. , and Parameter combinations.

[0042] In one embodiment of the present invention, the process of generating the corresponding switch drive signal involves coordinate transformation and modulation. The compensation current reference command is inversely transformed from the rotating coordinate system back to the stationary three-phase coordinate system to obtain the three-phase instantaneous compensation current reference value. The actual output instantaneous value of the three-phase current of the energy storage converter is obtained through a current sensor. The three-phase instantaneous compensation current reference value is compared with the actual output instantaneous value of the three-phase current to obtain the three-phase current error value. The three-phase current error value is input to the current inner loop controller, whose parameters are also set by the dynamic control parameter set. The current inner loop controller outputs a three-phase modulated wave signal. The three-phase modulated wave signal is compared with a triangular carrier wave to generate switch drive signals for controlling the power switching devices to turn on and off. The method also includes an online self-updating process for the voltage disturbance mode classifier and control strategy mapping table. The disturbance feature vector, disturbance category and severity level output by the classifier, and the actual recovery status of the grid voltage after executing the corresponding control strategy are continuously recorded during historical operation. When the actual recovery status deviates continuously from the control expectation, or when disturbance features not defined in the mapping table frequently occur, the self-updating process is triggered. The self-updating process uses newly accumulated data to incrementally train the voltage disturbance pattern classifier, optimizing its classification boundary or adding new disturbance categories. Simultaneously, it optimizes or supplements the voltage control target set and dynamic control parameter set under the corresponding entries in the control strategy mapping table based on the newly accumulated data.

[0043] In practice, the process of generating the corresponding switch drive signals is based on the compensation current reference command output by the voltage control outer loop algorithm and the actual output current of the energy storage converter. The compensation current reference command is inversely transformed from the rotating coordinate system back to the stationary three-phase coordinate system to obtain the three-phase instantaneous compensation current reference values. The instantaneous value of the three-phase current actually output by the energy storage converter is obtained through a current sensor. The reference value of the three-phase instantaneous compensation current. Compared with the actual output instantaneous value of three-phase current By comparison, the three-phase current error values ​​are obtained. The three-phase current error value The input current inner loop controller, whose parameters are also set by the dynamic control parameter set, outputs a three-phase modulated wave signal. The three-phase modulated wave signal With triangular carrier The comparison generates switching drive signals to control the power switching devices to turn on and off. In practical implementation, the current inner loop controller typically uses proportional-integral regulation, and its output modulated wave signal... The discrete calculation expression for the k-th control cycle is:

[0044] in: This represents the output value of the modulated wave signal corresponding to phase x (a, b, c) in the current control cycle. This represents the proportional coefficient of the inner current loop, set by the dynamic control parameter set. This represents the x-phase current error value in the current control cycle. This represents the integral coefficient of the inner current loop, set by the dynamic control parameter set. Indicates the sampling period of the control system. This represents the cumulative sum of historical values ​​of the x-phase current error from the initial moment to the current moment. In specific implementations, the online self-update process of the voltage disturbance mode classifier and control strategy mapping table runs continuously, continuously recording the disturbance feature vectors, disturbance categories and severity levels output by the classifier, and the actual recovery status of the grid voltage after executing the corresponding control strategy. When there is a persistent deviation between the actual recovery status and the control expectation, or when disturbance features not defined in the mapping table frequently occur, the self-update process is triggered. The self-update process uses the newly accumulated data to incrementally train the voltage disturbance mode classifier, optimize its classification boundary, or add new disturbance categories. At the same time, it optimizes or supplements the voltage control target set and dynamic control parameter set under the corresponding item in the control strategy mapping table based on the newly accumulated data. In some embodiments, the actual recovery status is evaluated by comparing the key indicators of the grid voltage after control with the target values ​​in the voltage control target set. The key indicators include steady-state voltage deviation, recovery time, and overshoot. It can be understood that the judgment of persistent deviation can be based on statistical thresholds, such as the average voltage deviation after multiple consecutive controls exceeding a preset tolerance. Optionally, incremental training can be performed in the background during periods of low system load to avoid impacting real-time control performance. In some embodiments, a new disturbance category label and initial level are temporarily assigned to disturbance features not defined in the mapping table, and an initial set of voltage control targets and dynamic control parameters are generated based on the default control strategy. The self-update process then iteratively optimizes the corresponding control strategy mapping table entries based on subsequent control effect data of the new disturbance category. It is understood that newly accumulated data needs to be validated and screened before being used for self-update. Optionally, the self-update process can be designed to be manually triggered or run automatically periodically.

[0045] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A data acquisition-based adaptive control method for grid voltage of an energy storage converter, characterized in that, include: By deploying a multi-channel high-speed synchronous acquisition device on the AC side of the energy storage converter, the raw sampling data sequence of multiphase voltage and current at the power grid common connection point is acquired in real time. The original sampled data sequence is preprocessed to obtain the real-time electrical quantity data set; An improved Hilbert-Huang transform algorithm is applied to the real-time electrical quantity data set. The improved Hilbert-Huang transform algorithm dynamically adjusts the intrinsic mode function screening criteria based on the background harmonic content of the power grid to separate the dynamic characteristic components and steady-state fundamental components that characterize voltage disturbances from the data. Based on the dynamic feature components, a disturbance feature vector describing the transient process of grid voltage is constructed; The disturbance feature vector is input into a voltage disturbance pattern classifier based on an attention mechanism, and the voltage disturbance pattern classifier outputs the specific disturbance category and severity level of the current grid voltage. Based on the disturbance category and severity level, a preset control strategy mapping table is queried to obtain the corresponding voltage control target set and dynamic control parameter set.

2. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 1, characterized in that, The improved Hilbert-Huang transform algorithm dynamically adjusts the intrinsic mode function screening criteria based on the background harmonic content of the power grid, including: Empirical mode decomposition is performed on the standardized real-time electrical quantity data set to generate a series of candidate components of intrinsic mode functions; Perform spectrum analysis on the real-time electrical quantity data set to calculate the background harmonic content index of a specific frequency band; The background harmonic content index is compared with a preset threshold to determine whether the current power grid is in a high harmonic background or a low harmonic background state. When the background state is determined to be high harmonic, the screening criteria for intrinsic mode functions are tightened. That is, the number of extreme points and zero crossings of each candidate component must be strictly equal, and the local mean curve must be closer to the zero line, so as to suppress the mixing of high-frequency harmonic interference into the effective dynamic characteristic components. When the background is determined to be low harmonic, the intrinsic mode function screening criteria are relaxed, that is, the local mean values ​​of the upper and lower envelopes of the candidate components are allowed to have slight shifts, so as to retain more dynamic characteristic components that reflect subtle voltage fluctuations. Based on the adjusted screening criteria, valid intrinsic mode functions that meet the conditions are selected from the candidate components of the intrinsic mode functions and reorganized into the dynamic characteristic components. The remaining parts are regarded as the steady-state fundamental wave components.

3. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 2, characterized in that, Based on the dynamic feature components, a disturbance feature vector describing the transient process of the power grid voltage is constructed, including: Time-frequency analysis is performed on the dynamic feature components to extract their energy distribution characteristics, instantaneous frequency change rate characteristics, and envelope shape characteristics at different time scales; Calculate the start and end times, duration, peak amplitude, and cumulative energy value of the dynamic characteristic components; The energy distribution characteristics, instantaneous frequency change rate characteristics, envelope shape characteristics, start and end times, duration, peak amplitude, and cumulative energy value are arranged and normalized according to a predetermined dimensional order, and combined to form a multidimensional numerical vector, which is the disturbance feature vector.

4. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 3, characterized in that, Time-frequency analysis is performed on the dynamic feature components to extract their energy distribution characteristics, instantaneous frequency change rate characteristics, and envelope shape characteristics at different time scales, including: The dynamic feature components are analyzed using continuous wavelet transform to generate their time-frequency distribution spectrum; In the time-frequency distribution map, several representative time scale windows are selected; For each selected time scale window, the signal energy of the dynamic feature component within the time scale window is calculated. The signal energy is obtained by summing the squares of the magnitudes of all time-frequency coefficients within the time scale window. The signal energy distribution under all time scale windows constitutes the energy distribution feature. The instantaneous frequency is calculated based on the analytical signal of the dynamic feature components. The first derivative of the curve of instantaneous frequency changing with time is obtained to obtain the rate of change curve of instantaneous frequency changing with time. The maximum value, mean value and trend characteristics are extracted from the rate of change curve, which together constitute the instantaneous frequency rate of change characteristics. The envelope shape is obtained by calculating the upper and lower envelopes of the dynamic characteristic component waveform; The geometric features of the envelope shape are extracted, including the average width of the envelope, the variance of the width over time, and the slope features of the envelope at the start and end of the disturbance, which together constitute the envelope shape features.

5. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 4, characterized in that, The construction and operation process of the voltage perturbation pattern classifier includes: A training database containing various typical voltage disturbance samples is pre-built, and each disturbance sample is labeled with a disturbance category label and a severity level label; The voltage perturbation pattern classifier based on the attention mechanism is trained using labeled perturbation samples, so that it learns the mapping relationship from the perturbation feature vector to the perturbation category and severity level. In real-time operation, the voltage perturbation pattern classifier receives the currently constructed perturbation feature vector; The multi-head attention layer inside the classifier calculates the correlation weights between features of each dimension within the perturbation feature vector, highlighting the key features most relevant to the current perturbation. The classifier's output layer simultaneously generates a joint probability distribution of the perturbation category and severity level, and the combination with the highest probability is taken as the classification output.

6. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 5, characterized in that, The process of constructing and querying the control strategy mapping table includes: The control strategy mapping table is indexed by a combination of all disturbance categories and severity levels identified by the voltage disturbance pattern classifier. Under each index entry, a corresponding set of voltage control targets is stored. The set of voltage control targets includes the voltage amplitude target, phase target, unbalance limit target, and harmonic content limit target that need to be maintained. Under each index entry, a corresponding set of dynamic control parameters is also stored. The set of dynamic control parameters includes the proportional-integral parameters of the inner loop current controller, the control bandwidth of the outer loop voltage, and the adjustment rate limits of active power and reactive power. Based on the disturbance category and severity level output by the voltage disturbance mode classifier, a matching query is performed in the control strategy mapping table to obtain the voltage control target set and the dynamic control parameter set.

7. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 6, characterized in that, The method further includes: combining the steady-state fundamental component, the voltage control target set, and the dynamic control parameter set to calculate in real time the compensation current reference command that the energy storage converter should output; The compensation current reference command is input to the pulse width modulation controller of the energy storage converter to generate a corresponding switching drive signal. The switching drive signal is applied to the power switching device of the energy storage converter, causing it to output the compensation current corresponding to the compensation current reference command, so as to dynamically support and correct the grid voltage. The real-time calculation of the compensation current reference command that the energy storage converter should output includes: The amplitude and phase of the positive sequence component of the current grid voltage are calculated using the steady-state fundamental component. Calculate the desired grid voltage vector based on the voltage amplitude target and phase target in the voltage control target set; Compare the current grid voltage vector with the expected grid voltage vector, and calculate the voltage deviation. Based on the voltage deviation, and combined with the control bandwidth and adjustment rate limits in the dynamic control parameter set, the d-axis and q-axis reference values ​​of the compensation current are generated using the voltage control outer loop algorithm, which constitutes the compensation current reference command. During the calculation process, the limitations on unbalance and harmonics in the voltage control target set need to be considered simultaneously, and the corresponding negative sequence current compensation command and harmonic current compensation command components need to be added to the compensation current reference command.

8. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 7, characterized in that, The specific implementation process of the voltage control outer loop algorithm includes: The voltage deviation is decomposed into a d-axis voltage deviation component and a q-axis voltage deviation component in a rotating coordinate system; The d-axis voltage deviation component and the q-axis voltage deviation component are respectively subjected to proportional-integral regulation, and the parameters of the proportional-integral regulator are determined by the proportional-integral parameters in the dynamic control parameter set. The output of the proportional-integral controller is the initial d-axis current reference value and q-axis current reference value; The initial d-axis current reference value and q-axis current reference value are input into the limiting circuit constructed by the adjustment rate limit value in the set of dynamic control parameters to obtain the final d-axis and q-axis reference values.

9. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 8, characterized in that, The generation of the corresponding switch drive signal includes: The compensation current reference command is transformed from the rotating coordinate system back to the stationary three-phase coordinate system to obtain the three-phase instantaneous compensation current reference value; The instantaneous value of the three-phase current actually output by the energy storage converter is obtained through a current sensor; The three-phase instantaneous compensation current reference value is compared with the actual output instantaneous three-phase current value to obtain the three-phase current error value; The three-phase current error value is input into the current inner loop controller, and the parameters of the current inner loop controller are also set by the dynamic control parameter set. The current inner loop controller outputs a three-phase modulated wave signal. The three-phase modulated wave signal is compared with a triangular carrier wave to generate the switching drive signal that controls the power switching device to turn on and off.

10. The adaptive grid voltage control method for energy storage converters based on data acquisition according to claim 9, characterized in that, It also includes an online self-updating process for the mapping table between the voltage disturbance pattern classifier and the control strategy: The system continuously records the disturbance feature vectors, disturbance categories and severity levels output by the classifier, and the actual recovery of the grid voltage after implementing the corresponding control strategies during historical operation. When the actual recovery deviates continuously from the control expectation, or when disturbance characteristics not defined in the mapping table frequently occur, the self-update process is triggered. The self-updating process uses newly accumulated data to incrementally train the voltage perturbation pattern classifier, optimize its classification boundary, or add new perturbation categories. At the same time, the voltage control target set and dynamic control parameter set under the corresponding item in the control strategy mapping table are optimized or supplemented based on the newly accumulated data.

Images