Energy storage system operation management method and system for power quality fusion evaluation

Abstract

The application provides a power quality fusion evaluation energy storage system operation management method and system, time series signals of voltage and current are decoupled in frequency domain to obtain multiple frequency sub-band components; features of all frequency sub-band components are extracted to obtain a deep feature map, a type label of a composite power quality disturbance and an attention weight vector of all frequency sub-bands are determined based on the deep feature map; a dominant constraint is determined according to the type label, a frequency band weighting coefficient of each constraint term is determined according to the attention weight vector, each frequency band weighting coefficient is prioritized by the dominant constraint, and a set of modulated weighting coefficients is obtained; a virtual inertia and a power factor angle are determined based on the set of weighting coefficients, and power quality compensation is performed on the power grid. Based on the above scheme, the frequency domain decoupling features of the composite power quality disturbance can be introduced into the dynamic weighting fusion process of the energy storage control parameters to realize constraint priority modulation and adaptive compensation under disturbance scene perception.

Description

Technical Field

[0001] This application relates to the field of power quality monitoring data processing technology, and more specifically, to a method and system for the operation and management of energy storage systems for power quality fusion assessment. Background Technology

[0002] With the large-scale grid connection of distributed renewable energy sources such as wind power and photovoltaics, power quality disturbances in the power grid are becoming increasingly prominent. In particular, composite power quality disturbances formed by the superposition of multiple individual disturbances have severe aliasing and masking of different disturbance components in the time-domain waveform, making accurate disturbance identification extremely difficult. At the same time, as a flexible regulation resource for the power grid, energy storage systems, when participating in frequency regulation and voltage support, typically rely on fixed multi-constraint weighted fusion rules for their virtual synchronous generator control parameters. This makes it difficult to dynamically adjust according to the disturbance type and spectral distribution, resulting in insufficient adaptive regulation capabilities of energy storage systems under complex disturbance scenarios.

[0003] Existing methods for identifying complex power quality disturbances often employ time-frequency analysis combined with manual feature extraction or a single deep learning classification network. Their output only provides disturbance category labels and fails to integrate the identification results with energy storage operation control strategies in a closed-loop manner. Furthermore, multi-constraint control methods for energy storage virtual synchronous generators typically use preset fixed weighting coefficients, which cannot adaptively adjust the contribution ratio of each constraint term based on the energy distribution differences of the disturbance signal across different frequency bands. Therefore, how to incorporate the frequency domain decoupling characteristics of complex power quality disturbances into the dynamic weighted fusion process of energy storage control parameters to achieve constraint priority modulation and adaptive compensation under disturbance scenario perception has become a challenge for the industry. Summary of the Invention

[0004] This application provides a method and system for the operation and management of energy storage systems based on power quality fusion assessment. It can introduce the frequency domain decoupling characteristics of composite power quality disturbances into the dynamic weighted fusion process of energy storage control parameters to achieve constraint priority modulation and adaptive compensation under disturbance scenario perception.

[0005] In a first aspect, this application provides a method for the operation and management of an energy storage system based on integrated power quality assessment, comprising the following steps: The voltage and current timing signals at the grid connection point are acquired, and the voltage and current timing signals are decoupled in the frequency domain to obtain multiple frequency sub-band components. Hierarchical feature extraction is performed on all frequency sub-band components to obtain deep feature maps. Based on the deep feature maps, channel attention self-calibration and weighted classification mapping are performed to obtain the type label of composite power quality disturbance and the attention weight vector of all frequency sub-bands. The dominant constraints for virtual synchronous generator control are determined based on the type label, and the frequency band weighting coefficients for each constraint item are determined based on the attention weight vector. Then, the frequency band weighting coefficients are preferentially modulated through the dominant constraints to obtain a set of modulated weighting coefficients. The constraint terms are weighted and fused based on the weighted coefficient set to obtain the virtual inertia and power factor angle, and then the power grid is compensated for power quality based on the virtual inertia and the power factor angle.

[0006] In some embodiments, frequency domain decoupling of the voltage timing signal and the current timing signal to obtain multiple frequency sub-band components specifically includes: The voltage timing signal and the current timing signal are converted into voltage frequency domain coefficient sequences and current frequency domain coefficient sequences, respectively. Synchronous sliding sampling is performed on the voltage frequency domain coefficient sequence and the current frequency domain coefficient sequence to obtain multiple voltage frequency band spectrum sequences and current frequency band spectrum sequences covering different frequency ranges respectively; All voltage and current frequency spectrum sequences are converted back to the time domain to obtain multiple frequency sub-band components.

[0007] In some embodiments, hierarchical feature extraction is performed on all frequency sub-band components to obtain a deep feature map, specifically including: All frequency subband components are input into a hierarchical feature extraction network, and the first energy spectrum is obtained by local feature mapping through the first-level residual block; The receptive field of the first energy spectrum is gradually expanded by the second-level residual block to obtain the second energy spectrum; The deep feature map is obtained by capturing the long-range dependencies between the frequency band components in the second energy spectrum through the third-level residual block.

[0008] In some embodiments, channel attention self-calibration and weighted classification mapping are performed based on the deep feature map to obtain the type label of the composite power quality disturbance and the attention weight vector of all frequency sub-bands, specifically including: Multi-level channel attention self-calibration is performed on the deep feature map to obtain a channel importance-enhanced feature map; Residual feature extraction is performed on the channel importance enhancement feature map to obtain a residual refined feature map; Channel-level feature aggregation is performed on the residual refined feature map to generate multi-class original features; The original feature vectors of the multi-classification are input into the perturbation classifier for dual-path parallel mapping to obtain the type label of the composite power quality perturbation and the attention weight vectors of all frequency sub-bands.

[0009] In some embodiments, determining the dominant constraints for virtual synchronous generator control based on the type label specifically includes: Determine the set of constraints for virtual synchronous generator control; Based on the type label, select a dominant constraint for virtual synchronous generator control from the set of constraints.

[0010] In some embodiments, the virtual inertia and power factor angle are obtained by weighting and fusing the constraint terms based on the weighted coefficient set, specifically including: Based on the weighted coefficient set, the frequency change rate constraint and the energy storage state of charge constraint are weighted and fused to obtain the virtual inertia. The power factor angle is obtained by weighting and fusing the node voltage constraints based on the weighted coefficient set.

[0011] In some embodiments, power quality compensation of the power grid based on the virtual inertia and the power factor angle specifically includes: The virtual inertia and the power factor angle are input to the virtual synchronous generator control module of the energy storage converter to calculate the active power reference value and reactive power reference value of the energy storage converter. A modulation wave signal is generated based on the active power reference value and the reactive power reference value; The modulated wave signal drives the power semiconductor devices of the energy storage converter to perform charging and discharging operations.

[0012] Secondly, this application provides an energy storage system operation and management system for power quality fusion assessment, and a method for performing power quality fusion assessment of an energy storage system operation and management system, including: The frequency domain decoupling module is used to acquire the voltage timing signal and the current timing signal at the grid connection point, and to perform frequency domain decoupling on the voltage timing signal and the current timing signal to obtain multiple frequency sub-band components. The constrained modulation module is used to extract hierarchical features from all frequency sub-band components to obtain a deep feature map. Based on the deep feature map, channel attention self-calibration and weighted classification mapping are performed to obtain the type label of composite power quality disturbance and the attention weight vector of all frequency sub-bands. The constraint modulation module is further configured to determine the dominant constraint of virtual synchronous generator control based on the type label, determine the frequency band weighting coefficient of each constraint item based on the attention weight vector, and then perform priority modulation on each frequency band weighting coefficient through the dominant constraint to obtain a set of modulated weighting coefficients. The compensation module is used to perform weighted fusion of constraint terms based on the weighted coefficient set to obtain virtual inertia and power factor angle, and then perform power quality compensation on the power grid based on the virtual inertia and power factor angle.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described energy storage system operation management method for power quality fusion assessment.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described energy storage system operation management method for power quality fusion assessment.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application provides a method and system for the operation and management of an energy storage system for power quality fusion assessment, in which voltage and current time-series signals at the grid connection point are acquired, and frequency domain decoupling is performed on the voltage and current time-series signals to obtain multiple frequency sub-band components; Hierarchical feature extraction is performed on all frequency sub-band components to obtain deep feature maps. Based on the deep feature maps, channel attention self-calibration and weighted classification mapping are performed to obtain the type label of composite power quality disturbance and the attention weight vector of all frequency sub-bands. The dominant constraints for virtual synchronous generator control are determined based on the type label, and the frequency band weighting coefficients for each constraint item are determined based on the attention weight vector. Then, the frequency band weighting coefficients are preferentially modulated through the dominant constraints to obtain a set of modulated weighting coefficients. The constraint terms are weighted and fused based on the weighted coefficient set to obtain the virtual inertia and power factor angle, and then the power grid is compensated for power quality based on the virtual inertia and the power factor angle.

[0016] Therefore, in this application, power quality compensation for the power grid is performed based on the virtual inertia and the power factor angle. First, determining the frequency sub-band components yields a vector composed of a pair of sub-band voltage time-series signals and a pair of sub-band current time-series signals. By decoupling the voltage and current time-series signals in the frequency domain, the voltage and current time-series signals are decomposed into multiple frequency sub-band components. This effectively solves the problem of mutual masking and feature aliasing of various disturbance components in the time domain waveform during composite power quality disturbances. It separates disturbances with different frequency characteristics, such as transient rises, transient falls, harmonics, and oscillating transients, which were originally entangled, in the frequency domain, thereby facilitating subsequent hierarchical feature extraction and channel attention. Self-calibration provides clear and independent frequency band inputs; based on these frequency sub-band components, the attention mechanism can quantify the contribution of each frequency band to the current composite disturbance, thereby enabling each constraint term in the energy storage virtual synchronous generator control to obtain dynamic weighting coefficients that match the disturbance spectrum distribution. This ultimately achieves constraint priority modulation and adaptive control parameter calculation under disturbance scenario perception, significantly improving the accuracy of composite power quality disturbance identification and the adaptive operation and management capabilities of the energy storage system in complex disturbance environments. Then, determining the set of modulated weighting coefficients yields a set containing the modulated weighting coefficients of all constraint terms. This scheme achieves this through... The method dynamically selects the dominant constraint in the current scenario based on the type label of the composite power quality disturbance, and assigns frequency band-dependent weighting coefficients to the energy storage state of charge constraint, frequency change rate constraint, and node voltage constraint according to the energy distribution of each frequency sub-band reflected by the attention weight vector. Then, the dominant constraint is used to prioritize and modulate each weighting coefficient, enabling the modulated weighting coefficient set to adaptively adjust the contribution ratio of each constraint term according to the disturbance type and spectral characteristics. Compared to existing multi-constraint fusion methods that use fixed weighting coefficients and cannot perceive the disturbance type, this method effectively solves the problem of the inability of energy storage virtual synchronous generator control to adapt to multi-constraint scenarios. The problem of dynamically adjusting control weights based on the frequency domain characteristics of composite disturbances allows the calculation of virtual inertia and power factor angles to prioritize the control objectives of the dominant constraints while also taking into account the safety boundaries of other constraints. This improves the response speed and control accuracy of the energy storage system under complex disturbances such as frequency fluctuations and voltage over-limits, enhances the adaptive capability and robustness of power quality compensation, and ensures the safe and stable operation of the energy storage grid-connected system. In summary, based on the above scheme, the frequency domain decoupling characteristics of composite power quality disturbances can be introduced into the dynamic weighted fusion process of energy storage control parameters to achieve constraint priority modulation and adaptive compensation under disturbance scenario perception. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of an energy storage system operation management method for power quality fusion assessment according to some embodiments of this application; Figure 2 This is a flowchart illustrating the operation of determining a deep feature map according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the determination of dominant constraints according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of an energy storage system operation and management system for power quality fusion assessment according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a power quality fusion assessment method for the operation and management of an energy storage system, according to some embodiments of this application. Detailed Implementation

[0018] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figure 1 This figure is an exemplary flowchart of an energy storage system operation management method for power quality fusion assessment according to some embodiments of this application. The figure mainly includes the following steps: In step 101, the voltage timing signal and current timing signal at the grid connection point are obtained, and the voltage timing signal and the current timing signal are decoupled in the frequency domain to obtain multiple frequency sub-band components.

[0020] It should be noted that in this application, the voltage time-series signal is a digital waveform data characterizing the dynamic change sequence of voltage amplitude over time, which can be used to reflect the fundamental frequency and low-frequency disturbance characteristics such as voltage swell, voltage drop, and interruption; the current time-series signal is a digital waveform data characterizing the dynamic change sequence of current amplitude over time, which can be used to reflect the mid-to-high frequency disturbance characteristics such as harmonics, flicker, and oscillation; by using the voltage time-series signal and the current time-series signal together as the input of the subsequent frequency domain decoupling module, it is possible to simultaneously capture complete information of composite power quality disturbances in different frequency bands, thereby improving the accuracy of disturbance type identification and the precision of energy storage control parameter adaptation, and realizing an integrated closed loop of power quality fusion assessment and energy storage operation management.

[0021] In some embodiments, the voltage and current timing signals at the grid connection point can be obtained in the following manner: First, the rated voltage and rated current signals on the primary side of the grid connection point can be converted into weak voltage and weak current analog signals with good linear conversion characteristics using voltage and current transformers, respectively. Simultaneously, a low-pass filter circuit is used to perform anti-aliasing filtering on the weak voltage and weak current analog signals to remove high-frequency noise components and avoid spectral aliasing during subsequent analog-to-digital conversion due to non-compliance with the sampling theorem. Second, phase-locked loop (PLL) technology is used to synchronously track the grid fundamental frequency in real time, dynamically generating synchronous sampling control pulses based on grid frequency fluctuations. For example, when the grid fundamental frequency... When the frequency fluctuates within the range of 45 Hz to 55 Hz, the phase-locked loop automatically adjusts the frequency multiplier output to ensure that the synchronous sampling pulse always maintains a definite phase-locked relationship with the fundamental frequency of the power grid, thereby ensuring that the sampling points are strictly distributed at integer multiples of the power grid signal cycle to eliminate spectral leakage errors. Then, triggered by the synchronous sampling control pulse generated by the phase-locked loop, an analog-to-digital converter is used to synchronously sample and convert the weak voltage analog signal and weak current analog signal after anti-aliasing filtering, converting the continuous-time analog signal into a discrete-time digital signal to obtain the voltage timing signal and current timing signal at the grid connection point. The sampling frequency of the analog-to-digital converter is set to 3200 Hz, and a total of 640 sampling points are collected every ten fundamental frequency cycles.

[0022] In some embodiments, frequency domain decoupling of the voltage timing signal and the current timing signal to obtain multiple frequency sub-band components can be achieved by the following steps: The voltage timing signal and the current timing signal are converted into voltage frequency domain coefficient sequences and current frequency domain coefficient sequences, respectively. Synchronous sliding sampling is performed on the voltage frequency domain coefficient sequence and the current frequency domain coefficient sequence to obtain multiple voltage frequency band spectrum sequences and current frequency band spectrum sequences covering different frequency ranges respectively; All voltage and current frequency spectrum sequences are converted back to the time domain to obtain multiple frequency sub-band components.

[0023] It should be noted that in this application, the voltage frequency domain coefficient sequence characterizes the energy distribution characteristics of the voltage signal at different frequency positions; the current frequency domain coefficient sequence characterizes the energy distribution characteristics of the current signal at different frequency positions, and is analyzed in conjunction with the voltage frequency domain coefficient sequence to fully describe the dual-channel information of the composite power quality disturbance in the frequency domain; the voltage frequency band spectrum sequence is a local frequency domain coefficient subset that only covers a certain continuous frequency range, used to extract the independent spectral components of the voltage signal in the specified frequency band, avoiding feature aliasing between disturbances in different frequency bands; the current frequency band spectrum sequence is a local frequency domain coefficient subset that strictly corresponds to the voltage frequency band spectrum sequence in terms of frequency range and sliding position, used to synchronously extract the spectral components of the current signal in the same specified frequency band, maintaining the phase and energy correspondence between voltage and current in the frequency band; the frequency sub-band component is a vector composed of a pair of sub-band voltage time-series signals and sub-band current time-series signals, with one frequency sub-band component corresponding to an independent frequency range, used to decompose the composite power quality disturbance into multiple independent time-domain waveform components according to the frequency band, thereby providing independent feature inputs for each frequency band for subsequent adaptive weighted control based on the attention mechanism.

[0024] In practical implementation, firstly, a weighted sum-based discrete cosine transform (DCT) can be performed on both the voltage and current time-series signals. This involves multiplying the instantaneous value of the voltage time-series signal at each sampling point with the corresponding element in a pre-calculated and stored orthogonalized cosine weighting coefficient matrix. Then, all multiplications are summed to obtain a DCT coefficient at the corresponding frequency domain index position. After traversing all frequency domain index positions, all DCT coefficients are arranged in frequency domain index order to form a numerical sequence of the same length as the voltage time-series signal, serving as the voltage frequency domain coefficient sequence. Similarly, the same process is performed on the current time-series signal. The same weighted sum operation yields the current frequency domain coefficient sequence. Then, a pre-defined multi-band sampling window is used to synchronously slide the voltage and current frequency domain coefficient sequences. This involves setting a rectangular sliding window function with a width of 150 Hz. Starting from the beginning of the voltage frequency domain coefficient sequence, the window slides sequentially with a step size of 120 Hz. At each sliding window position, the rectangular sliding window function is multiplied element-wise with the voltage frequency domain coefficient sequence to extract a subset of coefficients within the frequency range covered by the window. This yields the voltage frequency band spectrum sequence corresponding to the first frequency range, from 0 to 150 Hz. Then, the sliding window... The sliding window is incremented by 120 Hz to extract the voltage frequency band spectrum sequence covering the second window frequency range, from 20 Hz to 270 Hz. This process is repeated until the sliding window covers the preset highest analysis frequency range. Simultaneously, the above window-by-window sampling operation is performed on the current frequency domain coefficient sequence to obtain the current frequency band spectrum sequence that corresponds one-to-one with each voltage frequency band spectrum sequence in time and frequency range. Finally, a discrete cosine inverse transform is performed on each voltage frequency band spectrum sequence, that is, the frequency domain coefficient sequence is multiplied and accumulated with the pre-calculated and stored orthogonalized cosine weight coefficient matrix according to the weighted sum operation rules. The coefficient values ​​at the frequency domain index positions are mapped back to the time domain sampling points point by point to obtain the sub-band voltage timing signal of the corresponding frequency range. Similarly, the same discrete cosine inverse transform is performed on the spectrum sequence of each current frequency band to obtain the sub-band current timing signal of the corresponding frequency range. Then, each pair of sub-band voltage timing signals and sub-band current timing signals covering the same frequency range are arranged in channels in order of increasing frequency range to form a frequency sub-band component. The bandwidth of each channel in the frequency sub-band component is 150 Hz, and there is a 30 Hz frequency overlap between adjacent channels, thus obtaining multiple frequency sub-band components.

[0025] In step 102, hierarchical feature extraction is performed on all frequency sub-band components to obtain a deep feature map. Based on the deep feature map, channel attention self-calibration and weighted classification mapping are performed to obtain the type label of composite power quality disturbance and the attention weight vector of all frequency sub-bands.

[0026] In some embodiments, reference Figure 2The figure is a flowchart illustrating the operation of determining a deep feature map according to some embodiments of this application. In this application, the deep feature map is obtained by performing hierarchical feature extraction on all frequency sub-band components using the following steps: All frequency subband components are input into a hierarchical feature extraction network, and the first energy spectrum is obtained by local feature mapping through the first-level residual block; The receptive field of the first energy spectrum is gradually expanded by the second-level residual block to obtain the second energy spectrum; The deep feature map is obtained by capturing the long-range dependencies between the frequency band components in the second energy spectrum through the third-level residual block.

[0027] It should be noted that in this application, the hierarchical feature extraction network is a stacked structure with three layers of one-dimensional convolutional residual blocks as the backbone, where each residual block consists of a main branch and a shortcut connection branch. This hierarchical feature extraction network is trained through supervised learning, based on a pre-constructed simulation dataset containing forty-two categories of power quality disturbances. After the deep feature map is input into the fully connected classifier, the difference between the predicted category and the true label is calculated using the multi-class cross-entropy loss function. The adaptive moment estimation gradient descent algorithm is then used to iteratively update the network weights under the conditions of a batch size of 256, 500 iterations, and adaptive adjustment of the learning rate, until the loss function converges.

[0028] In practical implementation, firstly, all frequency sub-band components can be stacked into a multi-channel input matrix according to the channel dimension. Each channel of this multi-channel input matrix corresponds to a sub-band component within a specified frequency range. For example, the first channel corresponds to the low-frequency fundamental component from 0 to 150 Hz, and the second channel corresponds to the mid-frequency harmonic component from 150 to 270 Hz. Then, this multi-channel input matrix is ​​input into a hierarchical feature extraction network. In the first-level residual block of the hierarchical feature extraction network, the main branch uses two stacked one-dimensional convolutional kernels to perform local feature mapping on the multi-channel input matrix. The one-dimensional convolutional kernel slides point-by-point along the signal sequence length dimension to extract local temporal patterns. The kernel size is set to 3 x 1, and the number of output channels of the first convolutional layer is set... The number of output channels in the second convolutional layer is also set to 32. This allows the original multi-channel frequency sub-band information to be encoded into a high-dimensional feature map through two nonlinear transformations. Simultaneously, the shortcut connection branch of the first-level residual block uses a one-dimensional unit convolution kernel to adjust the channel dimension of the multi-channel input matrix, ensuring that the number of channels output by this branch is consistent with the number of channels output by the main branch. Then, the feature map output by the main branch and the feature map output by the shortcut connection branch are added element-wise. After superposition, a nonlinear transformation is performed through the activation function of the linear correction unit to suppress the propagation of negative features and improve the forward propagation efficiency of the gradient. The feature map output after activation by the linear correction unit is used as the first energy map. Then, the first energy map is input... The second-level residual block of the hierarchical feature extraction network is input, and the number of output channels of the one-dimensional convolutional kernel of this residual block is set to sixty-four. When performing local feature mapping in the length dimension of the signal sequence, due to the increase in the number of convolutional layers, the effective time range covered by each convolutional operation gradually expands, enabling the network to simultaneously perceive the mutual influence between multiple frequency subbands in the first energy spectrum, such as whether the harmonic content in the mid-to-high frequency band changes synchronously when a sag event occurs in the low-frequency band at the same time. After the second-level residual block completes element-wise addition and activates the linear correction unit, it outputs the second-level feature spectrum as the second energy spectrum. Finally, the second energy spectrum is input to the third-level residual block, and the number of output channels of the one-dimensional convolutional kernel of the third-level residual block is further increased. The step size is set to 128. Through deeper nonlinear mapping, the local time-frequency pattern is gradually abstracted into perturbation pattern features with semantic meaning. Max pooling layers are used to downsample the feature map between the second-level and third-level residual blocks and within each residual block. The pooling window size is 2x1 and the step size is 2x1. After each max pooling operation, the size of the feature map in the length dimension of the signal sequence is halved, while the coverage of the receptive field in the sequence dimension is doubled. This step-by-step compression design enables the network to capture long-range temporal dependencies across the entire sampling period while maintaining computational efficiency. The feature map output after max pooling of the third-level residual block is then used as the deep feature map.

[0029] It should be noted that in this application, the first energy map is an intermediate feature representation of the local time-frequency energy of each frequency band encoded by the channel dimension. It is used to retain the fine-grained distribution information of the original disturbance signal in different frequency sub-bands, providing a high-resolution local feature basis for subsequent hierarchical abstraction. The second energy map is an intermediate feature representation with the number of channels extended to sixty-four dimensions. It is used to bridge the representation gap between local time-frequency details and global semantic abstraction. The deep feature map is a high-dimensional feature representation of the essential attributes of composite power quality disturbances. It enables the classifier to accurately distinguish multiple composite disturbance types based on complete coupled features, thereby improving the disturbance identification accuracy and the reliability of energy storage control parameter adaptation.

[0030] In some embodiments, the channel attention self-calibration and weighted classification mapping based on the deep feature map to obtain the type label of the composite power quality perturbation and the attention weight vector of all frequency sub-bands can be achieved by the following steps: Multi-level channel attention self-calibration is performed on the deep feature map to obtain a channel importance-enhanced feature map; Residual feature extraction is performed on the channel importance enhancement feature map to obtain a residual refined feature map; Channel-level feature aggregation is performed on the residual refined feature map to generate multi-class original features; The original feature vectors of the multi-classification are input into the perturbation classifier for dual-path parallel mapping to obtain the type label of the composite power quality perturbation and the attention weight vectors of all frequency sub-bands.

[0031] It should be noted that in this application, the channel importance enhancement feature map is a weighted feature map representing the key frequency band features related to the current composite power quality disturbance, used to suppress redundant channel information, thereby improving the accuracy of disturbance classification and the fineness of attention weight allocation; the residual refined feature map is a high-dimensional abstract feature map that retains the deep time-frequency domain structure information in the disturbance signal and avoids gradient vanishing and overfitting, used to enhance the model's ability to represent complex composite disturbances; the multi-classification original feature is to compress the most significant feature response of each channel into a fixed-length one-dimensional feature vector, used to simplify the feature dimension and retain key discrimination information, providing a stable and highly discriminative input for the classifier; the type label is the identifier of the category to which the composite power quality disturbance belongs, used to inform the energy storage controller of the specific disturbance type present in the current power grid, thereby guiding the selection of dominant constraints and the adaptive switching of control strategies; the attention weight vector is a set of normalized weights that quantify the contribution of different frequency bands to the current composite disturbance, with each weight corresponding to the importance of a frequency sub-band.

[0032] In practical implementation, firstly, multi-level channel attention self-calibration can be performed on the deep feature map. This involves using global mean pooling to compress the values ​​of all sampling points within each channel of the deep feature map into a single average value, obtaining a channel-level global description vector. Then, a compression-transformation technique is used to compress the dimension of the channel-level global description vector, with a compression ratio set to one-quarter, meaning the number of channels is reduced from the original value to one-quarter of the original value. Next, the compressed feature vector is dimension-restored to ensure the number of output channels matches the number of channels in the input deep feature map. Finally, a logistic activation function is used to normalize the restored feature vector, mapping the output value of each channel to the zero-to-one interval to form a channel attention weight vector. Finally, the channel attention weight vector is... The force weight vector is element-wise multiplied with the corresponding channel of the deep feature map to obtain the channel importance enhancement feature map. Next, the deep features of the channel importance enhancement feature map are extracted level by level by stacking three residual block structures. The main branch of each residual block uses two 3x1 one-dimensional convolutional layers, followed by a modified linear unit activation function to enhance nonlinear fitting capability. The number of output channels in the first residual block is set to 32, the second to 64, and the third to 128. Within each residual block, the shortcut branch uses a 1x1 one-dimensional convolutional layer, ensuring that the number of output channels in the shortcut branch is consistent with the number of output channels in the main branch. The output of the shortcut branch is element-wise added to the output of the main branch and then input into the next residual block. A 2x1 max-pooling layer is inserted between adjacent residual blocks to downsample the sequence dimension of the feature map by a factor of two, forcibly reducing the feature dimension and highlighting salient features. The feature map output from the third residual block is then used as the refined residual feature map. Next, global max-pooling is used to iterate through all sampling points in each channel of the refined residual feature map, comparing the values ​​of all sampling points in each channel and extracting the maximum value as the representative feature value of that channel. For a refined residual feature map with 128 output channels, global max-pooling yields 128 feature values, each corresponding to the maximum response intensity of a channel. The 128 feature values ​​are arranged in channel index order to form a one-dimensional vector as the original feature vector for multi-classification. Finally, the original feature vector for multi-classification is input into a perturbation classifier for dual-path parallel mapping. One path of the perturbation classifier is used for perturbation category identification, and the other path is used for attention weight generation. In the perturbation category identification branch, the original feature vector for multi-classification passes through the first fully connected layer and the dropout layer in sequence. The dropout layer randomly disconnects some neuron connections with a deactivation probability of 0.3 to prevent overfitting. Then, it is mapped to the output dimension equal to the total number of perturbation categories by the second fully connected layer. The output value is normalized to the probability distribution of each perturbation category by the flexible maximum activation function. The category corresponding to the maximum probability is taken as the type label of the composite power quality perturbation.In the attention weight generation branch, the original multi-class feature vectors are sequentially passed through the third fully connected layer and the modified linear unit activation function, then compressed to an output dimension equal to the number of frequency sub-bands by the fourth fully connected layer. The output values ​​are then normalized to the zero-to-one interval by the logistic activation function, yielding the attention weights for each frequency sub-band. Finally, the vector composed of the attention weights from all frequency sub-bands is used as the attention weight vector.

[0033] In step 103, the dominant constraints of virtual synchronous generator control are determined according to the type label, the frequency band weighting coefficients of each constraint item are determined according to the attention weight vector, and then the frequency band weighting coefficients are preferentially modulated by the dominant constraints to obtain the set of modulated weighting coefficients.

[0034] In some embodiments, reference Figure 3 The figure is an exemplary flowchart illustrating the determination of dominant constraints according to some embodiments of this application. The determination of dominant constraints for virtual synchronous generator control based on the type label in this application can be achieved using the following steps: In step 1031, the set of constraints for virtual synchronous generator control is determined; In step 1032, a dominant constraint for virtual synchronous generator control is selected from the set of constraints based on the type label.

[0035] It should be noted that in this application, the constraint set is a boundary condition library of virtual synchronous generator control parameters, composed of energy storage state of charge constraints, frequency change rate constraints, and node voltage constraints. This library provides optional constraint sources for subsequent weighted fusion calculations. By presetting this constraint set, multiple control objectives, including energy storage operation safety and grid power quality improvement, can be covered, laying the foundation for subsequent dynamic selection of control priorities. The dominant constraint refers to the constraint with the highest control priority under the current disturbance scenario, used to prioritize the frequency band weighting coefficients of other constraints. In this scheme, by selecting the dominant constraint, the control strategy of the energy storage converter can adaptively match the actual disturbance type, avoiding control conflicts between multiple constraints and improving the pertinence of power quality governance and the robustness of energy storage operation management.

[0036] In practical implementation, firstly, the preset parameters of the energy storage converter controller can be configured through offline calibration. This involves using the rated capacity, rated voltage, and safe operating range of the energy storage system's state of charge (SOC) as the basic parameters for SOC constraints. For example, the minimum allowable SOC value is set to 20%, and the maximum allowable SOC value to be 80%. Simultaneously, upper and lower boundaries of the frequency change rate constraint are constructed based on the allowable range of grid frequency deviation and frequency change rate, and upper and lower boundaries of the node voltage constraint are constructed based on the allowable range of grid connection point voltage deviation and voltage fluctuation. This forms a set of constraint terms including energy storage SOC constraints, frequency change rate constraints, and node voltage constraints. Then, the disturbance characteristics corresponding to the type labels are matched with a preset mapping rule base to obtain the dominant constraints for virtual synchronous generator control. If the same type label is simultaneously... When matching multiple constraint categories, the constraint with the most severe exceedance is selected as the final dominant constraint. For example, if the measured node voltage is 108% of the nominal voltage, exceeding the upper limit of the voltage by 110%, the exceedance is considered moderate, while the current state of charge (SNC) value of 85% is outside the safe range of 20% to 80%, the exceedance is considered mild. In this case, the node voltage constraint is selected as the dominant constraint. The mapping rule base pre-stores the correspondence between disturbance categories and dominant constraints. For example, when the type label contains any of the following: transient rise, transient fall, or interruption, the corresponding node voltage constraint is selected as the dominant constraint; when the type label contains frequency deviation or frequency fluctuation, the corresponding frequency change rate constraint is selected as the dominant constraint; when the type label contains harmonics or oscillatory transients, the corresponding energy storage SNC constraint is selected as the dominant constraint.

[0037] In some embodiments, determining the frequency band weighting coefficients of each constraint term based on the attention weight vector can be achieved using the following steps: The target frequency band set and the candidate frequency band set are determined based on the attention weight vector; The frequency band dependency weighting coefficients for each constraint term are determined based on the target frequency band set and the candidate frequency band set.

[0038] It should be noted that, in this application, the target frequency band set is a set of frequency sub-bands reflecting the main disturbance characteristics in composite power quality disturbances. It can guide the energy storage converter to prioritize the allocation of control resources to constraint terms strongly correlated with the dominant disturbance, thereby improving the speed and specificity of the disturbance response. The candidate frequency band set is a set of frequency sub-bands carrying the secondary disturbance characteristics or noise components in composite power quality disturbances. It can assist the energy storage converter in compensating for the node voltage constraint terms, thereby avoiding voltage over-limit or insufficient fluctuation suppression due to ignoring low-frequency fluctuation components. The frequency band-dependent weighting coefficient is a numerical parameter that quantifies the degree of influence of different frequency sub-bands on the energy storage state of charge constraint, frequency change rate constraint, and node voltage constraint. It is used to dynamically adjust the contribution ratio of each constraint term in the weighted fusion calculation, thereby achieving adaptive optimization configuration of virtual inertia and power factor angle.

[0039] In practice, firstly, all elements in the attention weight vector are sorted in descending order of their weight values ​​to obtain a frequency band weight sorting sequence with high priority. Then, based on a pre-set weight threshold, such as twice the average weight of all elements, frequency bands with weight values ​​greater than or equal to the threshold are selected from this frequency band weight sorting sequence and added to the target frequency band set. Meanwhile, the remaining frequency bands with weight values ​​less than the threshold are added to the candidate frequency band set. Next, for the energy storage state of charge constraint and the frequency change rate constraint, the attention weight values ​​corresponding to each frequency sub-band in the target frequency band set are multiplied element-by-element by the pre-set frequency band-constraint mapping matrix. The results of the element-by-element multiplication are summed to obtain the frequency band dependence weighting coefficients for the energy storage state of charge constraint and the frequency change rate constraint. For example, when the target frequency band set contains low-frequency components from 0 to 150 Hz and these low-frequency components are approximately related to the energy storage state of charge... When the constraint terms have high correlation, the corresponding multiplication coefficient in the mapping matrix takes a value greater than one, thereby selectively enhancing the weight influence of a specified frequency band for different constraint terms. For the node voltage constraint term, since the node voltage constraint term mainly focuses on the frequency band containing the fundamental frequency component and the low-frequency fluctuation component, and the candidate frequency band set mainly contains frequency bands with low weights, a piecewise compensation mapping method is adopted to perform numerical mapping transformation on the frequency bands in the candidate frequency band set whose weights are lower than the average value. That is, the attention weight value of each frequency band in the candidate frequency band set is multiplied by a compensation coefficient to obtain the compensation weight vector of the candidate frequency band set. Then, all elements in the compensation weight vector are summed, and the summation result is used as the frequency band dependence weighting coefficient of the node voltage constraint term. The compensation coefficient is pre-calibrated according to the actual correlation between the frequency band and the node voltage fluctuation. For example, the compensation coefficient of the low-frequency fluctuation frequency band below 50 Hz is set to two to make up for its underestimated weight in the attention weight vector.

[0040] In some embodiments, prioritizing the weighting coefficients of each frequency band according to the dominant constraint to obtain the modulated weighting coefficient set can be achieved by the following steps: Priority modulation is performed on the weighting coefficients of each frequency band using the dominant constraints to obtain the modulated weighting coefficients corresponding to each constraint term; The set of modulated weighted coefficients is determined based on all the modulated weighted coefficients.

[0041] It should be noted that, in this application, the modulated weighting coefficient is a weight parameter that quantifies the contribution of each constraint term in the subsequent weighted fusion calculation. It enables the energy storage control strategy to adaptively adjust the dominant position of each constraint term according to the type of composite disturbance, thereby achieving differentiated control under disturbance scenario perception. The weighting coefficient set is a set that includes the modulated weighting coefficients of all constraint terms.

[0042] In specific implementation, firstly, a preset proportional priority modulation technique is used to normalize the priority weight mapping of the frequency band weighting coefficients of each constraint item. The frequency band weighting coefficients of the dominant constraint are assigned the highest normalized priority weight, while the frequency band weighting coefficients of non-dominant constraints are allocated the remaining normalized priority weights according to the proportional relationship of the original values ​​of the frequency band weighting coefficients. For example, when the type label identification result is a transient disturbance, the dominant constraint is selected as the node voltage constraint. Then, the frequency band weighting coefficients corresponding to the node voltage constraint are assigned the highest normalized priority weight of 0.6. The frequency band weighting coefficients corresponding to the energy storage state of charge constraint and the frequency change rate constraint are allocated the remaining normalized priority weight of 0.4 according to the ratio of their original values ​​of 0.2 to 0.2. In this way, the frequency band weighting coefficients of each constraint item are transformed into normalized priority weights carrying priority modulation information, and the modulated weighting coefficients of each constraint item are obtained. Then, the modulated weighting coefficients of each constraint item are assembled into a unified data storage structure in the form of key-value pairs with one-to-one correspondence between constraint item identifiers and modulated weighting coefficient values, and the set of modulated weighting coefficients is obtained.

[0043] In step 104, the constraint terms are weighted and fused based on the weighted coefficient set to obtain the virtual inertia and power factor angle, and then the power grid is compensated for power quality based on the virtual inertia and the power factor angle.

[0044] In some embodiments, the virtual inertia and power factor angle can be obtained by weighting and fusing the constraint terms based on the weighted coefficient set using the following steps: Based on the weighted coefficient set, the frequency change rate constraint and the energy storage state of charge constraint are weighted and fused to obtain the virtual inertia. The power factor angle is obtained by weighting and fusing the node voltage constraints based on the weighted coefficient set.

[0045] It should be noted that in this application, virtual inertia is a control parameter that simulates the mechanical inertia of the rotor of a traditional synchronous generator in virtual synchronous generator control. It is used to characterize the response capability of the energy storage converter to actively adjust the output of active power when the grid frequency fluctuates. The power factor angle is the phase difference angle between the output voltage and the output current of the energy storage converter. It is used to characterize the regulation capability of the energy storage system to inject or absorb reactive power into the grid.

[0046] In specific implementation, firstly, the first sub-band weighted coefficient corresponding to the frequency change rate constraint term in the weighted coefficient set is multiplied by the frequency change rate constraint value, and the second sub-band weighted coefficient corresponding to the energy storage state of charge constraint term is multiplied by the energy storage state of charge constraint value. Then, the two products are added together using a linear weighted summation method to obtain the virtual inertia required for virtual synchronous generator control. The frequency change rate constraint value can be mapped to a constraint strength value using an exponential function model, and the energy storage state of charge constraint value can be mapped to a constraint strength value using a logistic regression function model. This automatically reduces the constraint strength to protect battery life when the energy storage approaches overcharge or over-discharge. Then, the third sub-band weighted coefficient corresponding to the node voltage constraint term in the weighted coefficient set is multiplied by the node voltage deviation constraint value, and based on... The power factor angle is obtained through mapping calculation using a five-segment adaptive power factor angle adjustment law. This law pre-divides the node voltage range into five segments: severely low voltage, low voltage, normal voltage, high voltage, and severely high voltage. Each segment corresponds to a preset power factor angle calculation function. For example, when the node voltage is in the severely high voltage segment, the power factor angle takes its maximum negative value to allow the energy storage inverter to absorb reactive power and thus lower the voltage. When the node voltage is in the severely low voltage segment, the power factor angle takes its maximum positive value to allow the energy storage inverter to inject reactive power and thus raise the voltage. Furthermore, the per-unit value of the voltage deviation can be nonlinearly transformed according to a pre-set voltage deviation-constraint strength mapping function to obtain the node voltage deviation constraint value.

[0047] In some embodiments, power quality compensation of the power grid based on the virtual inertia and the power factor angle can be achieved by the following steps: The virtual inertia and the power factor angle are input to the virtual synchronous generator control module of the energy storage converter to calculate the active power reference value and reactive power reference value of the energy storage converter. A modulation wave signal is generated based on the active power reference value and the reactive power reference value; The modulated wave signal drives the power semiconductor devices of the energy storage converter to perform charging and discharging operations.

[0048] It should be noted that in this application, the active power reference value is the target command value for controlling the magnitude of active power exchange between energy storage and the grid; the reactive power reference value is the target command value for controlling the magnitude of reactive power exchange between energy storage and the grid; the modulation wave signal is a sinusoidal waveform voltage command that drives the power semiconductor device to turn on and off, which can accurately convert the changes in virtual inertia and power factor angle into the actual charging and discharging actions of the energy storage converter, thereby realizing integrated closed-loop compensation for power quality fusion assessment and energy storage operation management.

[0049] In practice, firstly, the virtual inertia and power factor angle are input to the virtual synchronous generator control module of the energy storage converter. The virtual inertia is substituted into the synchronous generator rotor motion equation via the active power control loop, and combined with the power-frequency droop characteristic, to calculate the active power reference value of the energy storage converter. Simultaneously, the power factor angle is substituted into the excitation regulation equation via the reactive power control loop, and combined with the reactive power voltage droop characteristic, to calculate the reactive power reference value of the energy storage converter. Then, the active and reactive power reference values ​​are input to the virtual impedance control loop, which simulates the stator resistance and synchronous reactance of the synchronous generator to generate a virtual voltage drop. The original voltage command, superimposed with the virtual voltage drop, is input to the voltage outer loop proportional-integral controller, which adjusts the output voltage amplitude. Finally, the output of the voltage outer loop proportional-integral controller, along with the grid-connected current feedback value, is input to the current inner loop proportional-integral controller, which generates the voltage reference command value. The voltage reference command value is then subjected to sinusoidal pulse width modulation (PWM) transformation to generate a sinusoidal modulated wave signal. Finally, the modulated wave signal and the triangular carrier signal are input to a pulse width modulation comparator, which performs real-time amplitude comparison between the modulated wave signal and the triangular carrier signal. When the amplitude of the modulated wave signal is greater than that of the triangular carrier signal, a high-level switching control pulse is output; when the amplitude of the modulated wave signal is less than or equal to that of the triangular carrier signal, a low-level switching control pulse is output. The switching control pulse sequence consisting of the high-level and low-level switching control pulses is applied as a drive signal to the control electrode of the power semiconductor device in the energy storage converter. The power semiconductor device is controlled to switch on and off according to the frequency of the switching control pulse sequence, so that the DC-side energy storage battery of the energy storage converter and the AC-side grid exchange energy according to the active power reference value and the reactive power reference value. This achieves adaptive power quality compensation for frequency fluctuations and voltage over-limits of the grid based on virtual inertia and power factor angle.

[0050] Furthermore, in another aspect of this application, in some embodiments, this application provides an energy storage system operation and management system for power quality fusion assessment, with reference to... Figure 4The figure is a schematic diagram of the operation and management system of the energy storage system for power quality fusion assessment according to some embodiments of this application, including: frequency domain decoupling module 201, constraint modulation module 202 and compensation module 203, which are described below: Frequency domain decoupling module 201: In this application, the frequency domain decoupling module 201 is mainly used to acquire the voltage timing signal and the current timing signal at the grid connection point, and to perform frequency domain decoupling on the voltage timing signal and the current timing signal to obtain multiple frequency sub-band components. The constraint modulation module 202 in this application is used to perform hierarchical feature extraction on all frequency sub-band components to obtain a deep feature map, and to perform channel attention self-calibration and weighted classification mapping based on the deep feature map to obtain the type label of composite power quality disturbance and the attention weight vector of all frequency sub-bands. It should be noted that the constraint modulation module 202 is also used to determine the dominant constraint of virtual synchronous generator control according to the type label, determine the frequency band weighting coefficient of each constraint item according to the attention weight vector, and then perform priority modulation on each frequency band weighting coefficient through the dominant constraint to obtain the set of modulated weighting coefficients. The compensation module 203 in this application is mainly used to perform weighted fusion of constraint terms based on the weighted coefficient set to obtain virtual inertia and power factor angle, and then perform power quality compensation on the power grid based on the virtual inertia and the power factor angle.

[0051] The foregoing has detailed examples of the energy storage system operation management method and system for power quality fusion assessment provided in the embodiments of this application. It is understood that the corresponding apparatus, in order to achieve the above functions, includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specified application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specified application, but such implementation should not be considered beyond the scope of this application.

[0052] In some embodiments, this application also provides a computer device, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for calling and running the computer program from the memory, so that the computer device performs the above-described energy storage system operation management method for power quality fusion assessment.

[0053] In some embodiments, reference Figure 5The dashed lines in the figure indicate that the unit or module is optional. This figure is a structural schematic diagram of a computer device for implementing a power quality fusion assessment method for energy storage system operation and management according to an embodiment of this application. The power quality fusion assessment method for energy storage system operation and management described in the above embodiments can be achieved through… Figure 5 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a memory 302 and at least one communication unit 305. The computer device may be a terminal device, a server or a chip.

[0054] Processor 301 can be a general-purpose processor or a special-purpose processor. For example, processor 301 can be a central processing unit (CPU), which can be used to control computer devices, execute software programs, and process data from software programs. The computer device may also include a communication unit 305 for inputting (receiving) and outputting (transmitting) signals.

[0055] For example, the computer device may be a chip, and the communication unit 305 may be the input and / or output circuit of the chip, or the communication unit 305 may be the communication interface of the chip, which may be a component of a terminal device, network device or other device.

[0056] For example, the computer device may be a terminal device or a server, and the communication unit 305 may be a transceiver of the terminal device or the server, or the communication unit 305 may be a transceiver circuit of the terminal device or the server.

[0057] The computer device may include one or more memories 302 storing a program 304. The program 304 can be executed by a processor 301 to generate instructions 303, causing the processor 301 to execute the method described in the above method embodiments according to the instructions 303. Optionally, the memory 302 may also store data (such as a target audit model). Optionally, the processor 301 may also read data stored in the memory 302, which may be stored at the same storage address as the program 304, or it may be stored at a different storage address than the program 304.

[0058] The processor 301 and memory 302 can be configured separately or integrated together, for example, integrated on the system on chip (SOC) of the terminal device.

[0059] It should be understood that each step of the above method embodiment can be completed by hardware logic circuits or software instructions in the processor 301. The processor 301 can be a CPU, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, such as discrete gates, transistor logic devices, or discrete hardware components.

[0060] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] For example, in some embodiments, this application also provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the above-described energy storage system operation management method for power quality fusion assessment.

[0062] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0063] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for the operation and management of an energy storage system based on integrated power quality assessment, applied to an energy storage converter controller, characterized in that, The method includes the following steps: The voltage and current timing signals at the grid connection point are acquired, and the voltage and current timing signals are decoupled in the frequency domain to obtain multiple frequency sub-band components. Hierarchical feature extraction is performed on all frequency sub-band components to obtain deep feature maps. Based on the deep feature maps, channel attention self-calibration and weighted classification mapping are performed to obtain the type label of composite power quality disturbance and the attention weight vector of all frequency sub-bands. The dominant constraints for virtual synchronous generator control are determined based on the type label, and the frequency band weighting coefficients for each constraint item are determined based on the attention weight vector. Then, the frequency band weighting coefficients are preferentially modulated through the dominant constraints to obtain a set of modulated weighting coefficients. The constraint terms are weighted and fused based on the weighted coefficient set to obtain the virtual inertia and power factor angle, and then the power grid is compensated for power quality based on the virtual inertia and the power factor angle. The frequency band weighting coefficients of each constraint term are determined based on the attention weight vector using the following steps: The target frequency band set and the candidate frequency band set are determined based on the attention weight vector; The frequency band dependency weighting coefficients for each constraint term are determined based on the target frequency band set and the candidate frequency band set. The frequency band-dependent weighting coefficient is a numerical parameter that quantifies the degree of influence of different frequency sub-bands on the energy storage state of charge constraint, frequency change rate constraint, and node voltage constraint. Specifically, the virtual inertia and power factor angle are obtained by weighting and fusing the constraint terms based on the weighted coefficient set, including: Based on the weighted coefficient set, the frequency change rate constraint and the energy storage state of charge constraint are weighted and fused to obtain the virtual inertia. The power factor angle is obtained by weighting and fusing the node voltage constraints based on the weighted coefficient set.

2. The method as described in claim 1, characterized in that, Frequency domain decoupling of the voltage timing signal and the current timing signal to obtain multiple frequency sub-band components specifically includes: The voltage timing signal and the current timing signal are converted into voltage frequency domain coefficient sequences and current frequency domain coefficient sequences, respectively. Synchronous sliding sampling is performed on the voltage frequency domain coefficient sequence and the current frequency domain coefficient sequence to obtain multiple voltage frequency band spectrum sequences and current frequency band spectrum sequences covering different frequency ranges respectively; All voltage and current frequency spectrum sequences are converted back to the time domain to obtain multiple frequency sub-band components.

3. The method as described in claim 1, characterized in that, Hierarchical feature extraction is performed on all frequency sub-band components to obtain a deep feature map, specifically including: All frequency subband components are input into a hierarchical feature extraction network, and the first energy spectrum is obtained by local feature mapping through the first-level residual block; The receptive field of the first energy spectrum is gradually expanded by the second-level residual block to obtain the second energy spectrum; The deep feature map is obtained by capturing the long-range dependencies between the frequency band components in the second energy spectrum through the third-level residual block.

4. The method as described in claim 1, characterized in that, Based on the deep feature map, channel attention self-calibration and weighted classification mapping are performed to obtain the type label of composite power quality perturbation and the attention weight vector of all frequency sub-bands, specifically including: Multi-level channel attention self-calibration is performed on the deep feature map to obtain a channel importance-enhanced feature map; Residual feature extraction is performed on the channel importance enhancement feature map to obtain a residual refined feature map; Channel-level feature aggregation is performed on the residual refined feature map to generate multi-class original features; The original feature vectors of the multi-classification are input into the perturbation classifier for dual-path parallel mapping to obtain the type label of the composite power quality perturbation and the attention weight vectors of all frequency sub-bands.

5. The method as described in claim 1, characterized in that, The dominant constraints for virtual synchronous generator control, determined based on the type label, specifically include: Determine the set of constraints for virtual synchronous generator control; Based on the type label, select a dominant constraint for virtual synchronous generator control from the set of constraints.

6. The method as described in claim 1, characterized in that, Power quality compensation for the power grid based on the virtual inertia and the power factor angle specifically includes: The virtual inertia and the power factor angle are input to the virtual synchronous generator control module of the energy storage converter to calculate the active power reference value and reactive power reference value of the energy storage converter. A modulation wave signal is generated based on the active power reference value and the reactive power reference value; The modulated wave signal drives the power semiconductor devices of the energy storage converter to perform charging and discharging operations.

7. A power quality fusion assessment system for energy storage system operation and management, used to execute the power quality fusion assessment system operation and management method for energy storage systems as described in any one of claims 1 to 6, characterized in that, The system includes: The frequency domain decoupling module is used to acquire the voltage timing signal and the current timing signal at the grid connection point, and to perform frequency domain decoupling on the voltage timing signal and the current timing signal to obtain multiple frequency sub-band components; The constrained modulation module is used to extract hierarchical features from all frequency sub-band components to obtain a deep feature map. Based on the deep feature map, channel attention self-calibration and weighted classification mapping are performed to obtain the type label of the composite power quality disturbance and the attention weight vector of all frequency sub-bands. The constraint modulation module is further configured to determine the dominant constraint of virtual synchronous generator control based on the type label, determine the frequency band weighting coefficient of each constraint item based on the attention weight vector, and then perform priority modulation on each frequency band weighting coefficient through the dominant constraint to obtain a set of modulated weighting coefficients. The compensation module is used to perform weighted fusion of the constraint terms based on the weighted coefficient set to obtain the virtual inertia and power factor angle, and then perform power quality compensation on the power grid based on the virtual inertia and the power factor angle.

8. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the energy storage system operation management method for power quality fusion assessment as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the energy storage system operation management method for power quality fusion assessment as described in any one of claims 1 to 6.

Images