Network-constructed energy storage transient voltage active support regulation method with multi-working condition state perception
By using multi-condition state perception and dual-layer composite control, combined with reinforcement learning algorithms to optimize parameters, the problem of fixed parameters and insufficient coordination in voltage support control of grid-type energy storage is solved, improving the voltage stability of the power grid and the adaptability of the energy storage system, and realizing the smoothness of voltage recovery and equipment safety.
Patent Information
- Application Number
- CN202611380498.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-07
- Publication Date
- 2026-11-03
AI Technical Summary
Existing grid-type energy storage voltage support control has problems such as fixed parameters that cannot adapt to various types of disturbance conditions, single support dimension, insufficient coordination between inertia and reactive power, crude allocation strategy for multiple energy storage units, lack of closed-loop evaluation and self-learning optimization capabilities, abrupt mode switching that is prone to secondary impacts, and insufficient fine coordination between protection and support.
A multi-condition state-aware grid-type energy storage transient voltage active support and control method is adopted. By collecting grid connection point voltage and current data in real time, multi-dimensional feature quantities are extracted, the operating condition type is identified based on hierarchical judgment rules, and the corresponding control parameters are dynamically retrieved. A two-layer composite control architecture and multiple protection coordination are adopted, and reinforcement learning algorithms are combined to optimize parameters and smoothly switch modes, so as to achieve differentiated support and self-optimization.
It enables dynamic adjustment of control parameters based on actual operating conditions, improving the voltage stability and anti-disturbance capability of the power grid, enhancing the adaptability and overall efficiency of the energy storage system, and ensuring the smoothness of voltage recovery and the safety of the equipment.
Smart Images
Figure CN123034479A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power system energy storage control technology, specifically to a method for active support and regulation of transient voltage in grid-type energy storage with multi-condition state perception. Background Technology
[0002] With the continuous increase in the installed capacity of new energy power generation, the proportion of synchronous generators in the power system is declining, and the system inertia and short-circuit capacity are continuously decreasing, posing a severe challenge to the voltage stability and disturbance rejection capability of the power grid. Grid-connected energy storage, as a flexible grid-connected device that can simulate the external characteristics of synchronous generators, has the ability to provide active voltage support and inertia, and is an important technical means to improve the transient voltage stability of power grids with a high proportion of new energy.
[0003] Grid-based energy storage, through control strategies that simulate the operating characteristics of synchronous generators, can proactively provide reactive power and inertia support when disturbances such as voltage dips occur in the grid, helping the voltage at the grid connection point to recover quickly. However, existing grid-based energy storage voltage support control still has many shortcomings, such as fixed parameters that cannot adapt to various types of disturbance conditions; a single support dimension and insufficient coordination between inertia and reactive power; a crude allocation strategy for multiple energy storage units, resulting in underutilization of overall efficiency; a lack of closed-loop evaluation and self-learning optimization capabilities; abrupt mode switching that easily generates secondary impacts; and insufficient fine-grained coordination between protection and support. Summary of the Invention
[0004] To address the aforementioned technical problems, this application provides a multi-condition state-sensing method for active support and regulation of transient voltage in grid-type energy storage, thereby resolving the existing issues.
[0005] The multi-condition sensing method for active support and control of transient voltage in grid-type energy storage adopts the following technical solution: One embodiment of this application provides a method for active support and control of transient voltage in grid-type energy storage with multi-condition state awareness, including the following steps: S1: Real-time acquisition of grid connection point voltage and current data, extraction of multi-dimensional features of transient voltage disturbances, and identification of the current disturbance type based on hierarchical judgment rules; S2: Based on the identified operating condition type, dynamically retrieve three sets of core control parameters—virtual inertia, damping coefficient, and reactive power support gain—from the parameter matching library as the initial parameters for support control. S3: It adopts a dual-layer composite control architecture that provides virtual inertia support through outer loop virtual synchronous control and voltage support through inner loop fast reactive current injection, and simultaneously outputs active and reactive components to support the transient voltage at the grid connection point. S4: Calculate the comprehensive weight based on the three indicators of remaining capacity, response rate and battery capacity retention rate of each energy storage unit, and allocate the total supporting power demand to each energy storage unit according to the weight ratio; S5: Calculate the support effect score during the transient process, and use the support effect score as feedback to iteratively correct the parameter matching library to achieve continuous self-optimization of support performance.
[0006] Preferably, in step S1, the positive-sequence component and negative-sequence component of the voltage fundamental are extracted, and multi-dimensional characteristic quantities are calculated, including voltage sag depth, the ratio of the effective value of the positive-sequence component of the voltage to the rated voltage; disturbance duration, the length of time the voltage deviates from the rated value within ±10%; and voltage imbalance, the percentage of the negative-sequence voltage component to the positive-sequence voltage component.
[0007] Preferably, the step of identifying the operating condition type of the current disturbance based on the hierarchical judgment rule includes: slight voltage sag: sag depth The criteria for determining voltage sags are as follows: 1. PU and unbalance < 4%; 2. Severe voltage sag: sag depth < 0.7 PU and unbalance < 4%; 3. Voltage swell: voltage amplitude > 1.1 PU; 4. Unbalance fault: unbalance ≥ 4%. The priority of determination is as follows: first, determine the unbalance fault based on the unbalance degree; then, determine whether it is a voltage sag or swell based on the voltage amplitude; and finally, determine the degree of sag based on the sag depth.
[0008] Preferably, the dual-layer composite control architecture specifically includes: The control architecture consists of two layers: an outer loop and an inner loop. The outer loop is a virtual synchronous control layer that simulates the rotor motion equation of a synchronous generator and outputs an active power reference value through the active power-frequency droop characteristic. The inner loop is a fast reactive power control layer that outputs a reactive current reference value based on the grid connection point voltage deviation via a proportional-integral regulator and directly modulates the reactive power output of the converter.
[0009] Preferably, the transient process virtual inertia adaptively adjusts, specifically including: the virtual inertia is dynamically adjusted in real time according to the voltage drop depth during the transient process; the adjustment formula is as follows: Where J0 is the reference inertia corresponding to the current operating condition, ΔU is the voltage drop depth, which is the difference between the rated voltage and the actual voltage, and k is the inertia adjustment coefficient.
[0010] Preferably, in step S4, three indicators are obtained for each energy storage unit: remaining capacity, response rate, and battery capacity retention rate. The three indicators are weighted and summed using their respective weights to obtain the comprehensive weight of each energy storage unit. The supporting power allocated to each energy storage unit is equal to the total supporting power multiplied by the comprehensive weight of that energy storage unit and then divided by the sum of the comprehensive weights of all energy storage units.
[0011] Preferably, the support effect score specifically includes: Three support effectiveness evaluation indicators were calculated: voltage recovery time, which is the time it takes for the voltage at the grid connection point to recover to 0.9 per unit value from the moment the disturbance occurs; sag suppression ratio, which is the difference between the maximum sag depth and the expected sag depth without support, divided by the sag depth without support; and voltage overshoot rate, which is the proportion of the highest voltage exceeding the rated value during the recovery process. The support effectiveness score was obtained by normalizing the three indicators and then weighting and summing them.
[0012] Preferably, the iterative correction of the parameter matching library based on the support effect score specifically includes: The Q-learning reinforcement learning algorithm is used to achieve adaptive optimization of parameters. The state space consists of four types of working conditions, and the action space consists of the up and down adjustment of each set of parameters. The adjustment step size is a preset percentage of the baseline value. The reward function is the support effect score. After the cumulative learning samples reach a preset number of times, the optimized parameters are updated to the parameter matching library.
[0013] Preferably, it also includes a multi-mode smooth switching strategy, including: The system is set with three operating modes: normal grid connection mode, transient support mode, and fault ride-through mode. Mode switching adopts hysteresis comparison method. When the voltage is higher than the first preset voltage and lasts for a preset duration, it switches to normal mode. When the voltage is lower than the second preset voltage, it switches to transient support mode. When the voltage is lower than the third preset voltage, it switches to fault ride-through mode. During the switching process, the control parameters use a ramp function to smoothly transition.
[0014] Preferably, it also includes multi-constraint protection coordination, including: The triple protection constraints are set and executed in order of priority: the first priority is overcurrent protection, which immediately limits the current when the output current exceeds a preset multiple of the rated current; the second priority is overvoltage protection, which gradually reduces the reactive power injection in a linear proportional manner until it exits support when the grid connection point voltage exceeds a preset value; the third priority is SOC boundary protection, which exits active power support and only retains reactive power support when the SOC of the energy storage unit is lower than a preset first percentage, and exits charging mode when it is higher than a preset second percentage.
[0015] This application has at least the following beneficial effects: (1) Systematic innovation of the five-layer closed-loop control architecture. This application constructs a closed-loop control architecture of operating condition perception, parameter matching, two-layer control, optimized allocation, and self-learning correction, forming a complete technical chain from disturbance identification to parameter optimization. Unlike the fixed parameter mode of traditional open-loop control, this method first perceives the difference in operating conditions, then matches the corresponding parameters, executes coordinated control, optimizes power allocation, and finally evaluates the effect and corrects it in reverse. Each link is progressive and iterates in a closed loop. The closed-loop architecture enables the control system to dynamically adjust its behavior according to the actual operating conditions and support effect, and has the ability to adapt and evolve. The overall performance is far superior to the fixed parameter open-loop control scheme.
[0016] (2) Innovative Differentiated Control Based on Multi-Condition Sensing and Parameter Adaptive Matching. Power grid disturbances are diverse, and the support requirements for different operating conditions vary fundamentally, making it impossible to address all issues with fixed parameters. This application proposes a multi-dimensional disturbance feature extraction method, characterizing disturbance features from four dimensions: sag depth, duration, imbalance, and phase jump, and classifying them into four typical operating conditions. Furthermore, an operating condition-parameter mapping library is established, matching different inertia, damping, and reactive power gain parameters to different operating conditions to achieve differentiated support strategies. Increased support is applied during severe sags, while stability is emphasized during mild sags, and negative sequence control is strengthened during imbalance faults. Operating condition adaptive matching ensures that control parameters always match the current disturbance requirements, avoiding both insufficient support and oscillations caused by excessive support.
[0017] (3) Innovative Multi-Dimensional Collaborative Support of Dual-Layer Composite Control. Transient voltage recovery requires both rapid reactive power injection to boost the amplitude and active inertia support to maintain dynamic stability, with different time scales. This application designs a dual-layer composite architecture with an outer-loop virtual synchronous control and an inner-loop rapid reactive power control. The outer loop provides virtual inertia and damping on a time scale of hundreds of milliseconds to ensure the dynamic stability of the system; the inner loop rapidly injects reactive power on a time scale of milliseconds to directly support the voltage amplitude. The two layers of control perform their respective functions and output collaboratively, providing both rapid voltage amplitude response and a smooth dynamic transition process, solving the problem of limited support effect of a single dimension, and significantly improving support performance compared to the traditional single control architecture.
[0018] (4) Refined Control Innovation with Adaptive Adjustment of Transient Inertia. Virtual inertia is a core parameter of network-type control, and fixed inertia is difficult to adapt to the constantly changing voltage during transient processes. This application proposes an adaptive law that adjusts inertia in real time according to the voltage drop depth. The deeper the drop, the greater the inertia, and it drops synchronously after the voltage recovers. The adaptive inertia ensures that the maximum inertia support is obtained at the most severe moment of the transient, while the inertia naturally decreases when the voltage is close to recovery, avoiding overshoot caused by excessive support. Compared with the fixed inertia scheme, the adaptive inertia can significantly accelerate the voltage recovery speed without increasing overshoot, achieving a better balance between support strength and stability.
[0019] (5) Innovative power allocation mechanism for multiple energy storage units with weighted multi-indicator analysis. Energy storage power stations consist of multiple converters connected in parallel, and the remaining capacity, response rate, and battery capacity retention rate of each energy storage unit are different. Traditional average allocation or allocation by capacity does not take advantage of these differences, and the overall efficiency is not fully utilized. This application establishes a comprehensive weighted allocation model that includes three indicators: remaining capacity, response rate, and battery capacity retention rate. Energy storage units with larger capacities undertake more support tasks, energy storage units with faster responses take priority in handling transient peak loads, and energy storage units with higher battery capacity retention rates take on a larger share. The weighted allocation of the three indicators enables the optimal utilization of the capabilities of each energy storage unit, and improves the overall support capability and service life.
[0020] (6) Innovation of parameter self-optimization mechanism based on reinforcement learning. The theoretically designed control parameters are often not optimal in actual power grids and need to be adjusted according to the field conditions. This method introduces the Q-learning reinforcement learning algorithm, using the support effect score of each transient process as a reward to automatically iteratively optimize the control parameters under each operating condition. With the accumulation of operating experience, the parameters are continuously adjusted towards a better direction, and the support performance is continuously improved. The self-learning mechanism enables the control system to have the ability to continuously evolve, adapt to the slow changes in power grid characteristics, and improve long-term operating performance instead of decreasing it.
[0021] (7) Transient-friendly innovation in smooth multi-mode switching. Abrupt switching between different operating modes can easily cause secondary shocks. This application adopts a switching strategy of hysteresis comparison plus ramp transition. Hysteresis comparison avoids frequent back-and-forth switching when voltage fluctuates, and ramp transition ensures smooth changes in parameters and output power. The mode switching process is grid-friendly and does not introduce new transient disturbances, which helps the voltage to recover smoothly.
[0022] (8) Innovative Safe Operation with Coordinated Multiple Protection and Support. The effective use of support capabilities must be based on equipment safety. This method establishes a three-tiered priority protection coordination mechanism: overcurrent protection has the highest priority to ensure device safety, overvoltage protection is the next highest to prevent excessive voltage, and SOC protection is the lowest, taking into account both support capabilities and battery life. All protections operate with smooth limiting, preserving as much support capability as possible while limiting output, rather than simply shutting down all protections. This precise coordination between protection and support maximizes support capabilities within the safety boundary. Attached Figure Description
[0023] Figure 1 A flowchart illustrating the steps of the multi-condition state-aware grid-type energy storage transient voltage active support and control method provided in this application. Detailed Implementation
[0024] The following description, in conjunction with the accompanying drawings, details the specific scheme of the active support and control method for transient voltage of grid-type energy storage with multi-condition state sensing provided in this application.
[0025] This application provides an embodiment of a multi-condition state-aware network-type energy storage transient voltage active support and control method. For details, please refer to [link to specific implementation details]. Figure 1 This includes the following steps: Step S1: Collect grid connection point voltage and current data in real time, extract multi-dimensional features of transient voltage disturbances, and identify the operating condition type of the current disturbance based on the hierarchical judgment rules.
[0026] This step serves as an input to the control process, extracting disturbance characteristics and identifying operating condition types in real time to provide a basis for subsequent differentiated parameter matching. Accurate operating condition identification is a prerequisite for achieving adaptive control.
[0027] The grid-connected point voltage and current are fed into the digital controller via a sampling circuit. A phase-locked loop algorithm based on a second-order generalized integrator is used to extract the positive-sequence and negative-sequence components of the fundamental voltage in real time. Based on this, multidimensional characteristic quantities are calculated, including: The first item is the voltage sag depth. It is the ratio of the effective value of the positive sequence voltage component to the rated voltage, expressed as a per-unit value. The sag depth is the most intuitive and important indicator of disturbance intensity, directly determining the required support strength.
[0028] The second item is the duration of the disturbance. This refers to the length of time the voltage deviates from the rated value by ±10%. In this embodiment, the timing begins when the voltage amplitude drops out of the normal range of 0.9 to 1.1 per unit and ends when the voltage returns to the normal range. This timing period is the duration. The duration reflects the severity and energy scale of the disturbance.
[0029] The third item is voltage unbalance. It calculates the ratio of the effective value of the negative-sequence voltage component to the effective value of the positive-sequence voltage component, expressed as a percentage. The support strategies for three-phase unbalanced faults differ from those for three-phase symmetrical faults; the unbalance degree is a key indicator for distinguishing fault types.
[0030] Furthermore, based on various characteristic quantities, the operating conditions are classified according to the preset classification rules: slight voltage sag: sag depth The criteria for determining voltage sags are as follows: 1. PU and unbalance <4%; 2. Severe voltage sag: sag depth <0.7 PU and unbalance <4%; 3. Voltage swell: voltage amplitude >1.1 PU; 4. Unbalance fault: unbalance ≥4%. The priority for determining the fault is: first, determine the unbalance fault based on the unbalance degree; then, determine whether it is a voltage sag or swell based on the voltage amplitude; finally, determine the degree of sag based on the sag depth.
[0031] Step S2: Based on the identified operating condition type, dynamically retrieve the three sets of core control parameters—virtual inertia, damping coefficient, and reactive power support gain—from the parameter matching library for the corresponding operating condition, as the initial parameters for support control.
[0032] This step retrieves the corresponding control parameters from a pre-stored parameter matching library based on the identified operating condition type, thereby matching the support strength with the disturbance level. Differentiated parameter tuning is the core step in improving operating condition adaptability.
[0033] The parameter matching library pre-stores three sets of core control parameters corresponding to four operating conditions: virtual inertia J, damping coefficient D, and reactive power support gain Kq. The parameter baseline values are initially designed based on the rated capacity of energy storage and the grid short-circuit ratio, and are subsequently continuously optimized through self-learning.
[0034] The parameter settings for each operating condition are as follows: Mild voltage dip: The voltage drop is not deep, and the focus of support is on stability rather than strength. The virtual inertia is set to 1.0 times the baseline value, the damping coefficient to 1.2 times the baseline value (to increase damping and suppress oscillations), and the reactive power gain to 1.0 times the baseline value. This set of parameters emphasizes stability and avoids excessive support under small disturbances that could cause oscillations.
[0035] Severe voltage dips require the strongest possible support. The virtual inertia is set to 1.8 times the baseline value to provide stronger active inertia support; the damping coefficient is set to 0.8 times the baseline value to appropriately reduce damping and accelerate response; and the reactive power gain is set to 1.5 times the baseline value to significantly enhance reactive power injection capability. This set of parameters maximizes support strength while ensuring stability.
[0036] Voltage spurt condition: When the voltage is too high, reactive power needs to be absorbed to reduce the voltage. The virtual inertia is taken as 0.8 times the reference value, the damping coefficient is taken as 1.5 times the reference value, and the reactive power gain is taken as a negative value, that is, absorbing inductive reactive power and generating capacitive reactive power, which plays a role in reducing the voltage.
[0037] Unbalanced fault condition: An additional negative sequence reactive power control channel is added on the basis of positive sequence parameters. The negative sequence reactive power gain is 1.2 times that of the positive sequence gain, which is specifically used to suppress negative sequence voltage components and improve three-phase imbalance.
[0038] To avoid output jumps caused by parameter abrupt changes at the operating condition boundary, a transition band of ±5% is set near the boundary threshold. Within the transition band, linear interpolation is used to calculate parameter values to achieve a continuous and smooth transition of parameters.
[0039] Step S3: A dual-layer composite control architecture is adopted, which provides virtual inertia support through outer-loop virtual synchronous control and voltage support through inner-loop fast reactive current injection. At the same time, the active and reactive components of the output jointly support the transient voltage at the grid connection point.
[0040] This step is the core execution stage of the entire method. It utilizes a two-layer composite control architecture to simultaneously output active and reactive power support, achieving proactive voltage rise and stabilization. The two-layer architecture balances response speed and dynamic stability.
[0041] The control architecture consists of two layers: an outer ring and an inner ring, which handle support requirements at different time scales respectively. Specifically, the outer loop is a virtual synchronous control layer, which adopts a virtual synchronous generator control strategy to simulate the rotor motion equations of a synchronous generator: Where J is the virtual inertia, D is the damping coefficient, Pm is the mechanical power reference, Pe is the electromagnetic power, and ω is the electric angular velocity. By simulating the rotor's inertia and damping characteristics, the energy storage converter can provide active inertia support like a synchronous generator, suppressing rapid changes in frequency and voltage. The outer loop has a response time scale on the order of hundreds of milliseconds and its main function is to provide system inertia, improve dynamic damping, and maintain the stability of transient processes. Its output is the active current reference value.
[0042] The inner loop is a fast reactive power control layer, employing direct reactive power control based on voltage deviation. It collects the deviation between the grid-connected point voltage amplitude and its rated value in real time, calculates the reactive current reference value using a proportional-integral regulator, and directly feeds it into the current inner loop for execution. The inner loop has a high control bandwidth and a response speed down to the millisecond level, enabling it to rapidly inject reactive current after a transient event, thus raising the grid-connected point voltage amplitude. The inner loop's role is to quickly provide voltage amplitude support, which is crucial in determining the effectiveness of voltage sag suppression.
[0043] The outer loop outputs an active current reference, and the inner loop outputs a reactive current reference. Both are fed into the inner current loop controller, which simultaneously outputs the active and reactive components from the converter. Active current support maintains dynamic stability and inertia characteristics, while reactive current support increases voltage amplitude. The two work together to achieve comprehensive support for transient voltage.
[0044] Compared to traditional single virtual synchronous machine control, the inner loop fast reactive power channel added in this embodiment improves the voltage response speed; compared to single reactive power droop control, the outer loop inertia control retained in this embodiment ensures the damping characteristics and stability of the system.
[0045] Furthermore, regarding the adaptive adjustment of inertia during transient processes, the virtual inertia does not remain constant at the reference value during the transient process, but rather adjusts in real time according to the voltage drop depth. The adjustment formula is: Where J0 is the reference inertia corresponding to the current operating condition, ΔU is the voltage drop depth, i.e. the difference between the rated voltage and the actual voltage, and k is the inertia adjustment coefficient, which is 2.0 in this embodiment.
[0046] The beneficial effect of this adaptive adjustment is that the deeper the voltage drop, the more severe the disturbance, and the greater the inertia required to suppress the rate of change; as the voltage gradually recovers, the inertia decreases synchronously, avoiding overshoot and oscillation caused by excessive inertia in the later stages of recovery.
[0047] Step S4: Calculate the comprehensive weight based on the three indicators of remaining capacity, response rate and battery capacity retention rate of each energy storage unit, and allocate the total supporting power demand to each energy storage unit according to the weight ratio.
[0048] This step addresses the power distribution issue when multiple energy storage converters operate in parallel. Through comprehensive weight allocation, each energy storage unit undertakes its corresponding share of support tasks according to its own capabilities, thereby maximizing overall support efficiency.
[0049] The allocation model includes three weighting metrics: The first item is the weight of the remaining capacity, which is the current SOC value. The higher the SOC, the more energy the energy storage unit can release, and the more support output it can support. In this embodiment, the weight coefficient for this item is set to 0.5, which is the main basis for allocation, because the support continuity capability depends primarily on the remaining power.
[0050] The second term is the weight of the response rate, which is the reciprocal of the rise time of the step response to the rated power of the energy storage unit. Energy storage units with faster response times can reach their supporting power more quickly and contribute more to transient peak values; therefore, they should be allocated more weight to rapidly changing components. In this embodiment, this term is weighted at 0.3.
[0051] The third item is the weight of the battery capacity retention rate. The battery capacity retention rate is equal to the current actual capacity divided by the initial capacity and then multiplied by 100%, and is used to measure the degree of battery life degradation. Energy storage units with a high battery capacity retention rate have a lower degree of aging, greater overload capacity and cycle life margin, and can undertake more support tasks. In this embodiment, the weight coefficient for this item is set to 0.2.
[0052] Furthermore, the three indicators are weighted and summed using their respective weights to obtain the comprehensive weight of each energy storage unit. In this embodiment, the comprehensive weight of the i-th energy storage unit is... The calculation formula is: ;in, These represent the normalized remaining capacity, response rate, and battery capacity retention rate of the i-th energy storage unit, respectively.
[0053] The supporting power allocated to each energy storage unit is equal to the total supporting power multiplied by the comprehensive weight of that energy storage unit, and then divided by the sum of the comprehensive weights of all energy storage units. Preferably, the supporting power allocated to the i-th energy storage unit is... for: That is, total supporting power. Multiply by the proportion of the overall weight of this energy storage unit to the overall weight of all energy storage units.
[0054] In this embodiment, the allocation calculation is updated every 100 milliseconds, dynamically adjusting to changes in the SOC of each energy storage unit and the total support requirements.
[0055] Step S5: Calculate the support effect score during the transient process, and use the support effect score as feedback to iteratively correct the parameter matching library to achieve continuous self-optimization of support performance.
[0056] This step constitutes the feedback loop of the control closed loop. By quantitatively evaluating the support effect of each transient process, the control parameters are optimized accordingly to achieve continuous improvement in system performance.
[0057] Furthermore, this embodiment will perform a quantitative evaluation of the support effect. After each transient process, three support effect evaluation indicators will be automatically calculated: The first item is voltage recovery time. The recovery time is calculated from the moment the disturbance occurs until the grid connection point voltage recovers and stabilizes above 0.9 per unit. A shorter recovery time results in better support. In this embodiment, this item is weighted at 0.4. The second item is the transient suppression ratio. By comparing the actual maximum sag depth with the expected sag depth under the same fault conditions without energy storage support, the improvement ratio of the sag depth due to support is calculated. The formula is: Where D0 is the expected sag depth without support, and D1 is the actual sag depth. A higher suppression ratio indicates a more significant effect of support on voltage amplitude increase. In this embodiment, the weight of this item is set to 0.4. The third item is voltage overshoot. The ratio of the highest voltage value during voltage recovery to the rated voltage. A smaller overshoot results in a smoother recovery process. This item has a weight of 0.2 and is considered a negative indicator.
[0058] Furthermore, the three indicators are normalized separately and then weighted and summed to obtain a percentage-based support effectiveness score. The specific formula is as follows: ,in This is the normalization function.
[0059] Furthermore, for parameter self-learning correction, this embodiment uses the Q-learning reinforcement learning algorithm to iteratively optimize the parameter matching library.
[0060] Specifically, the state space consists of four working conditions, i.e., four states. The action space consists of adjustments to three parameters per group, with each parameter having three options: increase, remain unchanged, or decrease. In this embodiment, the adjustment step size is 5% of the baseline value. The reward function directly takes the support effect score; the higher the support effect score, the greater the reward. The algorithm aims to learn the optimal parameter combination for each working condition to maximize the expected reward. An ε-greedy strategy is used for action selection: new parameters are randomly explored with probability ε, and the current optimal parameter is selected with probability 1-ε. The exploration probability ε is initially set to 0.3 to ensure sufficient exploration in the early stages; as the number of learning iterations increases, ε gradually decays exponentially to 0.05, with later iterations focusing on utilizing the already learned optimal strategy. Each complete transient process is used as a learning sample, updating the Q-value of the corresponding state-action pair. When the cumulative learning samples reach 100, the parameter group corresponding to the action with the largest Q-value in each working condition is updated to the parameter matching library, completing one round of parameter iteration optimization. It should be noted that the specific optimization iteration process of the Q-learning reinforcement learning algorithm is existing technology, and this embodiment does not impose special restrictions, so it will not be elaborated here.
[0061] Through continuous online learning, the parameter matching library will be continuously optimized as operational experience accumulates, supporting performance to gradually improve and adapt to the slow changes in power grid characteristics.
[0062] Furthermore, to ensure a safe and smooth transition between different operating states and to maximize support capabilities within the equipment's safety boundaries, this embodiment constructs a multi-mode smooth switching mechanism, setting three operating modes: Normal grid-connected mode: When the voltage is normal, the grid-type control operates normally without activating enhanced support; Transient support mode: When the voltage drops slightly, enhanced support parameters are activated to maximize support capabilities; Fault ride-through mode: When the voltage drops extremely low, it switches to current-limiting ride-through mode to ensure that the equipment does not disconnect from the grid.
[0063] Therefore, mode switching employs hysteresis comparison logic. When the voltage is higher than a first preset voltage and remains above it for a preset duration, the system switches to normal mode; when the voltage is lower than a second preset voltage, it switches to transient support mode; and when the voltage is lower than a third preset voltage, it switches to fault ride-through mode. Preferably, in this embodiment, when the voltage is higher than 0.95 pu and remains above it for 200 ms, the system switches to normal mode; when the voltage is lower than 0.9 pu, it switches to transient support mode; and when the voltage is lower than 0.5 pu, it switches to fault ride-through mode. The hysteresis design avoids frequent switching when the voltage fluctuates near the threshold. The parameter switching process uses a 50-millisecond ramp transition, meaning the parameter changes linearly and gradually from the old value to the new value, rather than jumping instantaneously. The ramp transition ensures smooth changes in output power and avoids secondary impacts on the power grid.
[0064] Furthermore, this embodiment will implement multiple protection coordination. Specifically, three levels of protection constraints are set, arranged from high to low priority. The first priority is overcurrent protection, which immediately limits the current when the output current exceeds a preset multiple of the rated current. The second priority is overvoltage protection, which gradually reduces reactive power injection until it exits support when the grid connection point voltage exceeds a preset value. The third priority is SOC boundary protection, which exits active power support and retains only reactive power support when the SOC of the energy storage unit is lower than a preset first percentage, and exits charging mode when it is higher than a preset second percentage.
[0065] Preferably, in this embodiment, the first priority overcurrent protection is as follows: when the output current exceeds 1.5 times the rated current, the current is limited to 1.5 times by saturation limiting. Device safety is the top priority, and the current withstand limit must not be exceeded under any circumstances. Second priority overvoltage protection: In this embodiment, when the grid connection point voltage exceeds 1.2 pu, the reactive power injection is gradually reduced to prevent excessive support from causing the voltage to be too high. A linear proportional method is used to gradually reduce the load instead of immediately cutting off the load. The linear proportional method is as follows: the load reduction starting threshold is set to 1.2 pu. When the voltage is higher than this value, the reactive power injection is reduced by 10% for every 0.01 pu increase in voltage. The advantage of linear proportional load reduction is that it is continuous and smooth. The higher the voltage, the more it is reduced. This limits the maximum voltage and preserves the support capacity as much as possible, without any sudden drop to the threshold and disconnection. Third priority SOC boundary protection: When the SOC of the energy storage unit is below 10%, the active power support channel is exited, and only reactive power support is retained (reactive power does not consume active power); when the SOC is above 90%, the charging mode is exited to avoid overcharging.
[0066] All protections operate using a smooth limiting method, which limits the magnitude while preserving as much remaining support capacity as possible, rather than simply disengaging completely, thus maximizing support capacity under safety constraints.
[0067] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A method for active support and control of transient voltage in grid-type energy storage based on multi-condition sensing, characterized in that, Includes the following steps: S1: Real-time acquisition of grid connection point voltage and current data, extraction of multi-dimensional features of transient voltage disturbances, and identification of the current disturbance type based on hierarchical judgment rules; S2: Based on the identified operating condition type, dynamically retrieve three sets of core control parameters—virtual inertia, damping coefficient, and reactive power support gain—from the parameter matching library as the initial parameters for support control. S3: It adopts a dual-layer composite control architecture that provides virtual inertia support through outer loop virtual synchronous control and voltage support through inner loop fast reactive current injection, and simultaneously outputs active and reactive components to support the transient voltage at the grid connection point. S4: Calculate the comprehensive weight based on the three indicators of remaining capacity, response rate and battery capacity retention rate of each energy storage unit, and allocate the total supporting power demand to each energy storage unit according to the weight ratio; S5: Calculate the support effect score during the transient process, and use the support effect score as feedback to iteratively correct the parameter matching library to achieve continuous self-optimization of support performance.
2. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, In S1, the positive-sequence component and negative-sequence component of the voltage fundamental wave are extracted, and multi-dimensional characteristic quantities are calculated, including the voltage sag depth and the ratio of the effective value of the positive-sequence component of the voltage to the rated voltage. Disturbance duration, the length of time the voltage deviates from the rated value by ±10%; voltage imbalance, the percentage of negative-sequence voltage component to positive-sequence voltage component.
3. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 2, characterized in that, The classification rule for identifying the current disturbance's operating condition type includes: minor voltage sag: sag depth. The criteria for determining voltage sags are as follows:
1. PU and unbalance < 4%; 2. Severe voltage sag: sag depth < 0.7 PU and unbalance < 4%; 3. Voltage swell: voltage amplitude > 1.1 PU; 4. Unbalance fault: unbalance ≥ 4%. The priority of determination is as follows: first, determine the unbalance fault based on the unbalance degree; then, determine whether it is a voltage sag or swell based on the voltage amplitude; and finally, determine the degree of sag based on the sag depth.
4. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, The dual-layer composite control architecture specifically includes: The control architecture consists of two layers: an outer loop and an inner loop. The outer loop is a virtual synchronous control layer that simulates the rotor motion equation of a synchronous generator and outputs an active power reference value through the active power-frequency droop characteristic. The inner loop is a fast reactive power control layer that outputs a reactive current reference value based on the grid connection point voltage deviation via a proportional-integral regulator and directly modulates the reactive power output of the converter.
5. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, The transient process virtual inertia adaptive adjustment specifically includes: the virtual inertia is dynamically adjusted in real time according to the voltage drop depth during the transient process; the adjustment formula is as follows: Where J0 is the reference inertia corresponding to the current operating condition, ΔU is the voltage drop depth, which is the difference between the rated voltage and the actual voltage, and k is the inertia adjustment coefficient.
6. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, In step S4, three indicators are obtained for each energy storage unit: remaining capacity, response rate, and battery capacity retention rate. The three indicators are weighted and summed using their respective weights to obtain the comprehensive weight of each energy storage unit. The supporting power allocated to each energy storage unit is equal to the total supporting power multiplied by the comprehensive weight of that energy storage unit and then divided by the sum of the comprehensive weights of all energy storage units.
7. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, The support effect score specifically includes: Three support effectiveness evaluation indicators were calculated: voltage recovery time, which is the time it takes for the voltage at the grid connection point to recover to 0.9 per unit value from the moment the disturbance occurs; sag suppression ratio, which is the difference between the maximum sag depth and the expected sag depth without support, divided by the sag depth without support; and voltage overshoot rate, which is the proportion of the highest voltage exceeding the rated value during the recovery process. The support effectiveness score was obtained by normalizing the three indicators and then weighting and summing them.
8. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, The iterative correction of the parameter matching library based on the support effect score specifically includes: The Q-learning reinforcement learning algorithm is used to achieve adaptive optimization of parameters. The state space consists of four types of working conditions, and the action space consists of the up and down adjustment of each set of parameters. The adjustment step size is a preset percentage of the baseline value. The reward function is the support effect score. After the cumulative learning samples reach a preset number of times, the optimized parameters are updated to the parameter matching library.
9. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, It also includes a multi-mode smooth switching strategy, including: The system is set with three operating modes: normal grid connection mode, transient support mode, and fault ride-through mode. Mode switching adopts hysteresis comparison method. When the voltage is higher than the first preset voltage and lasts for a preset duration, it switches to normal mode. When the voltage is lower than the second preset voltage, it switches to transient support mode. When the voltage is lower than the third preset voltage, it switches to fault ride-through mode. During the switching process, the control parameters use a ramp function to smoothly transition.
10. The multi-condition state sensing method for active support and control of transient voltage in grid-type energy storage as described in claim 1, characterized in that, It also includes coordination of multiple constraint protections, including: The triple protection constraints are set and executed in order of priority: the first priority is overcurrent protection, which immediately limits the current when the output current exceeds a preset multiple of the rated current; the second priority is overvoltage protection, which gradually reduces the reactive power injection in a linear proportional manner until it exits support when the grid connection point voltage exceeds a preset value; the third priority is SOC boundary protection, which exits active power support and only retains reactive power support when the SOC of the energy storage unit is lower than a preset first percentage, and exits charging mode when it is higher than a preset second percentage.