Grid-connected and off-grid seamless switching control method and device for reconfigurable battery energy storage system

Abstract

The invention relates to a reconfigurable battery energy storage system grid-connected and off-grid seamless switching control method and device. The method comprises the following steps of: sensing a real-time power grid state, pre-judging a system operation mode, and acquiring a millisecond-level panoramic information flow of the power grid state through a high-speed sampling voltage sensor, a high-speed sampling frequency sensor, a high-speed sampling phase sensor and a high-speed sampling communication interface. And pre-judging the stability and disturbance trend of the power grid through a deep learning algorithm, executing a power distribution and voltage support algorithm in grid-connected and off-grid modes, and calculating an optimal switching opportunity point and a system reconstruction target topology scheme. Meanwhile, the harmonic distortion rate and the voltage sag depth are calculated in real time through discrete Fourier transform, so that a mode pre-judgment mechanism is automatically triggered. After switching, key parameters such as voltage, frequency and power are automatically corrected through real-time sampling and observation, and zero-sensing switching is achieved. The method can effectively improve the response speed and stability of the battery energy storage system when the power grid fails, and guarantees the safe operation of the power grid.

Description

Technical Field

[0001] The present invention belongs to the field of energy storage batteries, and more specifically relates to a reconfigurable battery energy storage system and an off-grid seamless switching control method and device. Background Art

[0002] Developing renewable energy is a key initiative to address energy and environmental challenges. Battery energy storage technology, in particular, is gaining widespread adoption due to its unique advantages, including high energy storage efficiency, long storage duration, and pollution-free power generation. Battery energy storage systems store electrical energy in batteries and release it to the power grid at appropriate times. They are widely used in distribution networks, microgrids, electric vehicles, and other fields. However, for grid-connected battery energy storage systems, how to quickly and stably switch to off-grid operation in the event of a sudden grid failure to avoid adverse impacts on the grid and loads has become a significant technical challenge.

[0003] Traditional battery energy storage system switching control methods typically disconnect the battery energy storage system from the grid upon detecting a grid anomaly, and then initiate off-grid operation. A disadvantage of this method is that the switching process results in a long grid outage, which can cause critical loads to lose power. Furthermore, traditional battery energy storage system switching control methods often employ fixed switching timing and sequences, failing to dynamically adjust switching strategies based on real-time grid and system status changes. This often leads to unnecessary voltage and frequency fluctuations, transient equipment overloads, and other issues during the switching process.

[0004] To address these issues and improve the grid's resilience to faults, scholars have proposed numerous design solutions. However, existing solutions often lack comprehensive consideration of multiple factors, such as the battery energy storage system's operating status and the grid's state. These solutions are unable to accurately predict and effectively control the seamless on-grid and off-grid switching of battery energy storage systems. Consequently, they suffer from slow switching speeds, poor stability, and significant impacts on system operation. Therefore, developing a novel control method and device for seamless on-grid and off-grid switching of reconfigurable battery energy storage systems is of great scientific and application value. Summary of the Invention

[0005] This invention primarily addresses how to enable battery energy storage systems to respond quickly and stably to sudden grid failures, enabling seamless switching of reconfigurable energy storage systems and real-time correction and control of key parameters such as voltage and frequency. Solving this technical problem can significantly improve the grid's resilience and stability to sudden failures, effectively reducing the impact of a fault on the grid and loads, and ensuring the safe operation of the battery energy storage system and the grid.

[0006] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions: the method comprises: Build a real-time grid status perception and system operation mode prediction unit. Through high-speed sampling voltage, frequency, and phase sensors and communication interfaces, it obtains millisecond-level panoramic information flow of grid status and calculates the optimal switching timing and system reconstruction target topology solution. Deploy high-precision voltage transformers, phase-locked loops, and fiber-optic communication interfaces to continuously collect three-phase voltage, current, frequency, and phase angle. These quantities are then converted into parameters in a stationary coordinate system through transformation. Using discrete Fourier transforms, the harmonic distortion rate and voltage sag depth are calculated in real time to predict switching timing and generate the target topology solution. Deep learning is used to predict disturbance trends and pre-calculate the control parameters of the two operating modes in parallel before switching occurs. If the risk of voltage collapse persists, the system is forced to switch back to the initial stable mode within a limited time. Self-checking and safety control work together. After switching, they automatically correct key parameters such as voltage, frequency, and power through real-time sampling and observation, shortening power recovery and frequency stabilization time. Zero-perception switching significantly shortens the system power recovery time and frequency stabilization time after switching.

[0007] In one solution, the data sampling rate of the real-time grid status perception unit is 100,000 times per second or more. Specifically, it continuously collects key grid parameters such as three-phase voltage, current, frequency and phase angle through the collaborative work of high-precision voltage transformers, phase-locked loops and optical fiber communication interfaces, and converts the three-phase quantities into parameters in a stationary coordinate system through Clarke transform. At the same time, discrete Fourier transform is used to calculate the harmonic distortion rate and voltage sag depth in real time.

[0008] In one solution, the disturbance prediction model uses a deep learning algorithm based on a long short-term memory network to process the collected time series data in a sliding window format, and outputs the grid stability probability and disturbance type vector within the next fifty milliseconds. The disturbance types include voltage swells, sags, frequency oscillations, harmonic excesses, and communication interruptions. When the predicted grid stability probability is lower than a set threshold or the disturbance risk is higher than a set threshold, the mode prediction mechanism is automatically triggered.

[0009] In one scheme, the main controller can run the power allocation algorithm in the grid-connected mode and the virtual synchronous machine control algorithm in the off-grid mode in parallel during the mode prediction stage, and calculate the switching timing and target topology scheme in advance by comparing the system state evaluation functions in the two modes.

[0010] In one solution, the deep learning model LSTM is used to predict disturbance trends before switching, and the control parameters for both grid-connected and off-grid operating modes are pre-calculated in parallel. Combined with cross-cycle rolling optimization, a switching plan is generated that prioritizes disconnecting non-critical load branches in a precise manner. If the risk of grid voltage collapse persists, the system is forced to switch back to the initial stable mode within ten milliseconds. All protection actions are based on fail-safe principles.

[0011] In one solution, through the coordinated operation of self-checking and safety control, high-frequency real-time sampling, observers and model predictive control algorithms are used to automatically correct key parameters such as voltage, frequency, and power, which can shorten the power recovery time after switching to less than eight milliseconds and the frequency stabilization time to less than fifteen milliseconds.

[0012] In one solution, the safety control layer has a multi-layer protection mechanism that can automatically perform fault isolation and redundancy compensation under extreme conditions such as communication interruption lasting two hundred milliseconds or complete failure of a single module.

[0013] In one solution, the method adopts a self-checking and safety-oriented control collaborative mechanism throughout the switching process. The system can automatically correct the control instructions based on the real-time detected operating parameters, achieve zero perception during the switching process, and quickly recover to the safe operating area after the switch.

[0014] In one solution, the method runs both grid-connected and off-grid control modes in parallel within the main controller, pre-calculates all key parameters and target topologies before switching occurs, and combines a distributed collaborative switching mechanism to achieve nanosecond-level synchronous issuance of switching instructions.

[0015] In another aspect, a reconfigurable battery energy storage system and an off-grid seamless switching control device are provided, wherein the device is applicable to the method described above and comprises: The real-time grid status perception and system operation mode prediction unit obtains millisecond-level panoramic information flow of the grid status through high-speed sampling voltage, frequency, and phase sensors and communication interfaces; A data processing module equipped with a high-precision voltage transformer, phase-locked loop, and fiber-optic communication interface; a disturbance prediction model based on a long short-term memory network; and power distribution and voltage support algorithms running in parallel in both on-grid and off-grid modes within the main controller; Deep learning predicts disturbance trends and combines cross-cycle rolling optimization to generate a switching plan that prioritizes disconnecting non-critical load branches in a precise manner. After switching, real-time sampling and observation are used to automatically correct key parameters such as voltage, frequency, and power, achieving zero-perception switching.

[0016] Beneficial effects of the present invention: 1. By building a real-time grid status perception and system operation mode prediction unit, it is possible to accurately predict the grid status and operation mode, and achieve seamless switching between the battery energy storage system and the grid.

[0017] 2. Through deep learning, disturbance trends are predicted and control parameters for the two operating modes are pre-calculated in parallel, which improves the speed and accuracy of switching and effectively reduces the impact on the power grid during system switching.

[0018] 3. Through real-time sampling and observation, self-checking and automatic correction of key parameters such as voltage, frequency, and power, the time for power recovery and frequency stabilization is shortened, and the operating stability of the system is improved.

[0019] 4. Through safety control, the fault impact energy is limited to the safety range of the semiconductor device tolerance limit, thereby improving the safety threshold of the energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention are described clearly and completely below. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, a control method for seamless on-grid and off-grid switching of a reconfigurable battery energy storage system includes the following steps: Step 1: Build a real-time grid status perception and system operation mode prediction unit. This unit continuously uses high-speed sampling voltage, frequency, and phase sensors and communication interfaces to build a millisecond-level panoramic information flow of grid status. Simultaneously, it combines deep learning or fuzzy logic algorithms to perform ultra-short-term predictions of grid stability and assess disturbance trends. Power distribution and voltage support algorithms for both on-grid and off-grid modes run in parallel within the main controller, pre-calculating the optimal switching timing and target topology for system reconstruction. This forward-looking perception and "mode prediction" mechanism replaces the traditional passive response mode and lays the foundation for subsequent seamless switching.

[0023] The system deploys high-precision voltage transformers (sampling rate ≥ 100kHz), phase-locked loops (PLLs) and optical fiber communication interfaces to continuously collect three-phase voltage , current , frequency f and phase angle ). Through Clarke transformation ( ) Convert the three-phase quantity into the stationary coordinate system :

[0024] Synchronously calculate the harmonic distortion rate THD and voltage sag depth in real time using discrete Fourier transform (DFT) The above millisecond-level raw data stream is input into the disturbance prediction model based on the long short-term memory network (LSTM). The model processes the time series data in the form of a sliding window and outputs the probability of grid stability within the next 50ms. and perturbation type vector (corresponding to voltage swell, sag, frequency oscillation, excessive harmonics and communication interruption respectively). or When , the mode prediction mechanism is triggered. At this time, the main controller performs two core calculations in parallel: 1. Solve the optimal power allocation in grid-connected mode:

[0025]

[0026] 2. Constructing the Virtual Synchronous Generator (VSG) equation in off-grid mode:

[0027]

[0028] By comparing the system status evaluation function under the two modes , the predicted switching timing is , and generate the target topology solution at the same time , For battery cluster connection status, This process refreshes the prediction result every 10ms to achieve forward-looking decision-making.

[0029] Different from traditional passive detection based on threshold comparison, this step predicts disturbance trends through deep learning and completes parallel pre-calculation of the control parameters of the two operating modes before switching occurs. Combined with cross-cycle rolling optimization, it generates an active switching strategy that includes precise time points and topology configurations, eliminating the response delay during the mode transition period from the root.

[0030] Step 2: Establish a unified interface architecture for flexible on-grid and off-grid battery clusters based on multi-port power electronic converters. Leveraging the isolated multi-port bidirectional DC / DC converters (such as dual active bridges (DABs)) between the battery cluster and the DC bus in the reconfigurable battery energy storage system, and the power conversion system (PCS), a unified interface architecture is designed at the hardware level. This architecture presets two operating logics at the control software level: grid-connected power dispatch mode and off-grid voltage source mode. It allows each battery cluster and its corresponding converter module to instantly switch between the two control logics under system-level instructions without physically switching contactors. This flexible interface based on power electronic converters is the core physical carrier for achieving seamless switching.

[0031] In terms of hardware architecture, each battery cluster ( ) is connected to the DC bus through an independent isolated dual active bridge (DAB) converter, and its power transmission is determined by the phase shift angle control:

[0032] in is the transformer ratio, is the battery voltage, is the DC bus voltage, is the switching angular frequency, The DAB output is connected in parallel to a multi-port power distribution unit (MPDU), which is directly connected to the three-phase PCS bridge arm via a pre-charge resistor. At the control software layer, each module has two core algorithms: 1. Grid-connected mode (PQ control): The outer loop receives system-level power commands , the inner loop adopts current decoupling control:

[0033] in , , is the grid voltage feed-forward compensation term.

[0034] 2. Off-grid mode (VF control): Generate voltage reference based on the principle of virtual synchronous machine:

[0035] Then the voltage tracking is achieved through the quasi-PR controller:

[0036] in is the cutoff frequency, is the fundamental frequency.

[0037] When the system instruction triggers the switch (such as switching from grid-connected to off-grid), the main controller sends the mode switch command to the FPGA of each module through the optical fiber synchronization signal. Freeze the current moment The instantaneous value of the three-phase voltage is obtained by inverse coordinate transformation ;exist time( is the control cycle), directly switch to VF mode for the initial value and load the pre-stored virtual inertia parameters At the same time, the working state of DAB changes from "current source type" (maintain Constant) into "voltage source type", by dynamically adjusting the phase shift angle Make the output impedance satisfy:

[0038] This process does not require mechanical contactor action, and the functional conversion can be completed within 200μs by simply reconstructing the modulation logic of the power electronic switch. In addition, the DAB of each battery cluster in off-grid mode adopts an active voltage support strategy: when it detects When the fluctuation exceeds ±2%, droop control is automatically enabled: Ensure the stability of multiple modules in parallel.

[0039]

[0040]

[0041] Step 3: Implement dynamic power path decoupling and zero-switching impact energy buffering strategies. When an impending mode switch is predicted (triggered by step 1), system control instructions first direct some battery clusters and their converter modules in an "idle" state to activate their target modes tens of milliseconds in advance (for example, when switching from grid-connected to off-grid, some modules are pre-activated to enter voltage source mode). By precisely controlling the power output of these "pre-activated" modules, their electrical characteristics (voltage and phase) accurately track the current load or microgrid demand. The system's filter capacitors, inductors, or supercapacitor modules are then used to form a very short-term "virtual bus" to actively absorb or release instantaneous power differences. This achieves "connection" and "decoupling" of the old and new power paths at the moment of switching, essentially transforming the switching process into a controlled power transfer process rather than a traditional opening / closing process, achieving zero-current impact switching.

[0042] At the switching moment predicted in step 1 The first 30ms (i.e. ), the main controller according to the target topology solution Select a subset of battery clusters that are idle . Control instruction guide The corresponding converter module is activated to the target mode in advance (for example, switching to VF mode when switching from grid-connected to off-grid), and the phase calibration signal is injected synchronously:

[0043] in is the current grid phase, It is the preset virtual angular velocity compensation. , pre-activated module output power is set to:

[0044] This exponential ramp command allows the power output to smoothly increase from 0 to the target value. At the electrical characteristics level, the pre-activation module accurately tracks the load voltage demand through dual closed-loop control: Voltage outer loop: Generating a current reference

[0045] Current inner loop: Space Vector Modulation (SVM) Drives IGBTs Synchronous use of filter capacitors in the system Constructing dynamic virtual bus: When power difference is detected When the capacitor energy buffer control is triggered:

[0046] in is the hysteresis adjustment coefficient The bigger the , this strategy enables the capacitor current to actively compensate for the instantaneous power gap. At the moment the main contactor operates, the new and old power paths meet the following decoupling conditions:

[0047] At this time the load current The transfer process is described by the following dynamic equation:

[0048] Since the voltage difference and resistance difference are controlled within the limit range ( ), the forced component of this equation approaches zero, achieving current continuity.

[0049]

[0050] Step 4 triggers millisecond-level distributed coordinated switching logic and topology reconstruction. At the moment the dynamic power buffer reaches equilibrium, the master controller issues nanosecond-level coordinated switching instructions via a high-speed communication bus (such as EtherCAT). The converter modules corresponding to each battery cluster synchronously switch their control mode (e.g., PQ to V / f, or vice versa) and power flow direction according to the pre-installed target topology solution (provided in Step 1), and may also change internal switch states to reconfigure the DC link or AC connection mode. This ensures that all modules complete functional conversion and topology adjustment within the same electrical cycle, ensuring that the entire system presents a continuous and uninterrupted appearance.

[0051] When the dynamic power buffer in step 3 reaches the equilibrium point (satisfying and phase difference When a timer is connected, the master controller broadcasts a timestamp-locked collaborative instruction via the EtherCAT high-speed bus. The instruction consists of two core components: 1. Target topology identifier (generated by step 1), using binary coding to clearly specify the target connection mode of each module (such as DC side series / parallel flag, AC side star / delta configuration).

[0052] 2. Atomic action sequence template, which defines the time base point (Synchronization accuracy ±100ns) Standardized set of operations to be performed, including: control mode switching instructions (PQ VF), power direction flag (charge / discharge status), and the binary on / off code of the internal switch matrix.

[0053] Distributed collaborative execution The FPGA of each converter module is at the time base point Synchronously trigger the following actions: Seamless switching of control modes: Instant switching of control algorithms based on preloaded parameter mapping tables (e.g. when switching from PQ mode to VF mode, the initial value of the voltage injected into the latch) ), to ensure that the controller output has no step jump.

[0054] Power flow direction reversal: bidirectional energy flow switching is achieved by instantaneously reversing the sign of the phase shift angle:

[0055] in is the phase shift angle of the dual active bridge (DAB), the sign function Updated according to target power direction.

[0056] Dynamic topology reconfiguration: driving a 4-switch crossbar matrix based on SiC MOSFETs (e.g., when the DC side is connected in series, executing the switching logic: ,exist The electrical connection path is reconstructed internally and the switching spike voltage is absorbed by the mirror capacitor.

[0057] Synchronization guarantee mechanism To maintain switching of all modules within a single electrical cycle (20ms@50Hz): 1. Phase continuity control: Each module inherits the current angular frequency of the system at the moment of switching , and add dynamic compensation:

[0058] in is the adaptive gain coefficient (typical value 0.2 rad / kW·s) to suppress frequency transients.

[0059] 2. Fault-tolerant communication protocol: The command frame adopts CRC32 checksum and retransmission mechanism (timeout window ), the abnormal module automatically switches to the local cached safety topology plan.

[0060] 3. Real-time safety monitoring: The module detects the DC bus voltage in real time through a high-bandwidth ADC (sampling rate 2MHz) and output current slope ,like or , immediately triggering the local IGBT soft shutdown and enabling the freewheeling path.

[0061] After the coordinated switching is completed, the transient characteristics of the system are as follows: Voltage stability: AC bus voltage sag depth , recovery time , meeting the IEEE 1547 standard.

[0062] Power continuity: grid-connected load power pulsation , off-grid frequency deviation .

[0063] Signal quality: The total harmonic distortion (THD) of the output voltage maintains the level before switching.

[0064] This collaborative architecture combines the nanosecond response of power electronics (SiC MOSFET switching delay ) and Time Sensitive Network (TSN) communication technology, compressing the cascade delay of traditional centralized switching (20-100ms) to .

[0065]

[0066] Step 5: Execute post-seamless switching operational status self-verification and safety-oriented control. After the mode switch is complete, the system immediately enters a closed-loop self-verification state, utilizing high-precision sensors to monitor key node parameters such as voltage, current, and frequency in real time. Model prediction or state observer algorithms are used to evaluate the system's stability and performance in the new mode. If a slight deviation is detected, a rapid fine-tuning algorithm is activated, dynamically adjusting the output power of some battery clusters or activating backup modules to restore the system to the ideal operating point within tens of milliseconds. Simultaneously, a safety-oriented control layer is embedded. Upon detecting anomalies at any stage (such as communication delays, component failures, or sudden load changes), the system can instantly enforce optimal protection actions (such as switching back to stable mode, activating redundant modules, and performing staged load shedding) based on pre-set "fail-safe" strategies, ensuring system stability and equipment safety while minimizing potential risks.

[0067] At the moment the mode switch is completed (accurate to within ±100μs), the system immediately starts the closed-loop self-calibration process. The high-precision sensor network captures key node parameters (such as DC bus voltage) in real time at a 1MHz sampling rate. , AC side current , grid frequency ), and multi-dimensional state assessment based on the pre-installed digital twin model: reconstructing the output current dynamic equation through the Luenberger Observer

[0068] , and combined with the model predictive control (MPC) algorithm to calculate the stability index (including voltage fluctuation , power oscillation amplitude and frequency offset integral When a deviation exceeding a safety threshold is detected (such as voltage deviation or frequency deviation>0.1Hz), the system immediately triggers the fast fine-tuning mechanism - by dynamically adjusting the coefficient Redistribute battery cluster power instructions: If ), the backup module is called to compensate for the difference in power; if the voltage regulation requirement , then the droop control parameter adaptive algorithm is enabled , the operating point converges to the target operating range (deviation < 0.5%) within 20ms.

[0069] The safety-oriented control layer, which runs in parallel, forms the core line of defense for system resilience. This layer implements three layers of protection based on a real-time fault tree (FTA) model: Communication abnormal response: When EtherCAT message delay > 50μs or CRC error rate > 10 4When the local controller automatically switches to the pre-stored compensation timing table and reconstructs the clock reference through the Time Sensitive Network (TSN) resynchronization protocol; Component level fault tolerance: If the IGBT drive circuit reports an overcurrent ( ) or the junction temperature exceeds the limit ( ), immediately enable the N+1 redundant module to take over the load and trigger the SiC MOSFET soft shutdown sequence ( is the switching cycle); System-level protection: against sudden load changes In the event of a large disturbance, a hierarchical unloading strategy is implemented: non-critical load branches are disconnected first (response time < 5ms), and if the voltage collapse risk persists ( ), then it will be forced to switch back to the initial stable mode within 10ms. All protection actions are based on the principle of Fail-Safe, ensuring that the system stabilizes to the safe operating area (SOA) within 150ms, limiting the fault impact energy to (Less than 20% of the tolerance limit of semiconductor devices).

[0070] The coordinated operation of self-checking and safety control enables the system to achieve "zero-perception switching"—measured data shows that power recovery time after switching is reduced to 8ms, frequency stabilization time is less than 15ms, and the core bus voltage sag is suppressed to within 1.8%. The safety control layer significantly reduces the risk of fault propagation, ensuring that the system maintains over 95% load power supply capacity even in the event of a 200ms communication interruption or complete failure of a single module. The fault isolation success rate exceeds 99.99%, meeting SIL-3 safety certification requirements.

[0071]

[0072] Based on a method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system, a device for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system is constructed. The device includes: a real-time grid state perception and system operation mode prediction unit, which obtains millisecond-level panoramic information flow of grid state through high-speed sampling voltage, frequency, and phase sensors and a communication interface; A data processing module equipped with a high-precision voltage transformer, phase-locked loop, and fiber-optic communication interface; a disturbance prediction model based on a long short-term memory network; and power distribution and voltage support algorithms running in parallel in both on-grid and off-grid modes within the main controller; Deep learning predicts disturbance trends and combines cross-cycle rolling optimization to generate a switching plan that prioritizes disconnecting non-critical load branches in a precise manner. After switching, real-time sampling and observation are used to automatically correct key parameters such as voltage, frequency, and power, achieving zero-perception switching.

[0073] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0074] It should be understood that the detailed description of the technical solutions of the present invention using the preferred embodiments above is illustrative and not restrictive. A person skilled in the art, after reading the present specification, may modify the technical solutions described in the embodiments or replace some of the technical features therein with equivalents; such modifications or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A control method for seamless on-grid and off-grid switching of a reconfigurable battery energy storage system, characterized in that: The method includes: Build a real-time grid status perception and system operation mode prediction unit. Through high-speed sampling voltage, frequency, and phase sensors and communication interfaces, it obtains millisecond-level panoramic information flow of grid status and calculates the optimal switching timing and system reconstruction target topology solution. Deploy high-precision voltage transformers, phase-locked loops, and fiber-optic communication interfaces to continuously collect three-phase voltage, current, frequency, and phase angle. These quantities are then converted into parameters in a stationary coordinate system through transformation. Using discrete Fourier transforms, the harmonic distortion rate and voltage sag depth are calculated in real time to predict switching timing and generate the target topology solution. Deep learning is used to predict disturbance trends and pre-calculate the control parameters of the two operating modes in parallel before switching occurs. If the risk of voltage collapse persists, the system is forced to switch back to the initial stable mode within a limited time. Self-checking and safety control work together. After switching, they automatically correct key parameters such as voltage, frequency, and power through real-time sampling and observation, shortening power recovery and frequency stabilization time. Zero-perception switching significantly shortens the system power recovery time and frequency stabilization time after switching.

2. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1 is characterized in that: The data sampling rate of the real-time grid status perception unit is 100,000 times per second or more. Specifically, it continuously collects key grid parameters such as three-phase voltage, current, frequency and phase angle through the collaborative work of high-precision voltage transformers, phase-locked loops and optical fiber communication interfaces, and converts the three-phase quantities into parameters in a stationary coordinate system through Clarke transform. At the same time, discrete Fourier transform is used to calculate the harmonic distortion rate and voltage sag depth in real time.

3. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1, characterized in that: The disturbance prediction model uses a deep learning algorithm based on a long short-term memory network to process the collected time series data in the form of a sliding window, and outputs the grid stability probability and disturbance type vector within the next 50 milliseconds. The disturbance types include voltage swell, sag, frequency oscillation, harmonic excess and communication interruption. When the predicted grid stability probability is lower than the set threshold or the disturbance risk is higher than the set threshold, the mode prediction mechanism is automatically triggered.

4. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1, characterized in that: During the mode prediction stage, the main controller can run the power allocation algorithm in the grid-connected mode and the virtual synchronous machine control algorithm in the off-grid mode in parallel, and calculate the switching timing and target topology scheme in advance by comparing the system state evaluation functions in the two modes.

5. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1, characterized in that: Before switching, the deep learning model LSTM is used to predict disturbance trends, and the control parameters for both grid-connected and off-grid operating modes are pre-calculated in parallel. Combined with cross-cycle rolling optimization, a switching plan is generated that prioritizes disconnecting non-critical load branches in a precise manner. If the risk of grid voltage collapse persists, the system is forced to switch back to the initial stable mode within ten milliseconds. All protection actions are based on fail-safe principles.

6. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1, characterized in that: Through the coordinated operation of self-checking and safety control, and the use of high-frequency real-time sampling, observers and model predictive control algorithms, key parameters of voltage, frequency and power can be automatically corrected, which can shorten the power recovery time after switching to less than eight milliseconds and the frequency stabilization time to less than fifteen milliseconds.

7. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1, characterized in that: The safety control layer has a multi-layer protection mechanism that can automatically perform fault isolation and redundancy compensation under extreme conditions such as a communication interruption lasting two hundred milliseconds or a complete failure of a single module.

8. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1, characterized in that: The method adopts a self-checking and safety-oriented control collaborative mechanism throughout the switching process. The system can automatically correct control instructions according to the operating parameters detected in real time, achieve zero perception during the switching process, and can quickly recover to the safe operating area after the switching.

9. The method for seamless on-grid and off-grid switching control of a reconfigurable battery energy storage system according to claim 1, characterized in that: The method runs the grid-connected and off-grid control modes in parallel in the main controller, completes the pre-calculation of all key parameters and target topology before switching occurs, and combines with the distributed collaborative switching mechanism to achieve nanosecond-level synchronous issuance of switching instructions.

10. A reconfigurable battery energy storage system on-grid and off-grid seamless switching control device, the device being applicable to the method according to any one of claims 1 to 9, characterized in that: The device comprises: The real-time grid status perception and system operation mode prediction unit obtains millisecond-level panoramic information flow of the grid status through high-speed sampling voltage, frequency, and phase sensors and communication interfaces; A data processing module equipped with a high-precision voltage transformer, phase-locked loop, and fiber-optic communication interface; a disturbance prediction model based on a long short-term memory network; and power distribution and voltage support algorithms running in parallel in both on-grid and off-grid modes within the main controller; Deep learning predicts disturbance trends and combines cross-cycle rolling optimization to generate a switching plan that prioritizes disconnecting non-critical load branches in a precise manner. After switching, real-time sampling and observation are used to automatically correct key parameters such as voltage, frequency, and power, achieving zero-perception switching.

Images