Coordinated control method of multi-port solid state transformer for ac-dc hybrid microgrid

Abstract

The application belongs to the technical field of electric power, and discloses a multi-port solid-state transformer cooperative control method for AC-DC hybrid microgrid, power mutation trend is predicted in advance by constructing a multi-port SST mathematical model and combining a source and load fluctuation prediction module; a feedforward decoupling matrix control and a least square iteration algorithm are adopted to dynamically update a coupling coefficient matrix, a cross-coupling feedforward compensation, an error correction factor and a coupling disturbance suppression link are set in a current inner loop controller, a multi-input multi-output coupling system is converted into multiple independent single-input single-output systems, and the decoupling precision and dynamic response speed are improved; an adaptive droop strategy is designed by fusing port operating states and load priorities, a voltage-active power characteristic is adopted for active power, a frequency-reactive power characteristic is adopted for reactive power, a droop coefficient is dynamically adjusted, a dynamic virtual impedance compensation is matched, and distribution deviation caused by line impedance difference is eliminated.

Description

Technical Field

[0001] This invention belongs to the field of power technology, specifically a collaborative control method for multi-port solid-state transformers for AC / DC hybrid microgrids. Background Technology

[0002] As an important carrier of new power systems, AC / DC hybrid microgrids possess the advantage of organically integrating AC and DC grids. Multi-port solid-state transformers (SSTs), as the core energy hub of AC / DC hybrid microgrids, integrate multiple functions such as voltage transformation, electrical isolation, bidirectional power flow, and power quality management. They can simultaneously connect medium-voltage AC distribution networks, medium-voltage DC buses, low-voltage AC loads, low-voltage DC loads, and energy storage systems. In existing technologies, Chinese invention patent CN112436508B discloses a solid-state transformer and its control method for uninterrupted operation under fault conditions, achieving smooth transition of submodule capacitor voltage during fault transients and suppression of capacitor voltage fluctuations under all operating conditions. The core journal article "Coordinated Operation Control Strategy for AC / DC Hybrid Distribution Systems Based on Solid-State Transformers" proposes a hierarchical coordinated control scheme for AC / DC hybrid distribution systems, achieving power balance within the subgrid and seamless switching between multiple operating modes. However, existing multi-port solid-state transformer collaborative control technologies still have the following technical problems: There are complex electrical coupling relationships between the converters in a multi-port SST. When the power of one port changes abruptly, it will cause fluctuations in the voltage and power of other ports, affecting the stability of the system. Existing decoupling technologies mostly rely on hardware structure optimization or simple control compensation, which have slow dynamic response speed and low decoupling accuracy.

[0003] Traditional droop control uses a fixed droop coefficient, which cannot adapt to dynamic changes on both the source and load sides, resulting in uneven power distribution between different ports, with some converters operating under overload while others operate under light load. Existing adaptive droop control mostly adjusts only for a single power type, without combining port operating status and load priority for coordinated optimization, and it does not solve the distribution deviation caused by line impedance differences.

[0004] Existing control strategies mostly control AC or DC bus voltages individually, lacking a coordinated regulation mechanism for AC and DC side voltages. When a large power disturbance occurs in the system, voltage over-limit problems are likely to occur. Existing hierarchical control or mode switching strategies have a lag in response and cannot achieve real-time linkage regulation of AC and DC bus voltages.

[0005] When a fault occurs in the distribution network, the existing SST is prone to disconnection due to overcurrent or overvoltage protection, which cannot provide continuous power supply to critical loads in the microgrid and reduces the power supply reliability of the system. Existing fault handling mostly only achieves fault isolation or simple mode switching, without forming a coordinated mechanism of hierarchical processing, port reconfiguration and critical load protection, and the fault tolerance capability is insufficient.

[0006] Existing technologies mostly adopt an architecture of independent control at a single port and passive linkage between ports only through bus voltage, lacking a global-level active coordination mechanism, and thus failing to achieve global linkage optimization of multiple control links. Summary of the Invention

[0007] The purpose of this invention is to provide a multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids to solve one or more of the problems mentioned in the background art.

[0008] To achieve the above objectives, this invention provides the following technical solution: a multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids. This method constructs a three-level closed-loop multi-port collaborative control architecture: the first level is an independent current inner loop decoupling control closed loop for each port; the second level is a collaborative closed loop for power distribution and voltage stability among multiple ports; and the third level is a global reinforcement learning scheduling optimization closed loop. The three-level closed loop adopts a hierarchical linkage signal interaction mechanism, achieving step-by-step transmission through differentiated control cycles: the shortest current inner loop cycle ensures rapid response, the middle power loop cycle achieves precise allocation, and the longest voltage and scheduling loop cycles take into account global optimization. The first-level closed loop is based on a feedforward decoupling matrix control method, combined with a least squares iterative algorithm to dynamically update the coupling coefficient matrix, and cross-coupling feedforward compensation and error correction are set in the current inner loop controller of each port. The factor and coupling disturbance suppression stage decouples the multi-input multi-output coupled system into multiple independent single-input single-output systems. The second-stage closed-loop design employs an adaptive droop control strategy, dynamically adjusting the droop coefficient based on voltage-active power and frequency-reactive power characteristics, and combining port operating status and load priority to adjust the droop coefficient. A virtual impedance compensation stage is also set up to optimize power allocation. The third-stage closed-loop uses the DDPG reinforcement learning algorithm to achieve adaptive optimization of power scheduling. The core dimensions of the reward function are port power coupling degree, power allocation deviation, AC / DC bus voltage deviation, and fault state weight. The output action space includes four types of control variables: port droop coefficient, decoupling coefficient, virtual impedance parameter, and mode switching threshold. The scheduling result is directly linked to the control parameters of the first and second-stage closed loops through the three-stage feedback path. Furthermore, in the power decoupling stage, a mathematical model of a multi-port solid-state transformer is established. Combined with the source-load fluctuation prediction module, dynamic data of distributed power output and load demand are collected in real time to predict power change trends in advance and analyze the power transmission characteristics and cross-coupling mechanism between each port. Simultaneously collect two types of characteristic data: real-time power coupling strength and bus voltage fluctuation amplitude at each port. Construct a multi-dimensional input sequence that includes source-load timing features and port electrical coupling features. Output prediction results and generate power coupling disturbance prediction labels, which are directly mapped to the coupling coefficient update trigger logic of the feedforward decoupling matrix.

[0009] The mathematical model is constructed based on the multi-port equivalent circuit theory. It independently models the medium-voltage AC port, medium-voltage DC port, low-voltage AC port, low-voltage DC port and energy storage port, respectively. It fully incorporates actual physical parameters such as converter conduction loss, switching loss, line parasitic resistance and parasitic inductance. At the same time, it accurately describes the dynamic characteristics of bidirectional power flow and electrical cross-connection between ports, so that the model can truly reflect the operating law of the system under different operating conditions such as steady state, transient state and disturbance.

[0010] Based on the feedforward decoupling matrix control method, combined with the least squares iterative algorithm, the coupling coefficient matrix between the control variables of each port and the output power is dynamically updated to achieve real-time adaptive adjustment of the coupling coefficient. In the current inner loop controller of each port, a cross-coupling feedforward compensation term and an error correction factor are set to transform the multi-input multi-output coupled system into multiple independent single-input single-output systems. At the same time, a coupling disturbance suppression circuit is set.

[0011] Least squares iterative update formula for coupling coefficient matrix: This represents the coupling coefficient matrix between the control variables and output power of the multi-port solid-state transformer obtained after the (k+1)th iteration. This represents the coupling coefficient matrix between the control variables and output power of the multi-port solid-state transformer obtained after the k-th iteration. This represents the step size factor of the least squares iterative algorithm, with a value ranging from 0 to 1. It is used to control the update rate of the coupling coefficient matrix and the convergence speed of the algorithm. This represents the actual output power vector of each port of a multi-port solid-state transformer, directly reflecting the true power output state of the port. This represents the control input vector of each port of the multi-port solid-state transformer, and signifies the control command signal applied by the system to each port. This represents the regularization coefficient, used to avoid the problem of singular matrices during matrix inversion and to ensure the stability of the algorithm. It represents the identity matrix and, in conjunction with regularization coefficients, performs matrix regularization, improving the reliability of matrix operations.

[0012] Furthermore, the power allocation stage is designed with an adaptive droop control strategy, which combines the port operation status evaluation module and the load priority weight coefficient to dynamically adjust the droop coefficient of each port according to the rated capacity, real-time operation status and system power demand of each port converter. For active power control, voltage-active power adaptive droop characteristics are adopted, and the droop slope is dynamically adjusted in combination with load priority; for reactive power control, frequency-reactive power adaptive droop characteristics are adopted, and a reactive power distribution error feedback mechanism is set up to calibrate the droop coefficient in real time. Both types of droop characteristics adopt a segmented adaptive adjustment mechanism, which divides the system into three ranges according to the system operating load level: light load, rated load, and heavy load. The light load range uses a gentle droop slope to maintain voltage and frequency stability, the rated load range uses a standard droop slope to ensure accurate power distribution, and the heavy load range uses a steep droop slope to prevent equipment overload. The different ranges are smoothly connected without abrupt changes.

[0013] Voltage-Active Power Adaptive Droop Formula: It represents the real-time active power output value of a single port of a multi-port solid-state transformer, which intuitively represents the current active power output of the port; It represents the rated active power reference value of a single port of a multi-port solid-state transformer, serving as the core reference standard for active power control; It represents the adaptive droop coefficient corresponding to voltage and active power, and its value is dynamically adjusted according to load priority and port operating status; This represents the rated voltage reference value of the AC / DC bus in an AC / DC hybrid microgrid, and is the standard reference value for bus voltage control. It represents the real-time voltage value of the AC / DC bus in the AC / DC hybrid microgrid, truly reflecting the current voltage operating status of the bus.

[0014] Frequency-reactive power adaptive droop formula: It represents the real-time reactive power output value of a single port of a multi-port solid-state transformer, intuitively representing the current reactive power output of the port; It represents the rated reactive power reference value of a single port of a multi-port solid-state transformer, serving as the core reference standard for reactive power control; It represents the adaptive droop coefficient corresponding to frequency and reactive power, and the value is calibrated and optimized in real time according to the reactive power distribution error. This represents the rated frequency reference value of the AC / DC hybrid microgrid system, which is the standard reference value for system frequency control. It represents the real-time frequency value of the AC / DC hybrid microgrid system, truly reflecting the current frequency operating status of the system.

[0015] The design incorporates a dynamic virtual impedance compensation mechanism that adaptively adjusts virtual impedance parameters based on real-time line impedance detection values. A power distribution balance monitoring module is also included, which automatically triggers a collaborative optimization mechanism between the droop coefficient and virtual impedance when a distribution deviation exceeds a threshold.

[0016] The real-time line impedance detection adopts a synchronous sampling and calculation method of port terminal voltage and output current. A high-precision electrical sampling unit is connected to the front end of each port output line to collect the voltage amplitude and phase information of the two ends of the line and the port output current data in real time. The equivalent impedance value of the line is calculated based on the voltage and current correlation characteristics under steady-state operation. The detection process is not affected by interference factors such as load switching and power fluctuation. The detection results are synchronized to the dynamic virtual impedance compensation unit in real time.

[0017] Furthermore, during the voltage stabilization phase, an AC / DC bus voltage collaborative control architecture is established, with the medium-voltage DC bus voltage as the core control variable of the system. Combined with the bus voltage prediction control module, the control parameters are adjusted in advance based on the source load fluctuation prediction data. In grid-connected mode, the medium-voltage AC port converter is responsible for maintaining the stability of the medium-voltage DC bus voltage and providing dynamic reactive power support to the distribution network. At the same time, it is linked with the low-voltage side bus voltage controller to realize the linkage regulation of the medium-voltage and low-voltage bus voltages. A bidirectional mutual feedback compensation mechanism for AC and DC bus voltages is constructed. When the deviation of the medium-voltage DC bus voltage exceeds a preset threshold, the droop coefficients of the low-voltage AC and low-voltage DC ports are adjusted synchronously. When the fluctuation of the low-voltage bus voltage exceeds the threshold, the feedforward compensation amount of the medium-voltage AC port converter is corrected in reverse, thereby realizing bidirectional linkage regulation of AC and DC bus voltages.

[0018] In islanded mode, the energy storage port converter switches to voltage construction mode, takes over the control of medium-voltage DC bus voltage, and simultaneously optimizes the control parameters of low-voltage AC and DC bus voltage. When a large-power disturbance occurs in the system, the AC and DC side voltage coordinated regulation plan is activated, and a bus voltage stability margin monitoring module is set up to evaluate the bus voltage stability in real time and trigger the regulation mechanism in advance.

[0019] Furthermore, the mode switching phase is based on the AC / DC bus voltage collaborative control architecture established in the voltage stabilization phase. It designs a multi-operation mode switching logic based on a state machine with multi-parameter fusion, defines three operation modes: grid-connected operation, islanded operation, and fault recovery, and combines them with the distribution network status monitoring module. Real-time acquisition of multi-dimensional parameters such as distribution network voltage, frequency, phase and fault signals; when a distribution network fault is detected, trigger mode switching; adopting predictive mode switching to adjust the control parameters of the energy storage port converter in advance. The medium-voltage AC port converter is switched from current control mode to blocking mode, and the energy storage port converter is switched from current control mode to voltage construction mode. When the distribution network returns to normal, pre-synchronization control is executed. The voltage amplitude, frequency and phase of the microgrid side are calibrated through the phase-locked loop. After matching, the grid connection switch is closed, and the mode switching fault self-diagnosis module and backup switching scheme are set.

[0020] The converter control mode switching is accompanied by a hardware linkage execution mechanism. When the software issues the mode switching command, it simultaneously triggers hardware actions such as converter drive signal latching, power switch device status verification, grid-connected switch auxiliary contact monitoring, and DC side capacitor voltage clamping. This ensures that the software control logic and hardware execution status are synchronized and matched, avoiding safety risks such as bus short circuits, power backflow, and voltage spikes caused by misalignment between software commands and hardware actions.

[0021] Furthermore, the fault traversal phase is designed with a hierarchical fault traversal strategy based on intelligent fault type identification. Combined with the fault diagnosis module, the fault type, severity, and fault location are identified in real time through wavelet transform algorithm, and differentiated control measures are set for different fault scenarios. When a temporary voltage drop fault occurs in the distribution network, the medium-voltage AC port converter is switched to low voltage ride-through mode to inject reactive current into the distribution network to support the grid voltage. At the same time, the output current is limited to within 1.5 times the rated value, and a current fluctuation suppression circuit is set. When a permanent fault occurs in the distribution network, the grid-connected switch is disconnected, the system switches to islanded operation mode, and the critical load priority protection mechanism is activated to prioritize power allocation to critical loads. When a fault occurs at a port within the SST, the drive signal of the corresponding port converter is blocked. Through intelligent port reconfiguration control, the power allocation strategy of non-faulty ports is dynamically adjusted to transfer the load of the faulty port to other normal ports. A fault recovery auxiliary module is set up so that port reconfiguration reset and power allocation recovery are automatically achieved after the fault is cleared.

[0022] Internal port fault diagnosis focuses on collecting two core insulation safety parameters: port insulation resistance and leakage current. Real-time detection values ​​are compared with the equipment's factory safety thresholds. At the same time, it combines multiple parameters such as port over-temperature, over-current, and over-voltage electrical signals for joint judgment. A single parameter abnormality only triggers an early warning, while multiple parameters exceeding limits confirm the fault. Multiple judgment conditions effectively avoid false blocking and false reconfiguration caused by single-point abnormality of sensors.

[0023] Furthermore, the power scheduling phase establishes a dynamic port power scheduling mechanism, which, combined with the intelligent scheduling decision module, uses a reinforcement learning algorithm to achieve adaptive optimization of power scheduling. Real-time acquisition of multi-dimensional parameters such as power of each port, bus voltage, and equipment operating status; combined with the current system operating mode, fault status, and load priority, dynamically allocating power reference values ​​for each port; and adjusting the control parameters of each port converter in real time in response to high-frequency fluctuations in source and load. A scheduling optimization evaluation module is set up to evaluate the rationality and economy of power scheduling in real time, linking power decoupling, power allocation, voltage stabilization, mode switching and fault ride-through.

[0024] The beneficial effects of this invention are as follows: 1. This invention constructs a multi-port SST mathematical model and combines it with a source-load fluctuation prediction module to predict power mutation trends. It uses feedforward decoupling matrix control and least squares iterative algorithm to dynamically update the coupling coefficient matrix. In the current inner loop controller, it sets up cross-coupling feedforward compensation, error correction factor and coupling disturbance suppression links to transform the multi-input multi-output coupled system into multiple independent single-input single-output systems, thereby improving decoupling accuracy and dynamic response speed, and eliminating voltage and power fluctuations caused by port power mutations.

[0025] 2. This invention designs an adaptive droop strategy that integrates port operating status and load priority. It dynamically adjusts the droop coefficient based on the voltage-active power characteristics for active power and the frequency-reactive power characteristics for reactive power. Combined with dynamic virtual impedance compensation, it eliminates the distribution deviation caused by line impedance differences. At the same time, it triggers a collaborative optimization mechanism through a power distribution balance monitoring module. It can flexibly allocate power according to converter capacity, real-time status, and power demand, avoiding inefficient operation of some converters being overloaded and others being lightly loaded, and adapting to dynamic fluctuations on both the source and load sides.

[0026] 3. This invention establishes an AC / DC voltage collaborative architecture with medium-voltage DC bus voltage as the core. It achieves seamless switching between grid connection / islanding / fault recovery through predictive pre-synchronization. It completes intelligent fault identification, port reconstruction and critical load protection through wavelet transform hierarchical fault ride-through strategy. It achieves adaptive optimization of power scheduling by combining reinforcement learning algorithm. It realizes voltage stability under high power disturbance, continuous power supply during faults and intelligent power optimization scheduling, and is suitable for complex operation scenarios of AC / DC hybrid microgrids. Attached Figure Description

[0027] Figure 1 This is the complete main flowchart of the multi-port solid-state transformer collaborative control of the present invention; Figure 2 This is a flowchart of the operation mode switching control sub-flowchart of the present invention.

[0028] Figure 1The complete main flowchart of the collaborative control shown covers six control stages: the power decoupling stage achieves power decoupling between ports through a feedforward decoupling matrix and a least-squares iterative algorithm; the power allocation stage achieves balanced power allocation across multiple ports through adaptive droop control and virtual impedance compensation; the voltage stabilization stage establishes a collaborative control architecture for AC and DC bus voltages with the medium-voltage DC bus voltage as the core; the mode switching stage achieves predictive seamless switching between three operating modes—grid-connected, islanded, and fault recovery—through a multi-parameter fusion state machine; the fault ride-through stage achieves graded fault ride-through and port reconstruction through wavelet transform fault diagnosis; the power scheduling stage achieves adaptive optimization of global power scheduling through the DDPG reinforcement learning algorithm; and the above six stages achieve signal interaction and parameter coordination through a hierarchical linkage mechanism of a three-level closed-loop architecture.

[0029] Figure 2 In the operation mode switching control sub-flowchart shown, the multi-parameter fusion state machine uses the distribution network voltage amplitude, frequency stability, phase deviation, fault signal, energy storage SOC state, and microgrid power balance state as core judgment parameters, and obtains the comprehensive judgment value for mode switching through weighted fusion calculation. The fault recovery mode is used as the intermediate state for the system to transition from islanding to grid connection after the distribution network fault is cleared. In this mode, the energy storage port voltage construction function is maintained to keep the internal power supply of the microgrid stable, and then the special control parameters of islanding operation are gradually rolled back, and the pre-synchronization calibration process is started simultaneously. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] like Figures 1 to 2 As shown, this embodiment of the invention provides a multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids. After the system starts up, it first performs a full-dimensional initialization operation, sequentially completing hardware port self-test, converter control parameter reset, AC / DC bus voltage reference value setting, source-load prediction module calibration, and control cycle synchronous calibration. At the same time, it detects the operating status of each sensor and communication module. After confirming that there are no abnormalities in all units, it enters the steady-state control preparation stage. If an initialization abnormality is detected, an audible and visual warning is triggered and the faulty port is locked.

[0032] In this embodiment of the invention, the power decoupling stage establishes a mathematical model of a multi-port solid-state transformer, and combines it with a source-load fluctuation prediction module to collect dynamic data on the output of distributed power sources and load demand in real time, predict power change trends in advance, and analyze the power transmission characteristics and cross-coupling mechanisms between ports. The source-load fluctuation prediction module adopts time-series prediction logic. It takes historical output data of distributed power sources, meteorological parameters, load power consumption patterns and time period characteristics as input information, completes short-cycle power fluctuation prediction according to the system control cycle, and outputs the port power fluctuation amplitude, mutation probability and trend direction in the future multiple control cycles. The prediction results are synchronously transmitted to the power decoupling and voltage stability control link.

[0033] Based on the feedforward decoupling matrix control method, combined with the least squares iterative algorithm, the coupling coefficient matrix between the control variables of each port and the output power is dynamically updated to achieve real-time adaptive adjustment of the coupling coefficient. In the current inner loop controller of each port, a cross-coupling feedforward compensation term and an error correction factor are set to transform the multi-input multi-output coupled system into multiple independent single-input single-output systems. At the same time, a coupling disturbance suppression circuit is set.

[0034] The cross-coupling feedforward compensation term generates compensation commands based on the real-time coupling strength and power transmission direction of each port, and directly superimposes them into the current inner loop setpoint to offset the coupling effect in advance; the error correction factor tracks the actual deviation of power and voltage after decoupling in real time, and dynamically adjusts the correction coefficient according to the magnitude of the deviation. The larger the deviation, the stronger the correction amplitude, thus reducing the error between the theoretical control value and the actual output value. The cross-coupling feedforward compensation parameters will be adaptively tuned according to the system operating conditions and changes in coupling strength. In the stable grid-connected operation scenario, steady-state tuning parameters are used, and in dynamic operating conditions such as sudden changes in source load power and mode switching, they will automatically switch to dynamic tuning parameters. At the same time, the compensation parameter base value is configured according to the differences in the type and power level of the port access device.

[0035] The error correction factor adopts a segmented adaptive adjustment rule for deviation, dividing the power and voltage deviations into three levels: small deviation, medium deviation, and large deviation. Small deviations are corrected only by a small amount of smoothing, medium deviations are corrected by increasing the correction coefficient linearly, and large deviations are corrected by activating the limit correction coefficient and triggering an emergency update of the coupling coefficient matrix. At the same time, an upper limit threshold for the correction coefficient is set to prevent overcorrection from causing system oscillation.

[0036] The system is specially equipped with a decoupling effect closed-loop verification unit, which collects the cross-correlation degree and fluctuation deviation data of power and voltage of each port before and after decoupling in real time. By quantitatively calculating the decoupling residual interference rate, the current decoupling accuracy level is accurately determined. When the decoupling residual rate is higher than the preset safety tolerance value, the system immediately and automatically triggers the expedited iterative update process of the coupling coefficient matrix to keep the decoupling accuracy stable within the target high standard range.

[0037] In this embodiment of the invention, the power allocation stage is designed with an adaptive droop control strategy. Combining the port operation status evaluation module and the load priority weight coefficient, the droop coefficient of each port is dynamically adjusted according to the rated capacity, real-time operation status and system power demand of each port converter. The port operation status assessment module performs real-time assessments from multiple dimensions, including converter operating temperature, switching frequency, output current margin, port insulation level, and power transmission efficiency, and quantifies and generates port health and load-bearing capacity scores. The score results directly serve as the core basis for adjusting the droop coefficient.

[0038] The adaptive droop coefficient setting includes upper and lower limit constraints and a graded adjustment step size mechanism. During the adjustment process, small step size fine-tuning is used first to achieve a smooth transition. Only when the power distribution deviation continues to expand and exceeds the smooth adjustment range will it automatically switch to a large step size fast correction mode. At the same time, it is combined with the maximum value of the lock coefficient of the port rated load capacity to avoid problems such as port overload and abnormal voltage deviation caused by excessive coefficient.

[0039] For active power control, voltage-active power adaptive droop characteristics are adopted, and the droop slope is dynamically adjusted in combination with load priority; for reactive power control, frequency-reactive power adaptive droop characteristics are adopted, and a reactive power distribution error feedback mechanism is set up to calibrate the droop coefficient in real time. The load priority weighting coefficient is divided into three levels according to the importance of the load power supply: critical load, important load, and general load. The critical load has the highest weight and the general load has the lowest weight. The weighting coefficient can be manually set or automatically adjusted according to the system operation scenario and power supply plan. When power is scarce, the power supply needs of high-weight loads are given priority, and when power is sufficient, the power demand of each level of load is met in a balanced manner.

[0040] The reactive power distribution error feedback mechanism collects the deviation data between the reactive power output value and the theoretical distribution value of each port in real time. After filtering, the deviation data is transmitted to the droop control unit. The reactive power droop coefficient is finely adjusted periodically according to the direction and magnitude of the deviation to reduce the reactive power distribution deviation.

[0041] The design incorporates a dynamic virtual impedance compensation mechanism. Based on real-time detection of line impedance, the virtual impedance parameters are adaptively adjusted to eliminate power distribution deviations caused by line impedance differences. A power distribution balance monitoring module is also included. When a distribution deviation exceeds a threshold, a collaborative optimization mechanism between the droop coefficient and the virtual impedance is automatically triggered.

[0042] The power distribution balance monitoring module collects parameters such as output power, rated capacity ratio, and line transmission loss of each port in real time, calculates the power distribution balance index and deviation rate, presets the allowable threshold for power distribution deviation, and continuously compares the actual deviation with the threshold.

[0043] The dynamic virtual impedance and adaptive droop coefficient adopt a time-series coordinated adjustment rule. When a power distribution deviation is detected, the virtual impedance fine-tuning is initiated first to quickly eliminate the basic deviation caused by the line impedance. After the virtual impedance parameter stabilizes, the droop coefficient is adjusted in conjunction to achieve precise distribution. The two adjustment processes cooperate with each other without interfering with each other, avoiding power oscillation and voltage fluctuation caused by simultaneous adjustment.

[0044] In this embodiment of the invention, during the voltage stabilization phase, a collaborative control architecture for AC and DC bus voltage is established. The medium-voltage DC bus voltage is used as the core control variable of the system. Combined with the bus voltage prediction control module, the control parameters are adjusted in advance based on the source load fluctuation prediction data. The bus voltage prediction and control module takes the real-time AC / DC bus voltage, predicted source load power, and port operating status as inputs to predict the bus voltage change trend over multiple control cycles, sets a safe and stable voltage range, and initiates adjustment actions in advance when the voltage is predicted to exceed the limit, adjusting the control parameters according to a preset step size.

[0045] The AC / DC bus voltage coordinated control is set with a clear voltage protection priority. The priority is from high to low as follows: medium voltage DC bus, medium voltage AC bus, low voltage DC bus, and low voltage AC bus. When the system power is scarce or a disturbance occurs, the voltage of the high-level bus is guaranteed to be stable within the rated range first, and then the voltage of the low-level bus is adjusted step by step.

[0046] In grid-connected mode, the medium-voltage AC port converter is responsible for maintaining the stability of the medium-voltage DC bus voltage and providing dynamic reactive power support to the distribution network. At the same time, it is linked with the low-voltage side bus voltage controller to realize the linkage regulation of the medium-voltage and low-voltage bus voltages. In islanded mode, the energy storage port converter switches to voltage construction mode, takes over the control of medium-voltage DC bus voltage, and simultaneously optimizes the control parameters of low-voltage AC and DC bus voltage. When a large-power disturbance occurs in the system, the AC and DC side voltage coordinated regulation plan is activated, and a bus voltage stability margin monitoring module is set up to evaluate the bus voltage stability in real time and trigger the regulation mechanism in advance.

[0047] The bus voltage stability margin monitoring module uses the rated voltage as a benchmark and calculates the difference between the real-time voltage and the safe range as the voltage stability margin. It presets a margin warning threshold and a critical threshold. When the margin is lower than the warning threshold, it initiates a slight adjustment. When the margin is lower than the critical threshold, it initiates a heavy adjustment. Through multi-dimensional linkage control, it quickly pulls the voltage back to the safe range.

[0048] The high-power disturbance coordinated control plan is implemented in stages according to the disturbance amplitude. First, the source and scope of the disturbance are quickly identified. The response rate of the medium-voltage DC bus voltage controller is increased simultaneously. Then, the control parameters of the low-voltage AC / DC bus voltage are adjusted in conjunction to suppress voltage deviation. At the same time, the power output of each port is coordinated to quickly smooth out the disturbance impact. The voltage recovery status is monitored in real time throughout the process. After the voltage returns to the stable range, the control parameters are gradually restored to the steady-state value.

[0049] In this embodiment of the invention, the mode switching stage is based on the AC / DC bus voltage collaborative control architecture established in the voltage stabilization stage. It designs a state machine multi-operation mode switching logic based on multi-parameter fusion, defines three operation modes: grid-connected operation, islanded operation, and fault recovery, and combines them with the distribution network status monitoring module. The multi-parameter fusion state machine uses the distribution network voltage amplitude, frequency stability, phase deviation, fault signal, energy storage SOC state, and microgrid power balance state as core judgment parameters. It obtains a comprehensive judgment value for mode switching through weighted fusion calculation and distinguishes between three states: grid-connected, islanded, and fault recovery according to the preset judgment threshold range, thus avoiding erroneous mode switching caused by fluctuations in a single parameter.

[0050] The fault recovery mode is used as an intermediate state for the system to transition from islanding to grid connection after the distribution network fault is cleared. In this mode, the energy storage port voltage construction function is maintained to keep the power supply inside the microgrid stable. Then, the special control parameters of islanding operation are gradually rolled back, and the pre-synchronization calibration process is started simultaneously. After the parameters of the distribution network and the microgrid are fully matched and there are no abnormalities in the system, the system is smoothly switched to grid-connected operation mode.

[0051] Real-time acquisition of multi-dimensional parameters such as distribution network voltage, frequency, phase and fault signals avoids erroneous switching caused by single parameter judgment. When a distribution network fault is detected, mode switching is triggered. Predictive mode switching is adopted to adjust the control parameters of the energy storage port converter in advance. The early warning signal for predictive mode switching comes from three key characteristic quantities of the distribution network: voltage drop rate, frequency drift rate, and voltage harmonic distortion rate. The linear trend fitting algorithm is used to predict the changing trend of the distribution network operation status in real time. When the characteristic quantity is detected to be continuously deteriorating and about to trigger the fault threshold, the switching preparation is initiated 1-2 system control cycles in advance.

[0052] The medium-voltage AC port converter is switched from current control mode to blocking mode, and the energy storage port converter is switched from current control mode to voltage construction mode. When the distribution network returns to normal, pre-synchronization control is executed. The voltage amplitude, frequency and phase of the microgrid side are calibrated through the phase-locked loop. After matching, the grid connection switch is closed, and the mode switching fault self-diagnosis module and backup switching scheme are set.

[0053] The mode switching fault self-diagnosis module monitors the voltage phase, control commands, switch status, and converter operating parameters in real time during the switching process to determine whether the switching is being performed normally. The backup switching scheme is automatically activated when the main switching logic fails, and performs degraded switching operations step by step according to the preset priority.

[0054] The pre-synchronization control sets a synchronization judgment threshold to control the voltage amplitude deviation, frequency deviation, and phase deviation between the microgrid and the distribution network within a very small industrial-grade allowable range. When the three key parameters have been stably met for several consecutive system control cycles without any drift trend, the synchronization locking mechanism is immediately activated to prevent synchronization failure caused by parameter fluctuations. After the locking is completed and a stable state is maintained, the grid connection switch closing operation is performed to ensure that there is no voltage surge or power fluctuation at the moment of grid connection.

[0055] In this embodiment of the invention, the fault traversal stage is designed with a hierarchical fault traversal strategy based on intelligent fault type identification. Combined with the fault diagnosis module, the fault type, severity and fault location are identified in real time through wavelet transform algorithm, and differentiated control measures are set for different fault scenarios. Based on fault characteristics, impact range, and duration, the fault diagnosis module classifies system operation faults into four levels: minor transient disturbance, moderate voltage drop, severe permanent fault, and internal port fault. Different fault levels correspond to different response speeds, control strategies, and protection actions. Minor faults only require disturbance suppression without switching operating modes. Moderate faults activate the low voltage ride-through support. Severe faults directly disconnect the grid-connected switch and switch to islanding mode. Internal faults immediately execute port blocking and reconfiguration.

[0056] When a temporary voltage drop fault occurs in the distribution network, the medium-voltage AC port converter is switched to low voltage ride-through mode to inject reactive current into the distribution network to support the grid voltage. At the same time, the output current is limited to within 1.5 times the rated value, and a current fluctuation suppression circuit is set. The current fluctuation suppression circuit collects the instantaneous fluctuation value and rate of change of the output current in real time, and generates suppression commands using a closed-loop regulation method. This quickly smooths out current spikes and oscillations, keeping the output current fluctuation amplitude within a safe range and preventing sudden current changes from impacting converter devices and load equipment.

[0057] The graded fault ride-through system configures dedicated protection thresholds and action delays for different fault types. For temporary voltage drop faults, 50% of the rated voltage is used as the critical threshold. When the voltage is higher than the threshold, reactive power support is activated and delayed by 100ms to prevent false tripping. When the voltage is lower than the threshold, islanding mode is switched directly. For internal port faults, abnormal insulation parameters and over-temperature are used as trigger thresholds. After a fault is detected, the port is instantly blocked and the load transfer is completed after a 50ms delay.

[0058] When a permanent fault occurs in the distribution network, the grid-connected switch is disconnected, the system switches to islanded operation mode, and the critical load priority protection mechanism is activated to prioritize power allocation to critical loads. When a fault occurs at a port within the SST, the drive signal of the corresponding port converter is blocked. Through intelligent port reconfiguration control, the power allocation strategy of non-faulty ports is dynamically adjusted to transfer the load of the faulty port to other normal ports. A fault recovery auxiliary module is set up so that port reconfiguration reset and power allocation recovery are automatically achieved after the fault is cleared.

[0059] The fault recovery auxiliary module continuously monitors the voltage, current, and insulation parameters at the fault point. After confirming that the fault has been completely eliminated and the system parameters have returned to normal, it issues a reset permission signal. According to the initial operating state of the port and the power allocation rules, it gradually releases the port block and restores the control commands.

[0060] Intelligent port reconfiguration control follows the reconfiguration principles of prioritizing load capacity, proximity to the faulty port, and minimizing losses. It prioritizes selecting normal ports with sufficient capacity, closest proximity to the faulty port, and lowest transmission loss to take over the transferred load. At the same time, it adjusts the control parameters and power limits of the receiving port to avoid overloading the receiving port due to the transferred load.

[0061] The load transfer process at the fault port adopts a power smoothing strategy. The receiving port gradually increases its output power according to a preset slope, without putting the transferred load at full load all at once. At the same time, the operating temperature, current change rate and voltage stability of the receiving port are monitored in real time, and the electrical shock caused by power transfer is suppressed through closed-loop regulation.

[0062] In this embodiment of the invention, the power scheduling stage is based on the power decoupling, power allocation, voltage stabilization, mode switching and fault ride-through control achieved in the previous stages. It establishes a port power dynamic scheduling mechanism and, in conjunction with the intelligent scheduling decision module, uses a reinforcement learning algorithm to achieve adaptive optimization of power scheduling. The reinforcement learning algorithm adopts the DDPG architecture, which consists of a state perception module, an action decision module, a reward evaluation module, and an experience playback module connected in sequence. The input of the state perception module is the power of each port, AC and DC bus voltage, load priority, fault status, and source-load fluctuation prediction value. The output of the action decision module is the power reference value of each port and the converter control parameters. The reward evaluation module uses power distribution balance, voltage stability margin, and system operating loss as multi-objective reward functions. The specific model training steps are as follows: Initialize the parameters of the experience replay pool, Actor network, and Critic network; collect system state samples and store them in the experience replay pool; use least squares iterative updates to update network weights, with a training step size of 0.001 to ensure a balance between the stability of network weight updates and convergence speed; set the capacity of the experience replay pool to 10000 to cover a sufficient system operating state sample space; when the fluctuation amplitude of the reward function in multiple consecutive training cycles is less than the preset convergence threshold, it is determined to be converged and training is stopped; the above parameters are selected based on the dynamic response characteristics of the SST multi-port system and the convergence analysis of the DDPG algorithm in the multi-port power control scenario; complete the training with the goal of system stability and optimal power allocation.

[0063] A customized DDPG reinforcement learning algorithm is adopted, with port power coupling degree, power distribution deviation, AC / DC bus voltage deviation, and fault state weight as the core dimensions of the reward function. The output action space includes four types of control variables: droop coefficient of each port, decoupling coefficient, virtual impedance parameter, and mode switching threshold. The scheduling result directly links the control parameters of the entire stage of power decoupling, power distribution, voltage stability, mode switching, and fault ride-through through a three-level feedback path.

[0064] The experience playback module has a built-in intelligent sample screening and deduplication optimization mechanism, which automatically removes invalid random fluctuation samples, duplicate and redundant samples, and sensor abnormal error samples, and retains only high-quality samples containing effective decision information such as typical power fluctuations, mode switching, fault ride-through, and load switching. At the same time, the playback pool is dynamically updated according to the timeliness of the samples, and old and failed samples are eliminated in a timely manner.

[0065] The weights of various indicators in the multi-objective reward function are dynamically adapted according to the system operation mode. In grid-connected mode, the weights of operating losses and voltage stability are increased to pursue economic and efficient operation. In islanded mode, the weights of power distribution balance and load protection are increased. In fault ride-through mode, the weights of voltage stability and equipment safety are strengthened.

[0066] Real-time acquisition of multi-dimensional parameters such as power of each port, bus voltage, and equipment operating status; combined with the current system operating mode, fault status, and load priority, dynamically allocating power reference values ​​for each port; and adjusting the control parameters of each port converter in real time in response to high-frequency fluctuations in source and load. A scheduling optimization evaluation module is set up to evaluate the rationality and economy of power scheduling in real time, linking power decoupling, power allocation, voltage stabilization, mode switching and fault ride-through.

[0067] The scheduling optimization evaluation module uses power scheduling efficiency, voltage stability level, operating loss, and load power supply guarantee rate as core evaluation indicators. It completes the quantitative evaluation of scheduling effect at fixed intervals and feeds the evaluation results back to the reinforcement learning power scheduling module in real time, providing a basis for the next round of power scheduling parameter optimization. Each core evaluation indicator has a clearly defined pass / fail threshold and an excellent threshold. Power dispatch efficiency is judged based on port power response delay, voltage stability level is judged based on the deviation of AC / DC bus voltage from the rated value, operating loss is judged based on the overall system energy loss rate, and load power supply guarantee rate is judged based on the continuous power supply duration of critical loads. The comprehensive dispatch score is obtained by weighted calculation of multiple indicators, and the score result directly serves as the core basis for the next round of decision optimization of the reinforcement learning model.

[0068] The system's stable operation closed-loop control is divided into three feedback paths: fast current inner loop feedback, medium power loop feedback, and slow voltage and scheduling loop feedback. The three feedbacks adopt differentiated cycles: the current loop has the shortest cycle to ensure fast response, the power loop has a medium cycle to achieve precise allocation, and the voltage and scheduling loop has the longest cycle to take into account global optimization. The three feedbacks are passed on step by step without interfering with each other.

[0069] Simulation platform and system parameters This simulation is based on a multi-port solid-state transformer simulation system built on the MATLAB / Simulink 2024a platform for AC / DC hybrid microgrids. The core parameters are as follows: Medium voltage AC port: 10kV / 50Hz, rated capacity 1MVA Medium voltage DC port: ±750V, rated transmission power 800kW Low-voltage AC port: 380V / 50Hz, load capacity 300kW Low-voltage DC port: 480V, load capacity 200kW Energy storage port: Rated voltage 600V, capacity 500kWh Switching frequency: 10kHz, simulation step size: 1μs 2. Comparative experimental design The control group consisted of the traditional PI decoupling and fixed droop control strategy, while the experimental group consisted of the collaborative control method proposed in this invention. Comparative tests were conducted from five dimensions: steady-state performance, dynamic response, power distribution, mode switching, and fault ride-through.

[0070] 3. Analysis of Simulation Experiment Data and Results Table 1: Steady-state operating performance test data

[0071] Table 2: Dynamic Response Test Data of Source Charge Mutation Test scenario: The low-voltage DC load suddenly increases by 150kW at t=0.5s and suddenly decreases by 100kW at t=1.0s.

[0072] Table 3: Test data on the accuracy of adaptive droop power distribution Test scenario: Total system load 600kW, rated capacities of 4 power output ports are 300kW, 200kW, 200kW, and 100kW respectively.

[0073] Table 4: Performance Test Data for Operating Mode Switching Test scenario: The system switches from grid-connected mode to islanded mode at t=0.5s, and restores grid connection at t=1.5s.

[0074] Table 5: Fault Trip Performance Test Data Test scenario: At t=0.5s, a three-phase short-circuit fault occurs in the medium-voltage AC distribution network, and the voltage drops to 20% of the rated value, lasting for 200ms.

[0075] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0076] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A collaborative control method for multi-port solid-state transformers in AC / DC hybrid microgrids, characterized in that, A three-level closed-loop multi-port collaborative control architecture is constructed: the first level is an independent current inner loop decoupling control closed loop for each port; the second level is a collaborative closed loop for power distribution and voltage stability among multiple ports; and the third level is a global reinforcement learning scheduling optimization closed loop. The three-stage closed loop employs a hierarchical signal interaction mechanism, achieving step-by-step transmission through differentiated control cycles: the shortest current inner loop cycle ensures rapid response, the middle power loop cycle achieves precise allocation, and the longest voltage and scheduling loop cycles ensure global optimization. The first-stage closed loop is based on a feedforward decoupling matrix control method, combined with a least-squares iterative algorithm to dynamically update the coupling coefficient matrix. Cross-coupling feedforward compensation, error correction factors, and coupling disturbance suppression mechanisms are set in the current inner loop controllers at each port, decoupling the multi-input multi-output coupled system into multiple independent single-input single-output systems. The second-stage closed loop designs an adaptive droop control strategy, employing different control cycles for each stage. The droop coefficient is dynamically adjusted based on the voltage-active power and frequency-reactive power characteristics. The droop coefficient is adjusted in conjunction with the port operating status and load priority, and a virtual impedance compensation loop is set to optimize the power allocation effect. The third-level closed loop uses the DDPG reinforcement learning algorithm to achieve adaptive optimization of power scheduling. The core dimensions of the reward function are port power coupling degree, power allocation deviation, AC / DC bus voltage deviation, and fault state weight. The output action space includes four types of control variables: droop coefficient of each port, decoupling coefficient, virtual impedance parameter, and mode switching threshold. The scheduling result is directly linked to the control parameters of the first and second-level closed loops through the three-level feedback path. Simultaneously, a collaborative control architecture for AC / DC bus voltage is established with the medium-voltage DC bus voltage as the core. In grid-connected mode, the medium-voltage AC port converter maintains the stability of the medium-voltage DC bus voltage, while in islanded mode, the energy storage port switches to voltage construction mode. A multi-parameter fusion state machine switching logic is designed, defining two main operating modes: grid-connected operation and islanded operation, as well as a fault recovery transition mode. Predictive mode switching and pre-synchronization control are achieved by combining distribution network monitoring parameters. A graded fault ride-through strategy is designed, and differentiated control measures are adopted based on fault diagnosis and identification results, setting up output current suppression and critical load protection links.

2. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 1, characterized in that, In the first-level closed loop, the system combines the source-load fluctuation prediction module to collect dynamic data on the output of distributed power sources and load demand in real time, predict the trend of power change in advance, and quantitatively analyze the power transmission characteristics and cross-coupling strength between each port. The source-load fluctuation prediction module simultaneously collects two types of characteristic data: real-time power coupling strength and bus voltage fluctuation amplitude of each port, constructs a multi-dimensional input sequence that includes source-load time sequence characteristics and port electrical coupling characteristics, and outputs prediction results while generating power coupling disturbance prediction labels, which are directly mapped to the coupling coefficient update trigger logic of the feedforward decoupling matrix.

3. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 2, characterized in that, The least squares iterative algorithm dynamically updates the coupling coefficient matrix using a regularized recursive least squares format. Cross-coupling feedforward compensation terms and error correction factors are set in the inner-loop controller of each port current. The error correction factor adopts a deviation-segmented adaptive adjustment rule, dividing power and voltage deviations into three levels: small deviation, medium deviation, and large deviation, each corresponding to a different correction coefficient. An upper limit threshold for the correction coefficient is set to prevent over-correction that could cause system oscillations. The coupling disturbance suppression stage is used to suppress residual coupling disturbances. The system is configured with a decoupling effect closed-loop verification unit, which collects real-time data on the cross-correlation degree and fluctuation deviation of power and voltage at each port before and after decoupling. The current decoupling accuracy level is determined by quantitatively calculating the decoupling residual interference rate. When the decoupling residual rate exceeds a preset safety tolerance value, an expedited iterative update process for the coupling coefficient matrix is ​​automatically triggered.

4. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 1, characterized in that, In the second-level closed loop, the droop coefficient of each port is dynamically adjusted based on the rated capacity of each port converter, real-time operating status, and system power demand, in conjunction with the port operation status assessment module and the load priority weight coefficient. The port operation status assessment module performs real-time assessment from multiple dimensions such as converter operating temperature, switching frequency, output current margin, port insulation level, and power transmission efficiency, and quantifies and generates port health and load-bearing capacity scores. The score results directly serve as the core basis for adjusting the droop coefficient.

5. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 4, characterized in that, In the adaptive droop control strategy, voltage-active power adaptive droop characteristics are adopted for active power control, and the droop slope is dynamically adjusted in combination with load priority; frequency-reactive power adaptive droop characteristics are adopted for reactive power control, and a reactive power allocation error feedback mechanism is set up to calibrate the droop coefficient in real time. Both types of droop characteristics adopt a segmented adaptive adjustment mechanism, which divides the system operating load level into three intervals: light load, rated load, and heavy load, and adopts gentle, standard, and steep droop slopes respectively, with smooth transitions between different intervals without abrupt changes.

6. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 5, characterized in that, In the virtual impedance compensation stage, the line impedance is detected in real time by using a synchronous sampling and calculation method of port voltage and output current. The equivalent impedance value of the line is calculated based on the voltage-current correlation characteristics under steady-state operation. The dynamic virtual impedance and adaptive droop coefficient adopt a time-series coordinated adjustment rule. When a power distribution deviation is detected, the virtual impedance fine-tuning is initiated first to quickly eliminate the basic deviation caused by the line impedance. After the virtual impedance parameters stabilize, the droop coefficient is adjusted in conjunction to achieve precise distribution. The two adjustment processes cooperate with each other without interfering with each other.

7. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 1, characterized in that, In the AC / DC bus voltage coordinated control architecture, the bus voltage prediction control module adjusts control parameters in advance based on source-load fluctuation prediction data. The bus voltage prediction control module takes the real-time AC / DC bus voltage, source-load prediction power, and port operating status as inputs to predict the bus voltage change trend over multiple control cycles, sets a safe and stable voltage range, and initiates adjustment actions in advance when the predicted voltage is about to exceed the limit. In grid-connected mode, the medium-voltage AC port converter simultaneously provides dynamic reactive power support to the distribution network and links with the low-voltage side bus voltage controller to achieve coordinated AC / DC bus voltage regulation. A bidirectional mutual feedback compensation mechanism for AC / DC bus voltage is constructed. When the medium-voltage DC bus voltage deviation exceeds a preset threshold, the droop coefficient of the low-voltage AC / DC port is adjusted synchronously. When the low-voltage side bus voltage fluctuation exceeds the threshold, the feedforward compensation amount of the medium-voltage AC port converter is corrected in the opposite direction. In islanded mode, after the energy storage port converter takes over the control of the medium-voltage DC bus voltage, it simultaneously optimizes the control parameters of the low-voltage AC / DC bus voltage.

8. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 1, characterized in that, The multi-parameter fusion state machine uses distribution network voltage amplitude, frequency stability, phase deviation, fault signal, energy storage SOC state, and microgrid power balance state as core judgment parameters, and obtains a comprehensive judgment value for mode switching through weighted fusion calculation; the early warning signal for predictive mode switching comes from three key characteristic quantities of distribution network voltage drop rate, frequency drift rate, and voltage harmonic distortion rate. The linear trend fitting algorithm is used to extrapolate the changing trend of distribution network operating status in real time. When the characteristic quantity is detected to be continuously deteriorating and about to trigger the fault threshold, the switching preparation is started 1-2 system control cycles in advance. The converter control mode switching is accompanied by a hardware linkage execution mechanism. When the software issues the mode switching command, it simultaneously triggers the converter drive signal latching, power switch device status verification, grid-connected switch auxiliary contact monitoring, and DC side capacitor voltage clamping hardware action. When the distribution network returns to normal, pre-synchronization control is executed. The voltage amplitude, frequency and phase of the microgrid side are calibrated through the phase-locked loop, and the grid-connected switch is closed after matching.

9. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 1, characterized in that, The fault diagnosis module uses wavelet transform algorithm to identify fault type, severity, and location in real time. Based on fault characteristics, impact range, and duration, it classifies system operation faults into four levels: minor transient disturbance, moderate voltage drop, severe permanent fault, and internal port fault. When a temporary voltage drop fault occurs in the distribution network, the medium-voltage AC port converter switches to low-voltage ride-through mode to inject reactive current into the distribution network to support the grid voltage, while limiting the output current to within 1.5 times the rated value. When a permanent fault occurs in the distribution network, the grid-connected switch is disconnected and switched to islanded operation mode, and the critical load priority protection mechanism is activated. When a fault occurs at an internal port of a solid-state transformer, the corresponding port converter drive signal is blocked. The internal port fault diagnosis focuses on collecting two core insulation safety parameters: port insulation resistance and leakage current. It is combined with multiple parameters linked to determine the port over-temperature, over-current, and over-voltage electrical signals. A single parameter abnormality only triggers an early warning, while multiple parameters exceeding the limit confirm the fault. The load on a faulty port is transferred to other normal ports through intelligent port reconfiguration control. The reconfiguration principle is to prioritize load capacity, power supply distance, and minimize losses. The load transfer process on the faulty port adopts a power smoothing strategy.

10. The multi-port solid-state transformer collaborative control method for AC / DC hybrid microgrids according to claim 1, characterized in that, In the third-level closed loop, the DDPG reinforcement learning algorithm is composed of a state perception module, an action decision module, a reward evaluation module, and an experience playback module connected in sequence. The state perception module takes as input the power of each port, AC / DC bus voltage, load priority, fault status, and source-load fluctuation prediction value. The action decision module outputs the power reference value of each port and converter control parameters. The reward evaluation module uses power distribution balance, voltage stability margin, and system operating loss as multi-objective reward functions. The experience playback module has a built-in intelligent sample screening and deduplication optimization mechanism, which automatically removes invalid random fluctuation samples, duplicate redundant samples, and sensor abnormal error samples, and retains only high-quality samples containing effective decision information on typical power fluctuations, mode switching, fault ride-through, and load switching. The weights of each indicator in the multi-objective reward function are dynamically adapted according to the system operating mode. In grid-connected mode, the weights of operating loss and voltage stability are increased; in islanded mode, the weights of power distribution balance and load protection are increased; and in fault ride-through mode, the weights of voltage stability and equipment safety are strengthened. The scheduling optimization evaluation module uses power scheduling efficiency, voltage stability level, operating loss, and load power supply guarantee rate as core evaluation indicators to complete the quantitative evaluation of scheduling effect on a fixed period.

Images