Cooperative control system for access of SST direct-hanging overcharge host to microgrid

Abstract

The invention, which relates to the technical field of micro-grid energy management, discloses a cooperative control system for access of an SST directly-hung overcharge host to a micro-grid, comprising a multi-time scale cooperative control architecture, an AI prediction-optimization module, plug-and-play modular hardware, a bidirectional vehicle-network interaction cooperative control unit, and a dynamic weight optimization and digital twinning operation and maintenance platform. According to the method, the SST port power instruction is dynamically generated, the self-adaptive capacity and the overall operation economy of the system for dealing with the uncertainty of the source and load double sides are remarkably improved, the priority of the optimization target can be automatically adjusted according to the scene through dynamic weight optimization and a digital twin operation and maintenance platform, and the optimization efficiency is improved. According to the method, a control strategy is optimized, a 1: 1 digital twinborn model is utilized to carry out real-time safety verification and fault pre-judgment on the control strategy, the reliability, safety and operation and maintenance intelligence level of the system are greatly enhanced, and in addition, through standardized plug-and-play hardware and a V2G cooperation unit, interaction support of on-site consumption of renewable energy sources and a power grid is further promoted.

Description

Technical Field

[0001] This invention relates to the field of microgrid energy management technology, and in particular to a collaborative control system for SST direct-connected supercharging host access to a microgrid. Background Technology

[0002] A microgrid is a miniaturized, autonomous power system centered around distributed generation (DG), energy storage systems (ESS), and controllable loads, connected through power electronic devices. It has the dual capability of operating independently (island mode) or connected to the main grid (grid mode). It aims to achieve reliable power supply, efficient consumption of renewable energy, and flexible interaction in local areas, playing a key role in promoting energy transition (such as "dual carbon" goals) and improving power supply resilience.

[0003] SST (Solid State Transformer), also known as Power Electronic Transformer (PET), is a new type of transformer based on high-frequency power electronic conversion (such as IGBT / SiC devices). It replaces traditional power frequency transformers and has core capabilities such as flexible AC / DC conversion, multi-port power routing, high-precision voltage / frequency regulation, and proactive power quality management. The SST direct-connect supercharging host is an integrated intelligent charging device that uses a solid state transformer as its core energy conversion and control unit, directly connected to a microgrid / grid bus, to provide DC supercharging services for electric vehicles.

[0004] In existing technologies, the coordinated control of rapidly changing loads such as SST direct-connected superchargers in microgrids typically employs a hierarchical but loosely coupled time-scale strategy. Global optimization, real-time scheduling, and instantaneous support are disconnected, making it difficult to form a closed loop across second-, minute-, and hour-level dynamic responses. Furthermore, traditional methods rely on fixed rules or offline optimization, lacking the dynamic decision-making capabilities to integrate high-precision prediction and online self-learning, resulting in difficulties in simultaneously achieving multiple objectives (stability, economy, and low carbon emissions). In addition, digital twin technology in traditional methods is mostly used for post-event monitoring and fails to be deeply integrated into the real-time control loop for strategy pre-simulation and safety verification. Therefore, this invention proposes a coordinated control system for SST direct-connected superchargers connected to a microgrid to address the problems existing in the prior art. Summary of the Invention

[0005] To address the aforementioned issues, the present invention aims to propose a collaborative control system for SST direct-connected supercharging host access to microgrids, which solves the problems of traditional system control technologies such as multi-timescale control disconnection, rigid optimization strategies, lack of deep digital twin-enabled real-time control, and difficulty in balancing stability, economy, and low-carbon goals.

[0006] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a collaborative control system for SST direct-connected supercharging host access to a microgrid, comprising:

[0007] The multi-timescale collaborative control architecture includes an upper-layer hourly-level global optimization module, a middle-layer minute-level adaptive coordination module, and a lower-layer second-level rapid response module.

[0008] The AI ​​prediction-optimization module includes a long short-term memory network prediction unit and a deep reinforcement learning optimization unit.

[0009] Plug-and-play modular hardware, including silicon carbide-based modular multilevel converter-solid-state transformer topology, bidirectional CLLC resonant converter and standardized communication interface;

[0010] A two-way vehicle-to-grid interactive and collaborative control unit, integrating a V2G communication interface and a hierarchical battery allocation strategy;

[0011] Dynamic weight optimization and digital twin operation and maintenance platform, including a scenario-adaptive multi-objective weight adjustment module and a digital twin model;

[0012] The optimization results of the upper-level hour-level global optimization module provide a power reference for the middle-level minute-level adaptive coordination module. The real-time adjustment commands of the middle-level minute-level adaptive coordination module and the virtual inertial support commands of the lower-level second-level fast response module work together on the plug-and-play modular hardware.

[0013] The prediction results of the Long Short-Term Memory Network prediction unit are input into the upper-level hourly global optimization module and the deep reinforcement learning optimization unit. The deep reinforcement learning optimization unit generates and outputs the SST port power adjustment command to the middle-level minute-level adaptive coordination module based on the real-time data of the middle-level minute-level adaptive coordination module and the dynamic weights provided by the digital twin operation and maintenance platform.

[0014] Further improvements are made in that: the upper-level hourly global optimization module uses 24 hours as the prediction time domain, inputs the photovoltaic / wind power prediction curve, the supercharger reservation power curve, the time-of-use electricity price signal and the initial value of the energy storage system's state of charge, and outputs the energy storage system's charging and discharging plan, the power reference value of each port of the SST and the power plan for interaction between the microgrid and the main grid.

[0015] Further improvements are made in that the mid-level minute-level adaptive coordination module is used to monitor the deviation between the actual output of photovoltaic / wind power and the actual load of supercharging in real time, and dynamically adjust the power allocation of source-storage-charging based on the adjustment instructions output by the deep reinforcement learning optimization unit.

[0016] Further improvements are made in that the lower-level second-level fast response module is used to simulate the inertial characteristics of a synchronous generator. When a sudden increase in overcharging load is detected, causing frequency fluctuations, the SST output power is adjusted within milliseconds to provide virtual inertia support.

[0017] Further improvements are made in that: the Long Short-Term Memory Network (LSTM) prediction unit is used to process historical photovoltaic / wind power output data, supercharging order data, and weather forecast data, and outputs future photovoltaic / wind power output curves and supercharging load curves.

[0018] Further improvements are made in that: the deep reinforcement learning optimization unit adopts a deep deterministic policy gradient algorithm, its state space includes microgrid voltage, frequency, energy storage SOC, supercharging power and distributed power output, its action space is the adjustment amount of SST port power, and its reward function is adjusted by the dynamic weight optimization and the dynamic weight provided by the digital twin operation and maintenance platform.

[0019] Further improvements include: in the plug-and-play modular hardware, the modular multilevel converter submodule adopts a redundant design and supports hot-swapping, and the standardized communication interface is based on the OPC UA unified architecture and supports multiple industrial protocols.

[0020] A further improvement is that the bidirectional vehicle-to-grid interactive cooperative control unit communicates with the electric vehicle via the ISO 15118 protocol and executes a hierarchical calling strategy to call on-board battery energy according to the battery status, while also integrating bidirectional metering function.

[0021] Further improvements are made in that: the scenario adaptive multi-objective weight adjustment module identifies the operating scenario based on fuzzy logic, and dynamically adjusts the weights of stability, economy and green energy consumption in the reward function of the deep reinforcement learning optimization unit; the digital twin model is used to map the physical system state in real time to support control strategy verification and fault diagnosis.

[0022] The beneficial effects of this invention are as follows: This invention utilizes LSTM high-precision prediction and DRL online learning to dynamically generate SST port power commands, significantly improving the system's adaptive capability to cope with uncertainties on both the source and load sides and its overall operational economy. Furthermore, through dynamic weight optimization and a digital twin operation and maintenance platform, the priority of optimization targets can be automatically adjusted according to the scenario, and a 1:1 digital twin model is used to perform real-time safety verification and fault prediction of the control strategy, greatly enhancing the system's reliability, security, and intelligent operation and maintenance level. In addition, through standardized plug-and-play hardware and V2G collaborative units, the system's expansion flexibility is improved, and the mobile energy storage potential of electric vehicles is fully explored, further promoting the local consumption of renewable energy and the interactive support of the power grid. It achieves seamless connection and collaborative optimization from hourly economic planning and minute-level dynamic coordination to second-level inertial support. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the framework structure of the collaborative control system for SST direct-connected supercharging host connected to microgrid of the present invention. Detailed Implementation

[0024] 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.

[0025] It should be noted that the technical means not described in detail in the following embodiments are all conventional means in the field, are not the key points of the invention, and will not be elaborated upon.

[0026] Example 1

[0027] according to Figure 1 As shown, this embodiment provides a collaborative control system for SST direct-connected supercharging host access to a microgrid. This system consists of a multi-timescale collaborative control architecture, an AI prediction-optimization module, plug-and-play modular hardware, a two-way vehicle-to-grid (V2G) collaborative control unit, and a dynamic weight optimization and digital twin operation and maintenance platform. The multi-timescale collaborative control architecture is responsible for hierarchical power management at different time granularities. The AI ​​prediction-optimization module provides intelligent forward-looking prediction and real-time decision-making. The plug-and-play modular hardware is the physical foundation for achieving efficient and reliable energy conversion. The two-way vehicle-to-grid collaborative control unit taps into the mobile energy storage potential of electric vehicles. The dynamic weight optimization and digital twin operation and maintenance platform provides strategy optimization and a virtual verification environment. Specifically:

[0028] The multi-timescale collaborative control architecture includes an upper-level hourly global optimization module, a middle-level minute-level adaptive coordination module, and a lower-level second-level fast response module. The optimization results of the upper-level hourly global optimization module provide a power reference for the middle-level minute-level adaptive coordination module. The real-time adjustment commands of the middle-level minute-level adaptive coordination module and the virtual inertial support commands of the lower-level second-level fast response module work together on the plug-and-play modular hardware.

[0029] The AI ​​prediction-optimization module includes a Long Short-Term Memory (LSTM) prediction unit and a Deep Reinforcement Learning (DRL) optimization unit. The prediction results of the LSTM prediction unit are input into the upper-level hourly global optimization module and the deep reinforcement learning optimization unit. The deep reinforcement learning optimization unit generates and outputs SST port power adjustment instructions to the middle-level minute-level adaptive coordination module based on the real-time data and dynamic weight optimization of the mid-level minute-level adaptive coordination module and the dynamic weights provided by the digital twin operation and maintenance platform.

[0030] Plug-and-play modular hardware, including silicon carbide (SiC) based modular multilevel converter (MMC)-solid-state transformer (SST) topology, bidirectional CLLC resonant converter, and standardized communication interface;

[0031] A two-way vehicle-to-grid interactive and collaborative control unit, integrating a V2G communication interface and a hierarchical battery allocation strategy;

[0032] The dynamic weight optimization and digital twin operation and maintenance platform includes a scenario-adaptive multi-objective weight adjustment module and a digital twin model.

[0033] This embodiment forms a deeply coupled closed loop through the above five components. Specifically, AI prediction provides input for long-term planning, while AI optimization generates instructions based on real-time status and dynamic weights to dynamically adjust control behavior at medium and short time scales. This closed-loop link of prediction-planning-control-optimization achieves the organic unity of global goals and local responses.

[0034] In this embodiment, the upper-level hourly global optimization module uses a 24-hour prediction time domain, inputting photovoltaic / wind power prediction curves, supercharger reserved power curves, time-of-use electricity price signals, and initial values ​​of the energy storage system's state of charge (SOC). It outputs the energy storage system's charging and discharging plan, the power reference values ​​of each port of the SST, and the power interaction plan between the microgrid and the main grid. Based on the prediction data for the next 24 hours, this module performs mathematical optimization calculations with the core objectives of optimizing the economic efficiency of the whole day's operation and maximizing the consumption of renewable energy. Its output is a pre-defined "operation script," including when the energy storage battery should charge and discharge, the reference power values ​​that should be transmitted at each port of the SST, and the planned exchange power between the microgrid and the main grid. This sets an authoritative and cost-effective benchmark for the subsequent real-time adjustment of the entire system.

[0035] The aforementioned mathematical optimization calculation specifically involves constructing and solving a multi-objective mixed-integer linear programming (MILP) model with a 24-hour scheduling cycle and a 1-hour time step. The model's primary comprehensive objectives are minimizing the total system operating cost and maximizing renewable energy consumption. Operating costs mainly include electricity purchase costs from the main power grid and depreciation costs of energy storage equipment; renewable energy consumption is measured by the actual power utilization of photovoltaic and wind power.

[0036] The model's decision variables include: the energy storage system's charging / discharging power at each time point, the transmission power at each port of the SST, and the interaction power between the microgrid and the main grid;

[0037] The core constraints of the model include:

[0038] Power balance constraint: The sum of the output of distributed generation, energy storage discharge, power purchase from the main grid, and V2G discharge power in a microgrid must be equal to the sum of overcharging load, energy storage charging, and power sold to the main grid.

[0039] Energy storage system operation constraints include upper and lower limits of energy storage state of charge (SOC), charging and discharging power limits, and constraints on the equality or set value of SOC at the beginning and end of the dispatch cycle, in order to ensure the sustainable operation of energy storage.

[0040] Equipment capacity constraints: The transmission power of each port of the SST and the power exchanged with the main power grid must not exceed its rated capacity;

[0041] Grid interaction constraints: Based on the grid connection agreement, set an upper limit on the power exchanged with the main grid;

[0042] Network operation safety constraints: Ensure that the voltage of critical nodes in the microgrid is maintained within the allowable range;

[0043] In solving the problem, scalarization methods such as linear weighting or ε-constraint are employed to transform the two objectives of economic efficiency and green energy consumption into a single-objective problem. The final output is a time-series energy storage charging and discharging plan for the next 24 hours, the SST port power baseline value, and the grid-connected power plan, which serve as the benchmark for subsequent minute-level real-time coordination.

[0044] In this embodiment, the mid-level minute-level adaptive coordination module is used to monitor the deviation between the actual output of photovoltaic / wind power and the actual load of supercharging in real time, and dynamically adjust the power allocation of source-storage-charging based on the adjustment instructions output by the deep reinforcement learning optimization unit. This module is responsible for monitoring the deviation between actual operation and upper-level plan (such as a sudden decrease in photovoltaic output or a surge in charging demand). It receives real-time adjustment instructions from the deep reinforcement learning optimization unit and quickly redistributes the instantaneous power flow between photovoltaic, energy storage and charging piles on a minute-level time scale. Its core strategy is to prioritize the use of readily available renewable energy (such as excess photovoltaic) to directly supply power, and then coordinate energy storage or V2G to supplement the insufficient part, thereby achieving minute-level rebalancing of system power.

[0045] In this embodiment, the lower-level second-level fast response module is used to simulate the inertial characteristics of a synchronous generator. When a sudden increase in overcharge load causes frequency fluctuations, it adjusts the output power of the SST within milliseconds to provide virtual inertia support. This module provides crucial frequency support for the power-electronic microgrid by simulating the physical inertial characteristics of a traditional synchronous generator (virtual synchronous machine technology). When the high-power connection of the overcharger causes a sudden change in the grid frequency, this module can quickly adjust the output power of the SST within 10 milliseconds, acting like a "damper" to rapidly suppress frequency fluctuations and ensure the instantaneous safe and stable operation of the microgrid.

[0046] In this embodiment, the Long Short-Term Memory (LSTM) network prediction unit is used to process historical photovoltaic / wind power output data, supercharging order data, and weather forecast data, and outputs future photovoltaic / wind power output curves and supercharging load curves. This unit utilizes an LSTM neural network that can handle long-term dependencies in time-series data, integrates historical operating data (past 72 hours) and future environmental information (weather forecasts), and generates high-precision renewable energy output predictions and electric vehicle charging load predictions. Its output 15-minute to 24-hour prediction curves are the fundamental basis for scientific decision-making in upper-level global optimization and real-time optimization.

[0047] In this embodiment, the deep reinforcement learning optimization unit adopts a deep deterministic policy gradient algorithm. Its state space includes microgrid voltage, frequency, energy storage SOC, supercharging power, and distributed power output. Its action space is the adjustment amount of SST port power. Its reward function is adjusted by dynamic weights provided by the dynamic weight optimization and digital twin operation and maintenance platform. This dynamic weight adjustment process is implemented through the following steps: The scenario adaptive multi-objective weight adjustment module in the dynamic weight optimization and digital twin operation and maintenance platform first analyzes the key operating indicators of the microgrid in real time based on fuzzy logic rules (such as frequency deviation, voltage limit exceedance, energy storage SOC, real-time renewable energy penetration rate, and electricity price signal), and classifies the system operating status into preset typical scenarios (such as emergency stable state, low-carbon priority state, or economic optimal state). Subsequently, based on the identified scenario, the module dynamically outputs a set of normalized weight coefficients (corresponding to stability, economy, and green electricity consumption targets, respectively) through the built-in fuzzy inference engine and defuzzification calculation. These real-time generated weight coefficients are directly injected into the reward function calculation of the deep reinforcement learning (DRL) optimization unit, specifically manifested as: reward function R t =w s *f s (Stability index) +w e *f e (Economic standard) + w g *f g (Green electricity consumption index), where the weight w s w e w g These are the parameters that the platform dynamically adjusts. Therefore, the optimization objective of the DRL agent will adaptively shift its focus according to the changing scenario; for example, in an emergency... s The level was significantly increased to prioritize stability, under a low-carbon priority environment. g The energy level is increased to maximize the absorption of green electricity, thereby achieving a dynamic optimal balance among multiple objectives.

[0048] In this embodiment, the MMC submodule in the plug-and-play modular hardware adopts a redundant design and supports hot-swapping. The communication interface is based on the OPC UA unified architecture and supports multiple industrial protocols. The physical layer of the plug-and-play modular hardware adopts the advanced SiC MMC-SST topology, which takes into account the high efficiency and high reliability of the high-voltage side and the adaptability of the low-voltage side to a wide range of electric vehicle batteries. The hot-swapping and redundancy design ensures that the system can run uninterrupted when a single submodule fails, greatly improving availability. The unified communication interface based on OPC UA solves the problem of interconnection between multiple devices and multiple manufacturers, and provides a standardized and high real-time data channel for the precise control of physical devices by upper-level high-level algorithms.

[0049] In this embodiment, the bidirectional vehicle-to-grid (V2G) interactive cooperative control unit communicates with the electric vehicle via the ISO 15118 protocol and executes a hierarchical calling strategy to call on-board battery energy according to the battery status. It also integrates bidirectional metering function. This unit acts as a "gateway" for interaction with the electric vehicle. It securely communicates with the vehicle battery management system (BMS) through the international standard protocol (ISO 15118) to obtain the core battery status and ensures that its energy is only called when the vehicle battery status allows it and is beneficial to the power grid. At the same time, the integrated bidirectional metering and incentive mechanism is a necessary technical means to promote user participation in V2G and realize a commercial closed loop.

[0050] In this embodiment, the scenario-adaptive multi-objective weight adjustment module identifies the operating scenario based on fuzzy logic and dynamically adjusts the weights of stability, economy and green electricity consumption in the reward function of the deep reinforcement learning optimization unit. It can automatically increase the influence of the corresponding objectives (such as stability and green electricity consumption rate) in DRL decision-making according to whether the system is in an emergency fault state, a low-carbon priority mode or normal economic operation.

[0051] Digital twin models are used to map the state of physical systems in real time. They can be used not only for daily monitoring and fault prediction, but more importantly, they provide a safe testing ground for the decision-making of deep reinforcement learning optimization units. Any new control strategy or optimization instruction can be simulated and tested in the digital twin model first. Only after confirming that it is safe and effective can it be sent to the real device, thereby greatly reducing the risks of online learning and optimization processes.

[0052] Example 2

[0053] This embodiment provides a collaborative control method for a collaborative control system of an SST direct-connected supercharging host connected to a microgrid, including the following steps:

[0054] S1. Multi-timescale forecasting and planning

[0055] Using a Long Short-Term Memory (LSTM) network model, based on historical data and external environmental information, the photovoltaic / wind power output and supercharging load for future periods are predicted. Based on the prediction results, global optimization is performed on an hourly time scale to generate a charging and discharging plan for the energy storage system, power reference values ​​for each port of the solid-state transformer (SST), and a microgrid grid-connection interaction plan.

[0056] S2, Minute-level Adaptive Coordination and Dynamic Optimization

[0057] On a minute-level timescale, the actual output of distributed power sources and the actual load of supercharging are monitored in real time and compared with the planned values ​​generated in step S1 to generate power deviations. Then, the deep reinforcement learning (DRL) algorithm is used in combination with dynamically adjusted multi-objective weights to optimize the source-storage-charging power allocation strategy online, generate and issue SST port power adjustment commands, and achieve minute-level power rebalancing.

[0058] S3, second-level rapid response and support

[0059] Using a time scale of seconds or less, virtual inertia support is provided for the microgrid by simulating the inertial characteristics of a synchronous generator. When frequency fluctuations caused by sudden changes in overcharging load are detected, the output power of the SST is adjusted within milliseconds based on the instructions or preset rules of the DRL algorithm to suppress frequency deviation.

[0060] S4, Vehicle-to-Network Interaction and System Maintenance

[0061] The system communicates with connected electric vehicles via standard protocols, executes a hierarchical battery allocation strategy based on battery status and microgrid requirements, and enables bidirectional energy flow and metering between the vehicle and the grid. Simultaneously, based on the constructed 1:1 digital twin model of the physical system, it synchronizes operational data for offline verification of control strategies, online fault prediction and diagnosis, and provides dynamic weight adjustment and decision safety verification for the DRL algorithm in step S2.

[0062] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A collaborative control system for SST direct-connected supercharging host to be connected to a microgrid, characterized in that, include: The multi-timescale collaborative control architecture includes an upper-layer hourly-level global optimization module, a middle-layer minute-level adaptive coordination module, and a lower-layer second-level rapid response module. The AI ​​prediction-optimization module includes a long short-term memory network prediction unit and a deep reinforcement learning optimization unit. The prediction results of the long short-term memory network prediction unit are input into the upper-level hourly global optimization module and the deep reinforcement learning optimization unit. The deep reinforcement learning optimization unit generates and outputs SST port power adjustment instructions to the middle-level minute-level adaptive coordination module based on real-time data and dynamic weights. Plug-and-play modular hardware, including silicon carbide-based modular multilevel converter-solid-state transformer topology, bidirectional CLLC resonant converter and standardized communication interface; A two-way vehicle-to-grid interactive and collaborative control unit, integrating a V2G communication interface and a hierarchical battery allocation strategy; Dynamic weight optimization and digital twin operation and maintenance platform, including a scenario-adaptive multi-objective weight adjustment module and a digital twin model; The optimization results of the upper-level hour-level global optimization module provide a power reference for the middle-level minute-level adaptive coordination module. The control commands of the middle-level minute-level adaptive coordination module and the lower-level second-level fast response module work together on the plug-and-play modular hardware.

2. The collaborative control system for SST direct-connected supercharging host access to microgrid according to claim 1, characterized in that: The upper-level hourly global optimization module outputs the energy storage system charging and discharging plan, SST port power reference values, and microgrid-to-grid interaction power plan based on the photovoltaic / wind power prediction curve, the supercharger reserved power curve, the time-of-use electricity price signal, and the initial value of the energy storage system's state of charge.

3. The collaborative control system for SST direct-connected supercharging host access to microgrid according to claim 1, characterized in that: The mid-level minute-level adaptive coordination module dynamically adjusts the power allocation between the source, storage, and charging based on the deviation between the actual output of photovoltaic / wind power and the actual load of supercharging, as well as the adjustment instructions output by the deep reinforcement learning optimization unit.

4. The collaborative control system for SST direct-connected supercharging host access to microgrid according to claim 1, characterized in that: The lower-level second-level fast response module simulates the inertial characteristics of a synchronous generator and adjusts the SST output power in a timely manner to provide virtual inertia support when the overcharge load changes abruptly.

5. The collaborative control system for SST direct-connected supercharging host to microgrid as described in claim 1, characterized in that: The Long Short-Term Memory Network (LSTM) prediction unit is used to predict future photovoltaic / wind power output and supercharging load based on historical and meteorological data.

6. The collaborative control system for SST direct-connected supercharging host access to microgrid according to claim 1, characterized in that: The state space of the deep reinforcement learning optimization unit includes microgrid voltage, frequency, energy storage SOC, supercharging power, and distributed power output. Its action space is the SST port power adjustment amount, and the weight of its reward function is dynamically adjusted by the digital twin operation and maintenance platform.

7. The collaborative control system for SST direct-connected supercharging host access to microgrid according to claim 1, characterized in that: The modular multilevel converter submodule in the plug-and-play modular hardware adopts a redundant hot-swappable design, and the standardized communication interface is based on the OPC UA architecture.

8. The collaborative control system for SST direct-connected supercharging host access to microgrid according to claim 1, characterized in that: The bidirectional vehicle-to-grid interactive control unit is used to communicate with electric vehicles and perform hierarchical allocation and bidirectional metering of battery energy.

9. A collaborative control system for SST direct-connected supercharging host access to a microgrid according to claim 1, characterized in that: The dynamic weight optimization and digital twin operation and maintenance platform is used to dynamically adjust the target weights and to perform strategy verification and system state mapping through a digital twin model.

Images