Flexible power distribution ac-dc networking device and method of use thereof

Abstract

This invention discloses a flexible AC / DC power distribution network device and its usage method. The device includes a multi-port adaptive energy router, a distributed AI control network, a multi-modal energy storage matrix, and a multi-modal communication and digital twin platform. The energy router employs wide-bandgap semiconductors and dynamic topology switching technology to achieve efficient AC / DC energy conversion; the distributed AI network dynamically allocates power and provides early warnings of faults through multi-objective optimization and fault prediction models; the energy storage matrix combines hierarchical scheduling and lifetime management to meet diverse load demands. The usage method includes system initialization, model building, dynamic networking, energy optimization, fault response, and system expansion steps, achieving intelligent control through mathematical models and federated learning. This invention solves the problems of low efficiency, slow response, and poor reliability in traditional power distribution systems, improving energy utilization, system stability, and economic benefits, and is applicable to various scenarios such as smart grids.

Description

Technical Field

[0001] This invention relates to the field of power grid technology, specifically to a flexible AC / DC power distribution network device and its usage method. Background Technology

[0002] In existing AC / DC hybrid power distribution systems, traditional power converters and centralized control architectures are typically used for energy transfer and management. However, these systems have significant limitations when handling complex loads and distributed power sources. First, traditional power converters often employ fixed topologies, making it difficult to dynamically adapt to changes in AC / DC loads, resulting in low energy conversion efficiency, especially in scenarios with multiple power sources and multiple loads, where conversion losses are significant. Furthermore, centralized control systems rely on preset rules and lack real-time response capabilities to dynamic loads and power fluctuations, making it difficult to achieve comprehensive optimization of efficiency, cost, and stability. This limits the application of these systems in highly dynamic scenarios such as data centers and electric vehicle charging stations.

[0003] Furthermore, existing power distribution equipment has shortcomings in energy storage management and fault prediction. Traditional energy storage systems typically use a single type of energy storage unit, which makes it difficult to simultaneously meet the requirements of transient high power demand and long-term stable power supply. Moreover, the lack of an effective lifespan management mechanism leads to accelerated aging of energy storage units and increased system operating costs. At the same time, existing fault detection methods mostly rely on threshold triggering, lacking predictability and failing to identify potential faults in advance, easily leading to unexpected downtime and decreased system reliability.

[0004] Furthermore, existing systems have limited communication and scalability capabilities, making it difficult to support distributed energy trading or modular upgrades, thus limiting their application potential in future smart grids. Current technologies also lack the ability to simulate and optimize system operation in real time, and traditional control strategies are difficult to dynamically adjust based on real-time electricity prices, load priorities, or power availability, reducing the system's economic efficiency and environmental friendliness. Summary of the Invention

[0005] The purpose of this invention is to provide a device and method for efficiently, flexibly and intelligently realizing AC / DC networking, so as to overcome the above-mentioned technical problems and improve the adaptability and reliability of the system.

[0006] To achieve the above objectives, the present invention proposes the following technical solution: a flexible AC / DC power distribution network device, comprising: a multi-port adaptive energy router, a distributed AI control network, a multi-modal energy storage matrix, and a multi-modal communication and digital twin platform;

[0007] The multi-port adaptive energy router adopts a gallium nitride-based multiphase interleaved parallel topology and integrates a dynamic impedance matching module and a hybrid modulation unit. The dynamic impedance matching module is used to detect and match the impedance characteristics of the power supply side and the load side in real time. The hybrid modulation unit is used to combine space vector modulation (SVM) and pulse width modulation (PWM) to support bidirectional AC and DC energy flow and can dynamically switch the circuit topology according to the instructions of the distributed AI control network to adapt to different loads and power requirements.

[0008] The distributed AI control network includes multiple locally deployed edge computing nodes, each equipped with a neural network processor (NPU), which respectively run an adaptive multi-objective optimization algorithm and a fault prediction model based on a long short-term memory network, and realizes local updates and global synchronization of model parameters through a federated learning mechanism;

[0009] The multimodal energy storage matrix includes lithium batteries, supercapacitors, and flywheel energy storage modules. Each energy storage module integrates an active thermal management module and an adaptive lifespan management unit. The active thermal management module maintains the temperature stability of the energy storage unit, and the adaptive lifespan management unit optimizes the charging and discharging strategy by monitoring the health status of each module in real time. The multimodal energy storage matrix allocates energy according to power allocation instructions containing a hierarchical energy scheduling strategy issued by a distributed AI control network.

[0010] The multimodal communication and digital twin platform supports 5G or 6G, power line carrier and low power wide area network communication protocols, and integrates a blockchain-based energy trading interface and digital twin model to realize real-time simulation of system operation status, strategy optimization and distributed energy peer-to-peer trading and optimization of system operation status.

[0011] The multi-port adaptive energy router, distributed AI control network, multimodal energy storage matrix, and multimodal communication and digital twin platform work together through preset communication protocols and data interfaces to achieve adaptive energy management, multi-objective optimization, and predictive maintenance of AC / DC hybrid power distribution systems.

[0012] Furthermore, the hybrid modulation unit of the multi-port adaptive energy router combines space vector modulation (SVM) and pulse width modulation (PWM), and adjusts the operating point in real time through a dynamic impedance matching module.

[0013] Furthermore, the objective function of the adaptive multi-objective optimization algorithm for the distributed AI control network is: ;

[0014] in, For the system energy conversion efficiency, P out and P in These are the input and output active power, respectively. The total operating cost reflects the economic cost incurred by the system due to electricity purchases at different time periods. t is the discrete-time index, used to represent different operating times or scheduling cycles. C t Let P be the unit electricity price or electricity cost coefficient at time t. grid,t Let be the active power purchased from the public grid at time t; Voltage stability is used to measure the degree to which the system voltage deviates from the rated value. ΔV is the deviation between the actual operating voltage and the rated voltage. nom The system rated voltage is given by 1−S, which represents the degree of voltage instability and is minimized in the objective function to improve power supply quality. w1, w2, and w3 are weighting coefficients that satisfy w1+w2+w3=1 and are dynamically adjusted according to the operational objectives. Constraints include: ,in These are power supply capacity, energy storage capacity, and load capacity, respectively. For real-time voltage and current, Within a safe range.

[0015] Furthermore, the fault prediction model of the distributed AI control network is based on a long short-term memory network, and the fault probability calculation formula is as follows: Where σ is the Sigmoid activation function, For model parameters, These are real-time voltage, current, equipment temperature, and energy storage health status, respectively. The fault prediction model updates its parameters periodically through a federated learning mechanism. Each edge node calculates the model gradient based on local data, uploads it to the central aggregation server after homomorphic encryption, and then the server executes the federated averaging algorithm to update the global model parameters before distributing them to each node.

[0016] Furthermore, the hierarchical energy scheduling strategy is as follows: the supercapacitor energy storage module handles transient loads, the lithium battery energy storage module supports continuous loads, and the flywheel energy storage module is used for high-power buffering; the adaptive life management unit monitors the health status of each module in real time and dynamically adjusts its charging and discharging current upper limit and depth based on the health status value.

[0017] Furthermore, in the multimodal communication and digital twin platform, the 5G or 6G communication module is used for low-latency control command transmission between distributed AI control nodes, the power line carrier PLC module is used for data transmission between fixed devices, and the low-power wide area network module is used to connect distributed sensing devices.

[0018] The blockchain energy trading interface is implemented based on smart contracts, supporting automatic peer-to-peer energy trading and settlement between distributed power sources and loads;

[0019] The digital twin model updates the virtual system state by receiving real-time sensor data and outputs optimized control parameters to the distributed AI control network based on simulation results.

[0020] The above-described method of using a flexible AC / DC power distribution network includes the following steps:

[0021] Step S1: System initialization and data acquisition. Connect the equipment to the power distribution network, connect the distributed power source and the load, collect real-time data through sensors, including voltage, current, load power, power supply power, electricity price and energy storage health status, and set the initial operation target through multimodal communication and digital twin platform.

[0022] Step S2: Model building and updating. The distributed AI control network builds a multi-objective optimization model and a fault prediction model based on the collected data, and updates the model parameters and global synchronization regularly through a federated learning mechanism.

[0023] Step S3: Dynamic networking and topology optimization. The multi-port energy router dynamically switches its circuit topology and adjusts impedance matching parameters according to the power allocation instructions output by the multi-objective optimization model (which are generated based on the energy multi-objective optimization model). At the same time, the distributed AI nodes collaboratively calculate and allocate the output power of each power source and energy storage. For the allocation of the energy storage matrix, transient, continuous and high-power tasks are assigned to the supercapacitor, lithium battery and flywheel module respectively according to the hierarchical energy scheduling strategy.

[0024] Step S4: Based on the solution results of the multi-objective optimization model, prioritize the scheduling of clean energy power supply; at the same time, according to the hierarchical energy scheduling strategy, allocate the transient fluctuation task of the load to the supercapacitor module, the continuous power supply task to the lithium battery module, and the high-power buffer task to the flywheel energy storage module.

[0025] Step S5: Fault prediction and response. Based on the fault probability output by the fault prediction model, when the fault probability is greater than 0.8, an early warning is triggered and the system switches to standby mode. At the same time, the security of the switching strategy is verified through simulation using multimodal communication and a digital twin platform.

[0026] Step S6: System expansion and continuous optimization. Energy storage and ports are expanded through hot-swappable modules. The distributed AI control network automatically identifies newly connected devices and updates and optimizes the model. Peer-to-peer trading of surplus electricity is realized through the blockchain energy trading interface, and the model parameters are continuously optimized by combining long-term operating data.

[0027] Furthermore, in step S2, the parameter synchronization process of the federated learning mechanism includes:

[0028] S2.1 Each edge node calculates the model gradient based on local data and encrypts the gradient using the Paillier homomorphic encryption algorithm;

[0029] S2.2 The encrypted gradient is uploaded to the central aggregation server through a secure channel;

[0030] S2.3 The server aggregates and calculates the encryption gradients, and executes the federated averaging algorithm to update the global model parameters;

[0031] S2.4. Distribute the updated global model parameters to each edge node to complete one round of model update.

[0032] Furthermore, in step S4, the energy storage matrix allocates tasks according to the load characteristics, the supercapacitors handle transient loads, the lithium batteries support continuous loads, the flywheel energy storage is used for high-power buffering, and the voltage fluctuation is controlled within ±0.5%.

[0033] Furthermore, in step S5, the digital twin platform updates the system virtual model in real time, verifies the optimization strategy and fault response effect, and records fault data through the blockchain interface when a fault occurs to ensure traceability.

[0034] Furthermore, in step S6, the excess electricity is traded peer-to-peer through the blockchain energy trading interface, and low-cost power is prioritized for allocation using a multi-objective optimization model.

[0035] Beneficial effects: The technical solution of this application has the following technical effects:

[0036] 1. Synergistic Mechanism of Multi-Objective Optimization, Energy Flow Model, and Hierarchical Scheduling: This invention utilizes a distributed AI control network to run a multi-objective optimization algorithm. This algorithm, with a multi-objective optimization model at its core, comprehensively considers efficiency, cost, and voltage stability to generate a globally optimal power allocation command. Within this command, the power allocation for the multi-modal energy storage matrix is ​​further refined and executed by a hierarchical energy scheduling strategy, ensuring that supercapacitors, lithium batteries, and flywheels each perform their respective functions, meeting dynamic load demands while extending energy storage lifespan. The organic combination of these three components enables the system to achieve efficient, stable, and economical operation under complex conditions.

[0037] 2. This invention significantly improves the performance of AC / DC hybrid power distribution systems through the collaborative design of a multi-port adaptive energy router, a distributed AI control network, a multimodal energy storage matrix, and a multimodal communication and digital twin platform. First, the multi-port adaptive energy router employs wide-bandgap semiconductors and dynamic topology switching technology, enabling it to flexibly adjust energy flow according to load and power supply characteristics, thus solving the problem of low efficiency in traditional fixed-topology converters. The distributed AI control network achieves a dynamic balance between efficiency, cost, and stability through multi-objective optimization algorithms and fault prediction models, overcoming the limitations of slow response in centralized control systems, and is particularly suitable for high-dynamic load scenarios.

[0038] 3. Furthermore, the multimodal energy storage matrix, combined with hierarchical scheduling and adaptive lifetime management technologies, effectively resolves the contradiction between high power and long lifespan in traditional energy storage systems. Energy storage units intelligently allocate tasks based on load characteristics, ensuring both transient and continuous power supply needs while extending energy storage lifespan and reducing long-term operating costs. Fault prediction models identify potential risks in advance through real-time data analysis, significantly improving system reliability, reducing unexpected downtime, and compensating for the shortcomings of traditional threshold detection methods.

[0039] 4. This invention also achieves real-time simulation and optimization of system operation status through multimodal communication and a digital twin platform, supporting distributed energy trading and modular expansion. The blockchain energy trading interface promotes the efficient utilization of clean energy, improving economic efficiency and environmental friendliness. The digital twin platform optimizes operating strategies through virtual simulation, reducing maintenance costs and enhancing system scalability. These combined advantages enable this invention to adapt to the diverse needs of future smart grids, providing a novel solution for efficient and reliable AC / DC networking.

[0040] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below can be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other.

[0041] The foregoing and other aspects, embodiments, and features of the teachings of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the invention, such as features and / or beneficial effects of exemplary embodiments, will become apparent from the following description or may be learned through practice of specific embodiments according to the teachings of the present invention. Attached Figure Description

[0042] The accompanying drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in the various figures may be denoted by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the invention will now be described by way of example and with reference to the accompanying drawings, wherein:

[0043] Figure 1 This is a schematic diagram of the system architecture of the present invention.

[0044] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0045] To better understand the technical content of this invention, specific embodiments are described below in conjunction with the accompanying drawings. Various aspects of the invention are described in this disclosure with reference to the accompanying drawings, which illustrate numerous illustrative embodiments. The embodiments of this disclosure are not necessarily defined to include all aspects of the invention. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed in this invention are not limited to any particular implementation. Furthermore, some aspects of this invention can be used alone or in any suitable combination with other aspects of this invention.

[0046] This embodiment provides a flexible AC / DC power distribution network device and its usage method, which is applied to a smart industrial park and includes AC loads such as 380V three-phase motors, DC loads such as 48V LED lighting, and distributed power sources such as photovoltaics and energy storage.

[0047] The hardware composition of the multi-port adaptive energy router is as follows: This module adopts a multi-phase interleaved parallel topology based on gallium nitride (GaN), including 4 AC ports supporting 380V three-phase AC and 3 DC ports supporting 48V to 800V DC. It integrates a dynamic impedance matching module and a hybrid modulation unit, combining space vector modulation (SVM) and pulse width modulation (PWM).

[0048] High-efficiency energy conversion is achieved through the low switching loss characteristics of GaN devices; the dynamic impedance matching module detects the load and power supply impedance in real time and optimizes the operating point to reduce transmission loss; the hybrid modulation unit reduces electromagnetic interference and supports dynamic topology switching, such as full bridge and half bridge, through the coordinated control of SVM and PWM, to adapt to the dynamic changes of AC and DC loads.

[0049] The dynamic impedance matching module is used to adaptively adjust the equivalent impedance of the energy transmission channel according to the real-time operating status of the AC / DC hybrid power distribution system, so as to reduce energy transmission loss and improve system stability. This module includes at least an electrical parameter detection unit, an adjustable impedance execution unit, and an impedance matching control unit.

[0050] (1) Principle of electrical parameter detection

[0051] The electrical parameter detection unit is used to collect voltage and current parameters at key nodes of the system in real time. It may include a voltage sampling circuit, a current sampling circuit, and a signal conditioning circuit. By synchronously sampling the collected voltage and current signals, the instantaneous impedance or equivalent impedance change trend of the corresponding node is calculated, providing a basis for subsequent matching control. The detection unit can be installed at key load buses, AC / DC interfaces, or energy router ports.

[0052] (2) Adjustable impedance circuit structure

[0053] The adjustable impedance execution unit is constructed using controllable power electronic devices, and its specific form may be as follows:

[0054] Based on the equivalent impedance adjustment structure of DC / DC or AC / DC power converter, the equivalent input or output impedance of the port is changed by adjusting the duty cycle.

[0055] Alternatively, impedance adjustment branches based on controllable reactance and resistor networks can be used to continuously or progressively adjust the equivalent impedance of the system within a certain range.

[0056] The above structure achieves dynamic adjustment of the equivalent impedance of the energy transmission channel by changing the conduction state or control parameters of the power device;

[0057] (3) Impedance matching control method

[0058] The impedance matching control unit is used to generate impedance adjustment commands based on the voltage and current information output by the electrical parameter detection unit. Its control logic includes:

[0059] Based on the preset target impedance range or minimum loss criterion, determine whether the current system impedance deviates from the ideal operating state;

[0060] When an impedance mismatch trend is detected, the adjustable impedance actuator adjusts the control parameters of the power device so that the system's equivalent impedance gradually approaches the target value.

[0061] During load changes, power fluctuations, or topology switching, continuous closed-loop regulation is performed to suppress voltage fluctuations and power oscillations.

[0062] The control process can be implemented using proportional adjustment, segmented control, or table lookup methods, and the specific algorithm can be adjusted and selected by those skilled in the art as needed.

[0063] (4) Collaborative Work Description

[0064] The dynamic impedance matching module can work in conjunction with the energy router control unit and the energy storage scheduling unit to synchronously adjust the energy flow direction and impedance parameters when the system operating status changes, thereby maintaining better energy transmission efficiency and voltage stability under different operating conditions.

[0065] The hybrid modulation unit and the dynamic impedance matching module do not work independently, but form a cooperative control relationship within the multi-port adaptive energy router to achieve efficient transmission and stable control of AC and DC energy.

[0066] (1) Functional division of labor

[0067] The hybrid modulation unit is used to modulate and control the switching devices of the power converter. By combining space vector modulation (SVM) and pulse width modulation (PWM), it can achieve fine adjustment of the output voltage amplitude, phase and harmonic characteristics.

[0068] The dynamic impedance matching module is used to adjust the equivalent input or output impedance of the energy router port according to the system operating status, so as to reduce reflection loss, voltage fluctuation and power oscillation during energy transmission.

[0069] (2) Information interaction and control sequence

[0070] During system operation, the dynamic impedance matching module first obtains the voltage, current and power information of the current port through the electrical parameter detection unit, and calculates the corresponding impedance deviation or impedance change trend.

[0071] When the impedance deviates from the preset target range, the dynamic impedance matching module generates an impedance adjustment command and transmits the command to the hybrid modulation unit.

[0072] Upon receiving an impedance adjustment command, the hybrid modulation unit, without altering the basic topology, changes the equivalent control characteristics of the power converter by adjusting the modulation methods or parameters (including duty cycle, modulation depth, or phase distribution) of SVM and PWM, thereby achieving continuous adjustment of the port equivalent impedance.

[0073] (3) Collaborative closed-loop control mechanism

[0074] The dynamic impedance matching module and the hybrid modulation unit form a closed-loop adjustment structure:

[0075] The dynamic impedance matching module, as the outer loop control, is used to determine the impedance matching status and provide the adjustment direction.

[0076] The hybrid modulation unit operates as the inner loop, performing impedance adjustment and energy control through real-time updates of modulation parameters.

[0077] During load changes, power fluctuations, or AC / DC mode switching, this closed-loop control mechanism can respond quickly, suppress transient voltage surges, and maintain the system in a high-efficiency, low-loss operating range.

[0078] (4) Synergistic effects in topology switching scenarios

[0079] When the energy router performs a topology switch (such as switching between a full-bridge and a half-bridge), the hybrid modulation unit first smoothly adjusts the modulation parameters according to the control strategy to reduce the switching impact. Subsequently, the dynamic impedance matching module reassesses the impedance matching state based on the changes in electrical parameters after the switch and further refines the modulation parameters to achieve rapid stabilization after the topology switch. Compared with traditional silicon-based converters, the energy conversion efficiency is significantly improved, the dynamic response speed is faster, and it can flexibly adapt to various loads and power supply scenarios, making it particularly suitable for the complex needs of AC / DC mixed loads in industrial parks.

[0080] The distributed AI control network hardware consists of 5 edge computing nodes, each equipped with a neural network processor (NPU) to run multi-objective optimization algorithms and a fault prediction model based on a long short-term memory (LSTM) network. The nodes achieve federated learning through an encrypted communication protocol.

[0081] In this invention, the distributed AI control network adopts a federated learning mechanism to achieve multi-node collaborative modeling. Its purpose is to achieve continuous training and updating of fault prediction models and energy optimization models without centralizing the original operating data, thereby balancing model performance and data security.

[0082] (1) System architecture and participating roles of federated learning

[0083] A federated learning system includes at least:

[0084] Multiple edge computing nodes: deployed on power distribution equipment, energy storage units, or energy routers respectively, for local model training;

[0085] Central aggregation node: Used for aggregating and distributing model parameters, it can be deployed in the digital twin platform or the main control server;

[0086] Communication and Security Module: Used for encrypted transmission of model parameters and authentication.

[0087] Each edge computing node uses only locally collected operational data (including voltage, current, temperature, load power, and energy storage health status) for model training, without uploading raw data.

[0088] (2) Local model training and parameter generation

[0089] Each edge computing node pre-stores uniform model structure parameters, including the number of neural network layers, node size, and loss function form.

[0090] In each round of federated learning, each edge node performs several rounds of model training based on local data to obtain the corresponding model parameters or gradient update amounts, which are used to reflect changes in local operating characteristics.

[0091] (3) Model aggregation algorithm

[0092] The central aggregation node uses the FedAvg algorithm to aggregate the model parameters uploaded by each edge node. The basic process is as follows:

[0093] Receive local model parameters from multiple edge nodes;

[0094] The parameters are weighted averaged according to the sample size of each node or the preset weight.

[0095] Generate new global model parameters to reflect the overall operating characteristics of the system;

[0096] The aggregation process does not rely on data from any single node, thus avoiding the risk of single point of failure caused by centralized control.

[0097] (4) Parameter encryption and privacy protection methods

[0098] To ensure the security of model parameters during transmission, this invention introduces a parameter encryption mechanism during federated learning.

[0099] Before uploading model parameters or gradients, each edge node encrypts the parameters using homomorphic encryption or equivalent security encryption, enabling the central aggregation node to complete the aggregation operation without decrypting the parameters of individual nodes.

[0100] In an alternative implementation, a differential privacy mechanism can be introduced during the model update process to further prevent the model from back-engineering local running data by injecting controlled noise into the parameters.

[0101] (5) Communication and Coordination Process

[0102] The communication coordination process for federated learning includes the following steps:

[0103] 1) The central aggregation node broadcasts the current global model parameters and training round instructions to each edge node;

[0104] 2) Each edge node trains the model based on local data and generates encrypted parameter update data;

[0105] 3) Encrypted parameters are uploaded to the central aggregation node via a secure communication link;

[0106] 4) The central aggregation node completes parameter aggregation and generates a new global model;

[0107] 5) The updated global model parameters are distributed to each edge node for the next round of training or for real-time control decisions.

[0108] The aforementioned communication process can be authenticated and recorded using blockchain or an equivalent distributed trusted communication mechanism to ensure the traceability and tamper-proof nature of the model update process.

[0109] (6) The combination of federated learning and system control

[0110] The updated global model parameters were used in the multi-objective optimization model and fault prediction model in the distributed AI control network.

[0111] During system operation, federated learning is executed periodically or event-triggered, and its output directly participates in power allocation decisions, fault warning judgments, and operational strategy adjustments, thereby achieving continuous adaptive optimization of the system.

[0112] Each node constructs a multi-objective optimization model and a fault prediction model based on locally collected data on voltage, current, temperature, and energy storage health status. The federated learning mechanism enables nodes to share optimization strategies while protecting data privacy and periodically update global model parameters.

[0113] The distributed architecture avoids the single point of failure risk of centralized control, and AI algorithms enable real-time optimization and predictive maintenance, significantly improving system response speed and reliability, and adapting to dynamic loads and power fluctuations.

[0114] The multimodal energy storage matrix hardware consists of a 50kWh lithium battery module, a 20kWh supercapacitor module, and a 10kW flywheel energy storage module, equipped with an active thermal management module combining liquid cooling and phase change materials, and an adaptive life management unit.

[0115] The tiered energy dispatch strategy allocates tasks based on load characteristics. Supercapacitor modules handle transient high-power demands, lithium battery modules support continuous power supply, and flywheel energy storage modules smooth out high-power fluctuations. Thermal management modules maintain stable energy storage unit temperatures, while adaptive lifespan management units optimize charging and discharging strategies by monitoring health status, extending lifespan. Multimodal design meets diverse load requirements, reduces voltage fluctuations, and improves system stability; lifespan management reduces energy storage aging and lowers long-term maintenance costs.

[0116] I. Layered Response Mechanism and Time Scale Limitations of Energy Storage Units

[0117] In this invention, the hierarchical energy dispatch strategy is not an abstract functional allocation, but a clear division based on the power density, energy density, and dynamic response characteristics of different energy storage units:

[0118] Supercapacitor module: Used to suppress sudden voltage disturbances and transient high power demands. Its charging and discharging response time is less than 10 ms, and it mainly undertakes the task of transient power compensation for durations of less than 1 s.

[0119] Flywheel energy storage module: used to buffer short-to-medium-term high power fluctuations lasting from 1 s to 60 s. It has a high cycle life and high power output capability, and is used to reduce power slope and smooth load changes.

[0120] Lithium battery module: Used to provide steady-state energy support for a duration of more than 1 minute, responsible for the system's basic energy supply and medium- to long-term power balance.

[0121] By identifying the magnitude, rate of change (dP / dt), and duration of load power changes, the system can allocate power demands at different time scales to the corresponding energy storage units, thereby preventing a single energy storage unit from operating under unsuitable conditions.

[0122] II. Monitoring and Optimized Control Methods for State of Health (SOH)

[0123] The adaptive lifespan management unit performs real-time assessments of the energy storage unit's health status based on multi-source operational data, and the health status includes at least:

[0124] Lithium battery module: capacity decay rate, internal resistance change, cycle count, temperature rise rate;

[0125] Supercapacitor Module: Equivalent Series Resistance (ESR), Capacitance Decay Trend;

[0126] Flywheel energy storage module: bearing vibration amplitude, speed decay characteristics, and energy loss rate.

[0127] Each energy storage unit acquires parameters such as voltage, current, temperature, and rotation speed through built-in sensors, and the edge nodes in the distributed AI control network extract features to form a health status index (SOH).

[0128] During the scheduling process, the adaptive lifetime management unit dynamically adjusts the scheduling weights based on the SOH value, for example:

[0129] When the SOH of a certain lithium battery module drops below a preset threshold, its participation in high-rate discharge conditions is reduced.

[0130] High-frequency charging and discharging tasks will be preferentially assigned to supercapacitors or flywheel energy storage modules.

[0131] The aging process can be slowed down by limiting the depth of charge and discharge (DOD) and rate of the battery.

[0132] III. Collaborative Control Mechanism of Thermal Management Module

[0133] The thermal management module is used to maintain each energy storage unit within a suitable operating temperature range. It works in conjunction with the adaptive lifespan management unit and the distributed AI control network, and specifically includes:

[0134] Real-time monitoring of energy storage unit temperature and temperature rise rate;

[0135] When an abnormal temperature or local hot spot is detected, thermal constraint information is fed back to the scheduling and control layer.

[0136] Based on this, the scheduling and control layer adjusts the power allocation strategy, reducing the load ratio of the corresponding energy storage unit, or transferring the power task to the energy storage unit with better temperature conditions.

[0137] By incorporating thermal state as one of the scheduling constraint parameters, performance degradation and lifespan reduction due to overheating can be avoided.

[0138] IV. Hierarchical Scheduling Implementation in Distributed AI Control Networks

[0139] The distributed AI control network serves as the system's decision-making core, achieving hierarchical scheduling of supercapacitors, lithium batteries, and flywheel energy storage through the following methods:

[0140] State perception layer: Collects load power characteristics, energy storage unit SOH and thermal state;

[0141] Decision analysis layer: Generates hierarchical scheduling instructions based on time scale discrimination, health status weights, and thermal constraints;

[0142] Execution control layer: Sends charging and discharging control signals to the power conversion interface of the corresponding energy storage unit to realize energy distribution.

[0143] This scheduling method meets load demands while also taking into account the lifespan management and safe operation of energy storage units, achieving comprehensive optimization of system performance and long-term reliability.

[0144] The hardware components of the multimodal communication and digital twin platform include 5G and power line carrier (PLC) communication modules, support for a blockchain energy trading interface, and an integrated digital twin server for real-time simulation. 5G and PLC provide low-latency communication, ensuring collaborative control of distributed AI nodes; the blockchain interface supports point-to-point trading of photovoltaic power and park loads, guaranteeing data security; the digital twin platform constructs a virtual model of the system, simulating operational status in real time to optimize control strategies and fault response. High-reliability communication supports millisecond-level response times, blockchain transactions improve economic efficiency, and the digital twin platform reduces maintenance costs and enhances system scalability through simulation.

[0145] The 5G communication module is primarily used to carry service data with high real-time and bandwidth requirements, including: control commands generated by the distributed AI control network, status data of key energy storage units (such as voltage, current, temperature, and SOH indicators), and real-time operating parameters required by the digital twin platform. Leveraging the low latency and high reliability of 5G, it ensures the timeliness of data during the control closed-loop and simulation synchronization processes.

[0146] The PLC communication module utilizes existing power lines for data transmission. It is primarily used to carry business data that is less sensitive to latency but requires high stability and continuity, including: basic operation monitoring data, energy metering information, transaction records related to blockchain energy trading, and ledger synchronization data. PLC communication does not rely on external wireless networks and has high anti-interference capabilities in local network environments.

[0147] During system operation, 5G and PLC form a primary-secondary cooperative relationship: under normal working conditions, the two communication methods work in parallel according to the preset business division; when an abnormality of the 5G link is detected or the wireless environment deteriorates, some non-real-time services can be switched to PLC link transmission, thereby improving the robustness of system communication and continuous operation capability.

[0148] The digital twin server maintains data synchronization with the physical system through the aforementioned multimodal communication interface, acquires key state parameters in real time and performs simulation calculations. Its simulation results and optimization strategies are then prioritized and sent to the control layer via the 5G link to achieve virtual-physical collaborative operation.

[0149] The usage method of this embodiment is performed according to the following steps, as follows:

[0150] Step S1: System Initialization and Data Acquisition

[0151] The equipment is connected to the industrial park's power distribution network, linking to a photovoltaic power generation system, energy storage matrix, AC motors, and DC LED lighting. Sensors collect real-time data, including voltage, current, load power, photovoltaic output power, electricity price, and energy storage health status. The operational target is set to prioritize efficiency through a digital twin platform.

[0152] Step S2: Model Building and Update

[0153] The distributed AI control network constructs a multi-objective optimization model based on collected data (the objective function comprehensively optimizes efficiency, cost, and stability); the fault prediction model is based on an LSTM network to predict the probability of potential faults. Each node trains its local model through federated learning, and the model parameters are transmitted encrypted hourly via a blockchain network. The optimization weights are dynamically adjusted based on the fault prediction results, prioritizing stability.

[0154] The objective function of the multi-objective optimization algorithm for the distributed AI control network is: ;

[0155] in, For system energy conversion efficiency, Total operating cost, For voltage stability; w1, w2, w3 are weighting coefficients that satisfy w1 + w2 + w3 = 1, and are dynamically adjusted according to the operating target; constraints include: ,in These are power supply capacity, energy storage capacity, and load capacity, respectively. For real-time voltage and current, Within a safe range.

[0156] The fault prediction model of the distributed AI control network is based on an LSTM network, and the fault probability calculation formula is as follows: Where σ is the Sigmoid activation function, For model parameters, These are real-time voltage, current, device temperature, and energy storage health status, respectively; the model updates its parameters periodically through federated learning.

[0157] The multi-objective optimization model coordinates system operation through a comprehensive objective function, balancing energy conversion efficiency, operating cost, and system stability. The core idea is to transform multiple operational objectives into a unified mathematical optimization problem, dynamically adjusting weights to adapt to different scenario requirements, such as efficiency-first or cost-first approaches. Based on real-time collected data such as voltage, current, load power, and electricity price, the model calculates the optimal power allocation scheme, ensuring efficient and stable energy flow between the power source and load. Constraints guarantee power balance and safe voltage and current ranges, preventing system overload or failure.

[0158] LSTM-based fault prediction models should employ at least one of the following network structures:

[0159] The fault prediction model is a multivariate time series LSTM neural network, whose network structure includes an input layer, at least one LSTM hidden layer, and an output layer. The input layer receives a sequence of operating state features arranged in chronological order. These operating state features include at least one or more of the following: voltage, current, power, temperature, vibration amplitude, harmonic content, and equipment health indicators. The input data forms a feature matrix in the form of a fixed time window.

[0160] LSTM hidden layers are used to model long-term dependencies in time series. They contain input gates, forget gates, and output gates. The number of hidden units is preferably 32 to 128 to achieve a balance between prediction accuracy and computational complexity. In some embodiments, two LSTM hidden layers can be cascaded to improve the ability to model complex fault evolution trends.

[0161] The output layer is either a fully connected layer or a Sigmoid activation layer, which maps the output of the LSTM hidden layer to a fault occurrence probability value. The output result is a continuous probability value between 0 and 1, which is used to represent the possibility of a potential fault occurring within the prediction time window.

[0162] In a federated learning scenario, each distributed node independently trains its local LSTM model on locally collected runtime data based on the unified network structure described above. It only uploads model parameters or parameter increments to the central coordination node for aggregation and updates, without uploading the original runtime data. This achieves collaborative training of the fault prediction model while protecting privacy.

[0163] The multi-objective optimization model described in this invention adopts at least the following comprehensive objective function form to coordinate the operation of multiple energy storage units and loads in an AC / DC hybrid power distribution system:

[0164] ;

[0165] in:

[0166] J: Comprehensive optimization objective function;

[0167] t: discrete time step, T is the length of the optimization time window;

[0168] V(t): The voltage of the critical bus or common coupling point measured at time t;

[0169] V ref : System rated reference voltage value;

[0170] α: Voltage stability weighting coefficient, used to adjust the importance of voltage deviation in the optimization objective;

[0171] P loss (t): The energy loss term of the system at time t, including power conversion loss and line loss;

[0172] β: Energy loss weighting coefficient;

[0173] C deg (t): The aging cost function of the energy storage unit at time t, which is used to characterize the lifespan consumption of the energy storage unit during the charging and discharging process;

[0174] γ: Lifetime cost weighting coefficient.

[0175] In the specific implementation process, the aging cost function C deg (t) can be constructed based on the charge / discharge rate, depth of discharge, and state of health (SOH) of the energy storage unit to limit the adverse effects of high-rate or deep cycling on the lifespan of the energy storage.

[0176] While satisfying the above objective function, the multi-objective optimization model is also subject to the following constraints:

[0177]

[0178] in:

[0179] P gen (t): Output power of the distributed power source at time t;

[0180] P sto (t): The charging and discharging power of the energy storage unit (positive value indicates discharging, negative value indicates charging);

[0181] P load (t): Load power requirement;

[0182] The minimum and maximum allowable power of the energy storage unit;

[0183] SOC(t): The state of charge of the energy storage unit at time t;

[0184] : SOC operating limits.

[0185] The model iteratively optimizes and, combined with distributed computing nodes processing real-time data, dynamically adjusts the operating mode of the energy router and the scheduling strategy of the energy storage units. This adaptive optimization relies on machine learning algorithms, which can predict load demand based on historical and real-time data and plan energy allocation paths in advance.

[0186] The fault prediction model utilizes Long Short-Term Memory (LSTM) networks to analyze equipment operating data, such as voltage, current, temperature, and energy storage health status, to predict the probability of potential faults. Its core lies in capturing the dynamic changes in equipment operating status through time-series data and identifying abnormal patterns. The model learns fault characteristics through training, generating probability outputs to determine whether preventative measures are necessary. Distributed nodes collaboratively train and periodically update model parameters to ensure that prediction results adapt to different operating environments.

[0187] The model processes multidimensional input data in real time, mapping features to a probability space using activation functions. The prediction results are fed back to the control system, dynamically adjusting operational strategies, such as reducing the load on high-risk modules to prevent failures.

[0188] The multi-objective optimization model and the fault prediction model work together to achieve system intelligence. The output of the fault prediction model serves as a dynamic input, adjusting the weights of the multi-objective optimization model to prioritize system stability. For example, when a potential fault risk is detected, the optimization model reduces its focus on efficiency and increases its weight on stability. This collaborative mechanism ensures that the system remains efficient and reliable in complex and dynamic scenarios.

[0189] The multi-objective optimization model overcomes the limitations of traditional single-objective control by comprehensively optimizing efficiency, cost, and stability. It is adaptable to various application scenarios, such as industrial parks and charging stations, enhancing system flexibility. The model dynamically adjusts its optimization strategy based on real-time data, addressing the inability of traditional fixed rules to handle load and power fluctuations, significantly improving energy utilization efficiency. By setting power balance and safety range constraints, the model ensures the system operates in a safe and reliable state, reducing the risk of overload or failure.

[0190] The advantages of fault prediction models include early identification of potential faults, avoiding the lag of traditional threshold detection, reducing unexpected downtime, and improving system reliability. Based on the analysis of multidimensional time series data, the model can capture subtle anomaly patterns, achieving higher prediction accuracy than traditional methods and making it suitable for complex operating environments. Through distributed node training and parameter updates, the model maintains high adaptability while protecting data privacy, making it suitable for smart grid scenarios with multi-device collaboration.

[0191] The fault prediction model adjusts the energy flow optimization strategy in real time, achieving a dynamic balance between efficiency and reliability, and overcoming the performance degradation problem of traditional systems in high-risk scenarios. The combination of distributed computing and federated learning significantly enhances the system's intelligence level, enabling it to adapt to complex scenarios and reduce maintenance costs. The collaborative mechanism allows the system to find the optimal balance between efficiency, cost, stability, and reliability, significantly improving economic efficiency and environmental friendliness, making it particularly suitable for the diverse needs of future smart grids.

[0192] Step S3: Dynamic Networking and Topology Optimization

[0193] The multi-port energy router dynamically switches between a full-bridge topology for high-power motor loads and a half-bridge topology for low-power LED loads based on the optimization model output. The impedance matching module optimizes energy transmission efficiency. Distributed AI nodes collaboratively allocate photovoltaic and energy storage power, prioritizing DC loads to reduce conversion losses.

[0194] Step S4: Multi-objective energy optimization

[0195] According to the optimization model, photovoltaic power is prioritized for DC LED lighting, with surplus power stored in an energy storage matrix. During peak periods, motors are supplemented by lithium batteries and flywheel energy storage, reducing reliance on the main grid. The energy storage matrix allocates tasks based on load characteristics, while supercapacitors handle the transient demands of motor startup.

[0196] Step S5: Fault Prediction and Response

[0197] When the fault prediction model detects an increase in the temperature of the energy storage module and the probability exceeds a threshold, it triggers an early warning, automatically reduces the module's utilization rate, and switches to standby mode. The digital twin platform verifies the response strategy, and the blockchain records fault data for traceability.

[0198] Based on statistical analysis of extensive historical operational data and digital twin simulation results, a fault probability threshold of 0.8 was determined. By comparing the false alarm rate and false negative rate under different fault probability thresholds, it was found that when the fault probability is less than 0.8, the system's accuracy in identifying potential faults is insufficient, posing a risk of false negatives. When the fault probability is greater than or equal to 0.8, the system can significantly improve the reliability of early fault warnings while maintaining a low false alarm rate. Furthermore, multi-condition simulation verification of the standby mode switching process was conducted using a digital twin platform, confirming that triggering the warning and executing the switch under this threshold condition does not adversely affect the system's safety and stability. Therefore, using a fault probability greater than 0.8 as the criterion for triggering warnings and standby mode switching is both engineeringly reasonable and feasible.

[0199] In this invention, the digital twin platform verifies the response strategy through virtual-real mapping modeling, strategy injection simulation, and security assessment.

[0200] First, the digital twin platform constructs a virtual model that corresponds one-to-one with the actual AC / DC hybrid power distribution system based on the physical system's structural parameters, operating parameters, and real-time acquired status data. The virtual model includes at least a power supply module, an energy storage module, a power conversion unit, and a load model, and its parameters are kept synchronized with the physical system through a multi-modal communication interface.

[0201] When the fault prediction model outputs a fault probability exceeding a preset threshold and generates a response strategy, this response strategy is injected into the digital twin model for pre-simulation verification before being deployed to the physical system. The verification process includes: simulating fault scenario triggering in the virtual model, executing the corresponding power switching or topology reconfiguration strategy, and monitoring changes in key indicators in real time.

[0202] The key indicators include at least: key bus voltage deviation, power fluctuation amplitude, energy storage unit load ratio, and system stabilization time. If the simulation results meet the preset safety constraints (such as voltage not exceeding limits, power change rate not exceeding thresholds, and energy storage units not entering abnormal operating ranges), the response strategy is determined to be an executable strategy and allowed to be applied to the physical system; otherwise, the response strategy is adjusted or reverted to a conservative control mode.

[0203] After the response strategy is executed, fault-related data is recorded via a blockchain interface for subsequent traceability and auditing. The recorded data includes at least: fault type, fault occurrence time, predicted probability value, the identifier of the adopted response strategy, digital twin simulation verification results, and key operational indicators after execution. This data is written to the blockchain ledger in block form to ensure the immutability and traceability of the records, providing a basis for subsequent fault analysis, strategy optimization, and liability determination.

[0204] Step S6: System Expansion and Continuous Optimization

[0205] New energy storage capacity can be added by hot-swapping modules, and the model automatically adapts to the new configuration, optimizing power allocation. The blockchain interface supports the trading of excess photovoltaic power, and the digital twin platform regularly analyzes operational data to update the AI ​​model to improve long-term performance.

[0206] The working principle of this embodiment is based on the collaborative optimization of hardware and software. The multi-port adaptive energy router, through GaN-based topology and hybrid modulation technology, achieves efficient AC / DC energy conversion and dynamic allocation, adapting to the diverse loads of industrial parks. The distributed AI control network utilizes multi-objective optimization models and fault prediction models to adjust power allocation in real time and provide early warnings of potential faults. The multi-modal energy storage matrix meets different load demands through hierarchical scheduling, while thermal management and lifespan management technologies ensure long-term stability. The multi-modal communication and digital twin platform ensure data security and efficient collaboration through 5G / PLC communication and blockchain transactions; the digital twin model optimizes operating strategies through real-time simulation, reducing maintenance costs. All modules of the system work collaboratively to achieve efficient energy management, early fault prevention, and flexible system expansion.

[0207] I. Experimental Objective

[0208] To verify the impact of the synergistic effect between the multi-port adaptive energy router, distributed AI control network, multimodal energy storage matrix, and digital twin platform on the performance of the AC / DC hybrid power distribution system, a comparative experiment was designed to compare the performance of the collaborative control system and the non-collaborative traditional system under the same operating conditions.

[0209] II. Experimental Subjects and Comparison Scheme

[0210] 1. Experimental System Configuration

[0211] Experimental System A (Collaborative System of this Invention)

[0212] Enabled:

[0213] Multi-port adaptive energy router;

[0214] Distributed AI multi-objective optimization control;

[0215] Multimodal energy storage matrix hierarchical scheduling;

[0216] Digital twin-assisted strategy verification;

[0217] Compare with system B (non-cooperative system).

[0218] Enabled:

[0219] Fixed topology power converter;

[0220] Traditional rule-based control (without AI optimization);

[0221] Single type of energy storage (lithium battery only);

[0222] No digital twin simulation support;

[0223] 2. Experimental Operating Conditions

[0224] project parameter AC side load 380V Three-phase Inductive Load DC-side load 48V / 380V mixed DC load Load changes Step change ±20% of rated power Distributed power Photovoltaics + Power Grid Energy storage status Initial SOC 50% Measurement points Critical load bus / PCC

[0225] III. Experimental Procedure

[0226] The system starts up and enters steady-state operation;

[0227] Under the same load conditions, run system A and system B respectively;

[0228] Introducing load step changes and power fluctuations;

[0229] Continuous data acquisition system efficiency, voltage fluctuation, response time, and operating costs;

[0230] Average performance indices under statistical steady-state and dynamic processes.

[0231] IV. Comparative Experimental Data (Core Supporting Factors)

[0232] Table 1 System performance comparison results

[0233] Note: Operating costs are normalized to 1 for traditional systems, while collaborative systems reduce overall costs through energy storage scheduling and energy optimization.

[0234] Experimental results show that the synergistic effect between the multi-port adaptive energy router, the distributed AI control network, the multimodal energy storage matrix, and the digital twin platform significantly improves system performance.

[0235] Among these features, the dynamic topology switching and impedance matching of the energy router reduce conversion losses; the distributed AI control network achieves real-time optimal power allocation through multi-objective optimization; the multi-modal energy storage matrix provides energy buffering capabilities at different time scales; and the digital twin platform reduces unnecessary control trial and error through simulation prediction. The coordinated operation of these modules enables the system to outperform non-cooperative control systems in key indicators such as efficiency, dynamic response, voltage stability, and operating costs, thereby significantly improving the overall performance of the AC / DC hybrid power distribution system.

[0236] This embodiment addresses the low efficiency and poor adaptability of traditional converters through a multi-port adaptive energy router, significantly improving energy conversion efficiency and dynamic response capabilities. The distributed AI control network optimizes power allocation and fault prediction through mathematical models, overcoming the slow response and insufficient reliability of centralized control, and adapting to highly dynamic load scenarios. The multimodal energy storage matrix, through hierarchical scheduling and lifetime management, meets transient and continuous power supply demands, extends energy storage life, and reduces maintenance costs. The multimodal communication and digital twin platform, through blockchain transactions and real-time simulation, enhances economic efficiency and system scalability, while promoting clean energy utilization and improving environmental friendliness. These combined advantages enable the system to efficiently and reliably serve complex scenarios such as smart industrial parks, providing innovative solutions for the future smart grid.

[0237] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. A flexible AC / DC power distribution network device, characterized in that, include: Multi-port adaptive energy router, distributed AI control network, multimodal energy storage matrix, and multimodal communication and digital twin platform; The multi-port adaptive energy router adopts a gallium nitride-based multiphase interleaved parallel topology and integrates a dynamic impedance matching module and a hybrid modulation unit. The dynamic impedance matching module is used to detect and match the impedance characteristics of the power supply side and the load side in real time. The hybrid modulation unit is used to combine space vector modulation (SVM) and pulse width modulation (PWM) to support bidirectional AC and DC energy flow and can dynamically switch the circuit topology according to the instructions of the distributed AI control network to adapt to different loads and power requirements. The distributed AI control network includes multiple locally deployed edge computing nodes. Each node is equipped with a neural network processor, which runs an adaptive multi-objective optimization algorithm and a fault prediction model based on a long short-term memory network, respectively. The model parameters are updated locally and synchronized globally through a federated learning mechanism. The multimodal energy storage matrix includes lithium batteries, supercapacitors, and flywheel energy storage modules. Each energy storage module integrates an active thermal management module and an adaptive lifespan management unit. The active thermal management module maintains the temperature stability of the energy storage unit, and the adaptive lifespan management unit optimizes the charging and discharging strategy by monitoring the health status of each module in real time. The multimodal energy storage matrix allocates energy according to power allocation instructions containing a hierarchical energy scheduling strategy issued by a distributed AI control network. The multimodal communication and digital twin platform supports 5G or 6G, power line carrier and low power wide area network communication protocols, and integrates a blockchain-based energy trading interface and digital twin model to realize real-time simulation of system operation status, strategy optimization and distributed energy peer-to-peer trading. The multi-port adaptive energy router, distributed AI control network, multimodal energy storage matrix, and multimodal communication and digital twin platform work together through preset communication protocols and data interfaces to achieve adaptive energy management, multi-objective optimization, and predictive maintenance of AC / DC hybrid power distribution systems.

2. The flexible AC / DC power distribution network equipment according to claim 1, characterized in that, The hybrid modulation unit of the multi-port adaptive energy router combines space vector modulation (SVM) and pulse width modulation (PWM), and adjusts the operating point in real time through a dynamic impedance matching module.

3. The flexible AC / DC power distribution network equipment according to claim 1, characterized in that, The objective function of the adaptive multi-objective optimization algorithm for the distributed AI control network is: ; in, For the system energy conversion efficiency, P out and P in These are the input and output active power, respectively. Let C be the total operating cost, t be the discrete-time index, and C be the total operating cost. t Let P be the unit electricity price or energy cost coefficient at time t. grid,t Let be the active power purchased from the public grid at time t; For voltage stability, ΔV is the deviation between the actual operating voltage and the rated voltage, V nom The system's rated voltage; w1, w2, and w3 are weighting coefficients that satisfy w1 + w2 + w3 = 1, and are dynamically adjusted according to the operational target; constraints include: ,in These are power supply capacity, energy storage capacity, and load capacity, respectively. For real-time voltage and current, Within a safe range.

4. The flexible AC / DC power distribution network equipment according to claim 1, characterized in that: The fault prediction model of the distributed AI control network is based on a long short-term memory network, and the fault probability calculation formula is as follows: Where σ is the Sigmoid activation function, For model parameters, These are real-time voltage, current, equipment temperature, and energy storage health status, respectively. The fault prediction model updates its parameters periodically through a federated learning mechanism. Each edge node calculates the model gradient based on local data, uploads it to the central aggregation server after homomorphic encryption, and then the server executes the federated averaging algorithm to update the global model parameters before distributing them to each node.

5. The flexible AC / DC power distribution network equipment according to claim 1, characterized in that: The hierarchical energy scheduling strategy is as follows: the supercapacitor energy storage module handles transient loads, the lithium battery energy storage module supports continuous loads, and the flywheel energy storage module is used for high-power buffering; the adaptive life management unit monitors the health status of each module in real time and dynamically adjusts its charging and discharging current upper limit and depth based on the health status value.

6. The flexible AC / DC power distribution network equipment according to claim 1, characterized in that: In the multimodal communication and digital twin platform, the 5G or 6G communication module is used for low-latency control command transmission between distributed AI control nodes, the power line carrier PLC module is used for data transmission between fixed devices, and the low-power wide area network module is used to connect distributed sensing devices. The blockchain energy trading interface is implemented based on smart contracts, supporting automatic peer-to-peer energy trading and settlement between distributed power sources and loads; The digital twin model updates the virtual system state by receiving real-time sensor data and outputs optimized control parameters to the distributed AI control network based on simulation results.

7. A method of using a flexible AC / DC power distribution network equipment as described in any one of claims 1-6, characterized in that, Includes the following steps: Step S1: System initialization and data acquisition. Connect the equipment to the power distribution network, connect the distributed power source and the load, collect real-time data through sensors, including voltage, current, load power, power supply power, electricity price and energy storage health status, and set the initial operation target through multimodal communication and digital twin platform. Step S2: Model building and updating. The distributed AI control network builds a multi-objective optimization model and a fault prediction model based on the collected data, and updates the model parameters and global synchronization regularly through a federated learning mechanism. Step S3: Dynamic networking and topology optimization. The multi-port energy router dynamically switches its circuit topology and adjusts impedance matching parameters according to the power allocation instructions obtained by the multi-objective optimization algorithm. At the same time, the distributed AI nodes collaboratively calculate and allocate the output power of each power source and energy storage. For the allocation of the energy storage matrix, transient, continuous and high-power tasks are assigned to the supercapacitor, lithium battery and flywheel module respectively according to the hierarchical energy scheduling strategy. Step S4: Based on the solution results of the multi-objective optimization model, prioritize the scheduling of clean energy power supply; at the same time, according to the hierarchical energy scheduling strategy, allocate the transient fluctuation task of the load to the supercapacitor module, the continuous power supply task to the lithium battery module, and the high-power buffer task to the flywheel energy storage module. Step S5: Fault prediction and response. Based on the fault probability output by the fault prediction model, when the fault probability is greater than the threshold, an early warning is triggered and the system switches to standby mode. At the same time, the security of the switching strategy is verified through simulation using multimodal communication and a digital twin platform. Step S6: System expansion and continuous optimization. Energy storage and ports are expanded through hot-swappable modules. The distributed AI control network automatically identifies newly connected devices and updates and optimizes the model. Peer-to-peer trading of surplus electricity is realized through the blockchain energy trading interface, and the model parameters are continuously optimized by combining long-term operating data.

8. The method of using a flexible AC / DC power distribution network equipment according to claim 7, characterized in that: In step S2, the parameter synchronization process of the federated learning mechanism includes: S2.1 Each edge node calculates the model gradient based on local data and encrypts the gradient using the Paillier homomorphic encryption algorithm; S2.2 The encrypted gradient is uploaded to the central aggregation server through a secure channel; S2.3 The server aggregates and calculates the encryption gradients, and executes the federated averaging algorithm to update the global model parameters; S2.

4. Distribute the updated global model parameters to each edge node to complete one round of model update.

9. The method of using a flexible AC / DC power distribution network equipment according to claim 7, characterized in that: In step S5, the digital twin platform updates the system virtual model in real time, verifies the optimization strategy and fault response effect, and records fault data through the blockchain interface when a fault occurs to ensure traceability.

10. The method of using a flexible AC / DC power distribution network equipment according to claim 7, characterized in that: In step S6, the excess electricity is traded peer-to-peer through the blockchain energy trading interface, and low-cost power is prioritized for allocation using a multi-objective optimization model.

Images