Composite resonant power supply and distribution system based on silicon carbide devices and control method thereof

Abstract

The embodiment of the present disclosure discloses a composite resonant power supply and distribution system based on silicon carbide devices and a control method thereof. It comprises a bridge resonant topology module composed of silicon carbide MOSFET, a dynamic parameter tuning module based on neural network, a blockchain verification module, a magnetic integrated composite insulation transformer and a multi-level digital twin monitoring module. The dynamic parameter tuning module predicts the optimal resonant frequency according to the collected load characteristic data and generates the adjustment instruction, the blockchain verification module verifies the consensus of the instruction, and the bridge resonant topology module adjusts the resonant frequency according to the verified instruction. Through the collaborative design of hardware topology and intelligent algorithm, the present application realizes the rapid response to load change and efficient power conversion, while ensuring the safety of the control instruction, and is suitable for high-frequency high-power scenes with high power supply quality requirements.

Description

Technical Field

[0001] Applications generally relate to the field of power electronics technology, particularly to composite resonant power supply and distribution systems based on silicon carbide devices and their control methods. Background Technology

[0002] Currently, composite resonant power supply and distribution technology utilizes the resonant characteristics of inductors and capacitors at specific frequencies to optimize and regulate the voltage and current waveforms in power systems, thereby improving power supply efficiency and power quality. This technology is widely used in power electronic converters, reactive power compensation devices, and new energy grid-connected systems. By rationally designing resonant parameters, it can effectively reduce losses, suppress harmonics, and improve system stability.

[0003] With the rapid development of high-frequency, high-power applications such as 5G communication, data centers, and edge computing, power supply and distribution systems face multiple challenges, including drastic dynamic load changes, high power quality requirements, and limited space. Traditional power supply and distribution technologies mainly include resonant converter methods, LC passive filtering technology, fixed-frequency resonant control technology, IGBT-based resonant topology technology, centralized reactive power compensation technology, single-resonant network tuning technology, and DSP-based static tuning technology. However, these traditional solutions have significant shortcomings in handling high-frequency dynamic response, adaptability to wide load ranges, and intelligent system control.

[0004] Specifically, existing power supply and distribution systems struggle to effectively support real-time interactive verification of high-frequency communication protocols and smart contracts, lacking mechanisms to ensure communication stability under complex electromagnetic environments. Furthermore, traditional control methods exhibit lag in response to sudden load changes, failing to achieve real-time coordination between dynamic resonant tuning and safety verification. In addition, in high-frequency, high-power scenarios, existing transformer designs suffer from large size, high losses, and limited insulation performance, making it difficult to meet the system's demands for high power density and compact design. While wide-bandgap devices such as silicon carbide have been applied in some fields, breakthroughs are still needed in optimizing their magnetic integration structure at multiple resonant frequencies and in their collaborative operation with intelligent control algorithms.

[0005] In summary, there is an urgent need to propose a novel power supply and distribution system that integrates composite resonant dynamic tuning, intelligent safety verification mechanisms, and magnetic integration optimization design to meet the comprehensive requirements for system efficiency, stability, and safety in high-frequency, high-power scenarios. Summary of the Invention

[0006] This disclosure addresses some deficiencies mentioned in the background art by providing a composite resonant power supply and distribution system based on silicon carbide devices and its control method.

[0007] In a first aspect, embodiments of this disclosure provide a composite resonant power supply and distribution system based on silicon carbide devices, comprising: The bridge resonant topology module, composed of silicon carbide MOSFETs, is used to perform high-frequency power conversion; A dynamic parameter tuning module, connected to the bridge resonant topology module, includes an IoT sensor array and a neural network model. The IoT sensor array is used to collect load characteristic data, and the neural network model is used to predict and generate adjustment commands for the optimal resonant frequency based on the load characteristic data. The blockchain verification module is communicatively connected to the dynamic parameter tuning module and is used to perform consensus verification on the adjustment command and output the verified adjustment command after the verification is successful. A magnetically integrated composite insulated transformer, operating at high frequency, is coupled to the bridge resonant topology module for voltage transformation and electrical isolation; and The multi-level digital twin monitoring module is communicatively connected to the bridge resonant topology module, dynamic parameter tuning module, blockchain consensus verification system, and magnetically integrated composite insulation transformer, and is used for real-time monitoring and simulation analysis of the system's operating status.

[0008] In one embodiment of the first aspect, the bridge resonant topology module includes: A full-bridge circuit composed of silicon carbide MOSFET switches; The series resonant circuit located on the primary side includes a primary-side leakage inductance and a primary-side resonant capacitor connected in series. The voltage divider resonant circuit located on the secondary side includes a secondary-side resonant inductor and a secondary-side resonant capacitor connected in series. The synchronous rectification circuit connected to the full-bridge circuit is used to perform synchronous rectification on the secondary side.

[0009] In one embodiment of the first aspect, the bridge resonant topology module further includes a real-time control unit connected to the control terminal of the blockchain verification module and the bridge resonant topology module, for generating a drive pulse according to the verified adjustment command to control the turn-on and turn-off timing of the silicon carbide MOSFET switch, thereby realizing the adjustment of the resonant frequency and the switching timing.

[0010] In one embodiment of the first aspect, the real-time control unit includes an instruction execution unit that employs a proportional-integral-derivative (PID) control algorithm to perform closed-loop adjustment of the resonant frequency according to the verified adjustment instruction.

[0011] In one embodiment of the first aspect, the real-time control unit is configured to execute a timing control strategy for the resonant converter, the timing control strategy comprising sequentially looping: In the first conduction stage, the first pair of switching transistors of the full-bridge circuit are turned on, so that energy is stored in the transformer core. During the dead time phase, all primary-side switches are turned off, and the synchronous rectifier circuit is turned on, so that the energy stored in the transformer core is released to the load. In the second conduction phase, the second pair of switching transistors of the full-bridge circuit are turned on, so that energy is stored in the transformer core again. During the secondary synchronous rectification stage, all primary-side switches are turned off, and the synchronous rectification circuit is turned on, so that the energy stored in the transformer core is released to the load again.

[0012] In one embodiment of the first aspect, the dynamic parameter tuning module includes: An Internet of Things (IoT) sensor array is used to collect load characteristic data, including at least current ripple rate, total harmonic distortion rate, and power factor. A neural network prediction unit, connected to the sensor array, has a built-in long short-term memory network model for predicting the optimal resonant frequency based on the load characteristic data. The instruction generation unit, connected to the neural network prediction unit, is used to generate adjustment instructions containing the target frequency based on the prediction results.

[0013] In one embodiment of the first aspect, the dynamic parameter tuning module further includes a data preprocessing unit for cleaning, normalizing and reducing the dimensionality of the collected load feature data, and inputting the processed data into the neural network model.

[0014] In one embodiment of the first aspect, the blockchain verification module adopts an authoritative proof consensus mechanism, which verifies the hash consistency of the adjustment instruction content through multiple authorized nodes, and signs and confirms the instruction after consensus is reached.

[0015] In one embodiment of the first aspect, the multi-level digital twin monitoring module is connected to the bridge resonant topology module, the dynamic parameter tuning module, and the blockchain verification module, and is used to collect system operating status data in real time, construct a digital twin model, and perform status simulation and fault early warning.

[0016] Secondly, the composite resonant power supply and distribution control method based on silicon carbide devices, applied to the composite resonant power supply and distribution system based on silicon carbide devices as described above, includes the following steps: S1: Collect load characteristic data through an IoT sensor array; S2: The dynamic parameter tuning module processes the load characteristic data based on a neural network model to generate an adjustment command for the optimal resonant frequency; S3: The adjustment command is verified through consensus by the blockchain verification module. After the verification is successful, the adjustment command is sent to the real-time control unit. S4: The real-time control unit drives the silicon carbide MOSFET in the bridge resonant topology unit according to the adjustment command, and adjusts its switching frequency and resonant network parameters to achieve dynamic response to load changes. S5: Real-time monitoring and simulation of system operation status through multi-level digital twin monitoring modules.

[0017] This disclosure proposes a composite resonant power supply and distribution system and method based on silicon carbide devices, which has the following beneficial effects: (1) By using silicon carbide MOSFETs to construct a composite resonant topology and combining the cooperative working mechanism of primary-side series resonance and secondary-side voltage divider resonance, the conversion efficiency of the system over a wide load range is significantly improved. At the same time, the dynamic parameter tuning algorithm based on neural networks can predict the optimal resonant frequency in real time. Combined with the fast-response execution unit, it can realize high-speed tracking and adjustment of load changes, effectively solving the problem of significant efficiency drop and response lag in traditional power supply and distribution systems when load changes suddenly.

[0018] (2) A blockchain consensus verification mechanism is introduced to verify the adjustment instructions through multiple nodes, ensuring the security and consistency of the instructions during transmission and execution. This mechanism can effectively resist the risk of malicious tampering or misoperation, improve the reliability of the system in complex electromagnetic environments, and overcome the control failure problem caused by communication interference or single point of failure in traditional control methods.

[0019] (3) A magnetically integrated composite insulation transformer is adopted. Through the combination of optimized magnetic circuit structure and high-performance magnetic core material, the core loss is effectively reduced under high-frequency operating conditions. This design significantly reduces the transformer volume while improving insulation performance, solving the problems of large size, high loss and poor insulation performance of traditional transformers in high-frequency applications.

[0020] (4) By collecting multi-dimensional load characteristic parameters through IoT sensor array and combining them with neural network for feature extraction and resonant frequency prediction, the system has self-learning and self-adaptive capabilities and can dynamically adjust operating parameters according to load characteristics, thus realizing intelligent control of power supply and distribution system and avoiding the limitations of traditional schemes that rely on manual experience for parameter tuning.

[0021] (5) A multi-level digital twin monitoring system was constructed to realize real-time monitoring and simulation analysis of the system's operating status. This monitoring system can provide data support for operation and maintenance decisions, and has the ability to provide fault early warning and trend prediction, making up for the lack of real-time diagnostic means and maintenance prediction capabilities in traditional power supply and distribution systems.

[0022] (6) The system operates at a high frequency and is suitable for high-frequency, high-power application scenarios with high power quality requirements and frequent load fluctuations, such as communication base stations, data centers, and electric vehicle charging stations. It meets the comprehensive requirements of modern power electronic systems for high efficiency, high power density, and high reliability. Attached Figure Description

[0023] Figure 1 This is a diagram of a composite resonant power supply and distribution system based on silicon carbide devices according to the present disclosure; Figure 2 The flowchart below illustrates the composite resonant power supply and distribution control method based on silicon carbide devices according to this disclosure. Figure 3 A schematic diagram of the silicon carbide main circuit topology of a composite resonant power supply and distribution system based on silicon carbide devices according to this disclosure; Figure 4 This is a schematic diagram of the timing logic of an LLC resonant converter in a composite resonant power supply and distribution system based on silicon carbide devices according to this disclosure. Detailed Implementation

[0024] The present application / disclosure will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application / disclosure and are not intended to limit the scope of the present application / disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present application / disclosure are shown in the accompanying drawings, not the entire structure.

[0025] Example 1 Figure 1 For the architecture diagram of the composite resonant power supply and distribution system based on silicon carbide devices according to this disclosure, as follows: Figure 1 As shown, this embodiment provides a composite resonant power supply and distribution system based on silicon carbide devices, mainly designed for high-frequency, high-power applications, particularly suitable for situations with drastic load characteristic changes, such as power supply units for communication base stations or data center cabinets. It includes a bridge resonant topology module 10, a dynamic parameter tuning module 20, a blockchain verification module 30, a magnetically integrated composite insulation transformer 40, and a multi-level digital twin monitoring module 50. The units are connected via a high-speed communication bus, collaboratively completing efficient conversion and control from power input to output. Bridge resonant topology module 10 adopts a full-bridge LLC resonant topology structure. For example... Figure 3As shown, its primary side consists of a full-bridge circuit composed of four silicon carbide MOSFET switches, labeled Q1, Q2, Q3, and Q4. Q1 and Q3 form one bridge arm, and Q2 and Q4 form the other. Silicon carbide devices were chosen because of their wide bandgap characteristics, which allow them to withstand higher switching frequencies and higher voltage stresses, while their on-resistance and switching losses are much lower than those of traditional silicon-based IGBTs or MOSFETs.

[0026] In a preferred embodiment of this disclosure, the bridge resonant topology module 10 has the following topology. In applications with higher input voltages and stricter requirements for device voltage stress, a three-level active midpoint clamping topology is used on the primary side instead of a full-bridge structure. This topology, through the series connection of multiple switches and the cooperation of clamping diodes, ensures that the voltage stress borne by each switch is only half of the input voltage. This allows the use of silicon carbide devices with lower voltage ratings but better conduction characteristics to handle high input voltages, further reducing conduction losses and improving system efficiency. Simultaneously, the output voltage waveform of the three-level topology has lower harmonic content, which helps to reduce the size of the filter components.

[0027] The output of the full-bridge circuit is connected to a resonant network. This resonant network consists of a resonant inductor Lr and a resonant capacitor Cr connected in series. It should be noted that the resonant inductor Lr is not an independent physical inductor; its inductance value is at least partially derived from the primary leakage inductance of the magnetically integrated composite insulated transformer. This design helps reduce the number of independent magnetic components, thus lowering system size and cost. One end of the resonant network is connected to the same-name terminal of the primary winding of transformer T, and the other end is connected to the opposite-name terminal of the primary winding, forming a complete series resonant circuit.

[0028] A synchronous rectification circuit is configured on the secondary side of transformer T. This circuit consists of a full-bridge rectifier structure composed of four switching transistors S1, S2, S3, and S4. Compared with traditional diode rectification, synchronous rectification uses MOSFETs with low on-resistance as the rectifier device, which can significantly reduce the on-state voltage drop and rectification loss, especially in applications with high output current. A voltage-dividing resonant capacitor Cr2 is also connected in parallel on the secondary side. This capacitor, together with the leakage inductance of the transformer secondary winding and the parasitic parameters of the rectifier circuit, forms an auxiliary resonant network on the secondary side, which is used to further optimize the output voltage waveform and reduce voltage spikes.

[0029] In one embodiment, the bridge resonant topology module 10 further includes a real-time control unit connected to the control terminal of the blockchain verification module 30 and the bridge resonant topology module 10. The control unit generates a drive pulse according to the verified adjustment command to control the turn-on and turn-off timing of multiple silicon carbide MOSFET switches, thereby adjusting the resonant frequency and switching timing.

[0030] In this embodiment, the operation of the bridge resonant topology module 10 involves multiple timing stages, such as... Figure 4 As shown, the circuit includes the Q1 / Q3 turn-on stage, the dead time stage, the Q2 / Q4 turn-on stage, and the secondary synchronous rectification stage. During the Q1 / Q3 turn-on stage, current flows from the input source through Q1, the transformer primary side, the resonant network, and back to Q3, storing energy in the transformer core. The secondary rectifier diodes are off, and energy has not yet been transferred to the load. During the dead time stage, all primary switches are off, and the resonant current charges and discharges the junction capacitance of the switches, creating conditions for the zero-voltage turn-on of the switches in the next stage. At this time, the secondary synchronous rectifier diodes turn on, releasing the energy stored in the core to the load. The Q2 / Q4 turn-on stage is another half-cycle of energy storage. The secondary synchronous rectification stage is a freewheeling stage where the secondary side operates independently. Through the cyclical execution of these four stages, efficient energy transfer from input to output is achieved. The dead time setting is crucial; it not only prevents shoot-through short circuits between the upper and lower bridge arms but also achieves zero-voltage turn-on of the primary switches through the resonant current, significantly reducing switching losses.

[0031] The hardware carrier of the dynamic parameter tuning module 20 is a high-performance mixed signal processor, which integrates a programmable logic array and a processing core, and can simultaneously handle high-speed signal acquisition and complex algorithm calculations.

[0032] In this embodiment, the dynamic parameter tuning module 20 includes a data acquisition unit 210, a neural network prediction unit 220, and an instruction execution unit 230. The data acquisition unit 210 acquires real-time operational data through an IoT sensor array distributed across key nodes of the system. Sensor nodes are deployed at the transformer output, load connection points, and key nodes of the resonant network, acquiring electrical parameters including, but not limited to, instantaneous values ​​of input and output voltage and current. Based on this raw sampled data, the data acquisition unit 210 internally calculates characteristic parameters characterizing the load state using digital signal processing algorithms, primarily including current ripple rate, total harmonic distortion rate of voltage, and power factor. Data acquisition is performed at a high frequency to ensure the capture of details of transient load changes. The acquired raw data is transmitted to the data preprocessing unit via an Industrial Internet of Things (IIoT) protocol. In this unit, outliers are removed through data cleaning, and sliding window filtering is used for smoothing to suppress measurement noise. Subsequently, linear normalization is used to map each characteristic parameter to a unified numerical range. Finally, a feature extraction algorithm extracts the time-frequency features of the signal, and principal component analysis is used to reduce the dimensionality of the feature vector, preserving key information while reducing the computational complexity of the subsequent neural network model.

[0033] In a preferred embodiment of this disclosure, the data acquisition unit 210 employs a communication protocol and data transmission method. Wireless communication can replace wired industrial IoT protocols. For example, a low-power wireless network can be deployed between sensor nodes and edge computing nodes. Sensor data is wirelessly aggregated to a gateway, which then forwards it to the dynamic parameter tuning module via fiber optic cable or high-speed Ethernet. This approach simplifies field cabling, improves deployment flexibility, and is particularly suitable for retrofit projects or space-constrained deployment environments.

[0034] As an example, the data acquisition frequency is 10kHz, using a high-precision ADC (16-bit resolution, sampling rate of 1MS / s) to ensure accurate capture of transient load changes. The acquired raw data is stored in CSV format and uploaded to the edge computing node for preprocessing via the MQTT protocol. Data cleaning includes removing outliers and noise interference, and smoothing using a sliding window mean filter (window length of 50 sampling points); normalization processing includes mapping CRR, THD, and PF to the [0,1] interval, using a linear normalization formula:

[0035] The neural network prediction unit 220 incorporates a pre-trained deep learning network model. In this embodiment, a Long Short-Term Memory (LSTM) network architecture is used to handle time series prediction problems. The network's input layer receives a pre-processed and dimensionality-reduced load feature vector, which contains load state information from the current time and several previous time points. The network contains multiple hidden layers, each composed of multiple LSTM units. These units, through their unique gating mechanism, can learn the long- and short-term dependencies of load changes. The network's output layer is a single node, and the output value is the predicted optimal resonant frequency. The model's training is completed offline. Using a large amount of historical operating data, the resonant frequency at which the actual operating efficiency is optimal is used as a label. The backpropagation algorithm and mean squared error loss function are used to optimize the network parameters, enabling the model to accurately predict the resonant frequency value that maximizes system efficiency based on the input load features. When the system is running online, this module receives the load feature vector from the data acquisition module in real time and calculates the target resonant frequency under the current operating conditions.

[0036] As an example, the structure of a dynamically tuned model based on a Long Short-Term Memory (LSTM) network is as follows: Input layer: receives a preprocessed load feature vector with dimensions (1, n), where n is the number of features (e.g., CRR, THD, PF, etc.); Hidden layer: contains two LSTM layers, each with 64 units, and uses the tanh activation function; Fully connected layer: maps the LSTM output to the predicted resonant frequency using the ReLU activation function; Output layer: outputs the predicted resonant frequency f.r The unit is kHz, and the range is [400, 600] kHz.

[0037] The model is trained using the backpropagation algorithm, and the loss function is the mean squared error (MSE).

[0038] in, The actual measured value is N, and the total number of samples is N.

[0039] In a preferred embodiment of this disclosure, considering that the load feature data may have more complex long-term dependencies, a Transformer model based on a self-attention mechanism is used instead of an LSTM model. The Transformer model, through its multi-head attention mechanism, can more effectively capture dependencies between distant locations in the input sequence, and its prediction accuracy may be superior for load data with complex periodic or abrupt changes.

[0040] The instruction execution unit 230 converts the target frequency output by the neural network prediction unit 220 into an actual switching control signal. Specifically, the instruction execution unit 230 includes a frequency tracking controller and a drive circuit. The frequency tracking controller receives the target frequency value and compares it with the current actual operating frequency to calculate the frequency deviation. Based on this deviation, the frequency tracking controller uses a proportional-integral-derivative (PID) control algorithm to calculate the required frequency adjustment and generate a corresponding switching timing signal. This timing signal, after being amplified by the drive circuit, is applied to the gate of the silicon carbide MOSFET to control its turn-on and turn-off times, thereby adjusting the operating frequency of the entire resonant converter unit to track the target frequency and quickly respond to rapid load changes.

[0041] In this embodiment, the training process of the LSTM model can be specifically executed as follows: Input data: Normalized load eigenvector; Output target: Optimal resonant frequency f at the corresponding time. r Training process: The Adam optimizer was used, with a learning rate of 0.001, a batch size of 32, and 100 training rounds.

[0042] The model predictions are converted into actual frequency values ​​using the following formula:

[0043] in, .

[0044] The blockchain verification module 30 is connected between the neural network prediction unit 220 and the instruction execution unit 230 of the dynamic parameter tuning module 20, and is used to securely verify the adjustment instructions to be issued. This unit includes multiple verification nodes, which can be deployed on edge computing devices to form a distributed small blockchain network.

[0045] After the neural network prediction unit 220 generates an adjustment instruction containing the target resonant frequency and an execution timestamp, the instruction is first sent to the blockchain verification module 30. Upon receiving the instruction, any node in the unit immediately performs a hash operation on the instruction content, generating a fixed-length hash value. Subsequently, this hash value, along with the instruction itself, is broadcast among all nodes in the network. Each node independently performs a hash operation on the received instruction and compares its calculated hash value with the received hash value to verify whether the instruction has been tampered with during transmission.

[0046] During the verification process, each node coordinates based on a preset consensus algorithm. This embodiment employs an authority proof mechanism, where several authoritative nodes with verification permissions are pre-defined in the network. Consensus is considered achieved when all or a majority of authoritative nodes agree on the hash value of the instruction. After consensus is reached, the master node selected in the current round digitally signs the original instruction and sends the signed instruction back to the instruction execution unit 230. Only instructions verified by digital signature are accepted and executed by the instruction execution unit. Any instruction that fails consensus verification or signature verification is considered invalid, discarded by the system, and logged. This mechanism effectively prevents the execution of erroneous instructions due to network attacks, communication interference, or system failures, ensuring the security and reliability of the system.

[0047] As an example, after the adjustment instructions, including the target resonant frequency and execution timestamp, are generated, they need to be verified by blockchain nodes. A lightweight blockchain architecture is used, with nodes communicating via RabbitMQ message queues. The verification process can be specifically executed as follows: (1) Instruction generation: The adjustment instructions are output by the LSTM model, which include the target frequency f_r and the execution timestamp; (2) Hash generation: Perform SHA-256 hash processing on the instruction content; (3) Node verification: Multiple blockchain nodes perform consistency verification on the hash value; (4) Consensus Reached: The PoA mechanism is adopted, and authorized nodes sign and confirm the instructions to ensure their legality; (5) Instruction issuance: After verification, the instruction is sent to the sub-microsecond dynamic response execution unit.

[0048] The magnetically integrated composite insulation transformer 40 is used to achieve high-frequency and miniaturized systems. The magnetic core of the magnetically integrated composite insulation transformer 40 is made of soft magnetic materials with excellent high-frequency characteristics, such as nanocrystalline or amorphous alloys. These materials have low core losses at high frequencies and can support operating frequencies of hundreds of kilohertz.

[0049] In this embodiment, the magnetic core is designed as a three-dimensional structure, comprising a central post and side posts. The primary and secondary windings are wound around the central post, employing a multi-layered, interleaved winding method to enhance coupling and reduce leakage inductance. However, in this design, leakage inductance is not completely eliminated; rather, its value is precisely controlled to function as a resonant inductor L. r A portion of the inductance participates in resonance. This is achieved by creating an air gap of a certain length on the center column. The length of the air gap is precisely calculated so that the leakage inductance of the transformer is exactly equal to or close to the design value of the required resonant inductance. In this way, the inductor element, which originally needed to be set separately, is integrated into the transformer, realizing the integrated design of the magnetic components.

[0050] Regarding insulation structure, insulation design is particularly important due to the high operating frequency and potential for high voltage stress in transformers. Composite insulation materials are placed between the primary and secondary windings, between the windings and the magnetic core, and between layers. These insulation materials are made of high-temperature resistant, high-dielectric-strength, and low-dielectric-loss thin films or fibers, forming a reliable insulation barrier through multi-layer composite construction. After the entire transformer winding is completed, it is placed in a mold for vacuum potting. The filler is made of epoxy resin with high thermal conductivity and high insulation. After curing, it forms a robust whole, which improves both mechanical strength and heat dissipation and insulation performance.

[0051] The multi-level digital twin monitoring module 50 builds a real-time monitoring platform based on the unified architecture (OPC UA) protocol, and interacts with the aforementioned units (100, 200, 300, 400) for data exchange. This unit collects operational data from each key node of the system in real time, including input and output voltage and current, temperature at each point, operating frequency, switch status, consensus status, etc.

[0052] In this embodiment, a digital twin model that fully corresponds to the physical system runs within the multi-level digital twin monitoring module 50. This model not only includes simulations of electrical characteristics but also integrates thermal and electromagnetic models, reflecting the system's true operating state from a multi-physics perspective. Driven by real-time data, the monitoring interface dynamically displays the system's 3D model, energy flow, key parameter curves, and health status assessment results. When certain parameters deviate from their normal range, or when the twin model predicts a potential failure at a future time, the system proactively issues warnings, prompting maintenance personnel to inspect or intervene. Furthermore, this unit also possesses historical data storage and trend analysis capabilities, providing data support for system optimization and maintenance decisions.

[0053] As an example, upon system startup, initialization is performed first, setting an initial resonant operating frequency, for example, 500kHz. Subsequently, the IoT sensor array begins collecting load characteristic data, including current ripple rate, harmonic distortion rate, and power factor, at a sampling frequency of 10kHz. The collected data is uploaded to the data acquisition unit 210 of the dynamic parameter tuning module 20 via the Industrial IoT protocol, where it undergoes a series of preprocessing steps, including cleaning, filtering, normalization, and dimensionality reduction.

[0054] The processed load feature vector is input into the LSTM model of the neural network prediction unit 220. After forward computation, the model outputs the predicted optimal resonant frequency value under the current load condition, for example, a predicted value of 520kHz. This frequency value, together with information such as the timestamp, constitutes an adjustment command.

[0055] The adjustment command is sent to the blockchain verification module 30. This unit uses a Proof-of-Agent (PoA) mechanism, where multiple authorized nodes perform hash comparisons on the command content. If they match, consensus is reached; otherwise, the command is discarded. After successful verification, the master node digitally signs the command and returns the signed command.

[0056] After receiving the verified and signed adjustment command, the instruction execution unit 230 extracts the target frequency value. The frequency tracking controller compares the target frequency with the current actual frequency, calculates the adjustment amount using a PID control algorithm, and adjusts the switching pulse frequency output to the silicon carbide MOSFET in real time. The drive circuit amplifies the control signal and drives the switching transistor, enabling the entire resonant converter unit to accurately track the target value at its operating frequency.

[0057] In this embodiment, a PID control algorithm is used to perform closed-loop regulation of the resonant frequency, and its control equation is as follows:

[0058] in, These are the proportional, integral, and derivative coefficients, which are tuned according to the system response characteristics.

[0059] Meanwhile, the magnetically integrated composite insulation transformer 40 operates stably at high frequencies. Its internally integrated resonant inductor and resonant capacitor work together to ensure efficient energy transfer. The synchronous rectification circuit on the transformer's secondary side, in conjunction with the secondary resonant capacitor, further optimizes the output characteristics.

[0060] Throughout the operation, the multi-level digital twin monitoring module 50 continuously collects data from each module, drives the virtual model to run synchronously, displays the system status in real time, and performs health assessments. If any abnormal trend is detected, an early warning message is immediately issued.

[0061] In a preferred embodiment of this disclosure, the LSTM model can be replaced with a Transformer architecture to improve the ability to extract long sequence features; 5G communication can also be used to replace the MQTT protocol to improve data transmission rate and reliability; in terms of deployment, some computing tasks can be migrated to the cloud to achieve distributed intelligent control.

[0062] In summary, this system achieves a high-frequency, high-efficiency, and highly adaptable power supply and distribution solution through deep integration of hardware and software, making it suitable for scenarios with high power quality requirements, such as data centers and electric vehicle charging stations.

[0063] This disclosure proposes a composite resonant power supply and distribution system based on silicon carbide (SiC) devices. It employs SiC MOSFETs to construct the resonant topology and combines them with a dynamic tuning algorithm to achieve a conversion efficiency of 98.2% ± 0.3% over a wide load range (10%-100%), significantly higher than traditional IGBT solutions (typically ≤95%). It is suitable for high-energy-consumption scenarios (such as data centers and 5G base stations), saving tens of thousands of kWh of electricity per site annually, directly reducing operating costs. The magnetically integrated transformer (PITv2) supports operating frequencies ≥500kHz, reduces volume by more than 40%, and achieves core losses as low as 15mW / cm³ (traditional solutions >50mW / cm³). It meets the needs of space-sensitive scenarios such as edge computing devices and on-board charging piles, improving device integration and deployment flexibility.

[0064] Example 2 Figure 2 To apply the composite resonant power supply and distribution control method based on silicon carbide devices disclosed herein to the system described above, the following will refer to... Figure 2 The present invention provides a detailed description of the composite resonant power supply and distribution control method based on silicon carbide devices.

[0065] In step 101, load characteristic data is collected through an IoT sensor array.

[0066] In one embodiment, after the system starts up, an IoT sensor array deployed at key locations in the power distribution system begins to operate. Sensor nodes are distributed at the transformer output, load connection points, and key nodes of the resonant network, acquiring instantaneous voltage and current values ​​in real time at a set sampling frequency. Based on these raw waveform data, several characteristic parameters reflecting the current load characteristics are calculated in real time using digital signal processing technology, primarily including current ripple rate, total harmonic distortion (THD) of voltage, and power factor.

[0067] In one embodiment, the acquired raw data is first cleaned to remove outliers caused by momentary sensor malfunctions or strong electromagnetic interference. Then, a sliding window mean filter is used to smooth the data to suppress high-frequency measurement noise. The cleaned and filtered data is then mapped to a uniform numerical range using linear normalization to eliminate the influence of different feature parameters' dimensions. To further improve the input quality of subsequent prediction models, wavelet transform is used to perform time-frequency analysis on the normalized signal, extracting its time-frequency domain features to form a high-dimensional feature vector. Finally, principal component analysis is used to reduce the dimensionality of this high-dimensional feature vector, retaining information that can explain the main variance of the original data, resulting in a low-dimensional and information-rich load feature vector, which serves as the input to the subsequent neural network model.

[0068] In step 102, the dynamic parameter tuning module processes the load characteristic data based on the neural network model to generate the adjustment command for the optimal resonant frequency.

[0069] In one embodiment, the load feature vector obtained in step 101 is input into a pre-trained Long Short-Term Memory (LSTM) network model. This LSTM model includes an input layer, multiple LSTM hidden layers, a fully connected layer, and an output layer. The input layer receives the load feature vector; the LSTM hidden layers learn the temporal dependencies of the load sequence through their internal forget gate, input gate, and output gate structures; the fully connected layer maps the output of the LSTM layer to the target dimension; and the output layer outputs a numerical value, namely the predicted optimal resonant frequency under the current operating conditions. This model has been trained offline using a large amount of historical operating data, with the training objective being to minimize the mean square error between the predicted frequency and the actual optimal frequency. During online operation, the model calculates the target resonant frequency in real time using forward propagation.

[0070] In step 103, the adjustment command is verified through consensus by the blockchain verification module. After successful verification, the adjustment command is sent to the real-time control unit.

[0071] In one embodiment, the target resonant frequency predicted in step 102 is combined with information such as the current timestamp and device identifier to form a complete adjustment command. A hash operation is performed on the complete content of this command to generate a fixed-length hash value. This hash value serves as the digital fingerprint of the command for subsequent integrity verification. The original command and its hash value are sent together to the blockchain consensus verification network.

[0072] In one embodiment, the blockchain consensus verification network consists of multiple verification nodes deployed on edge computing nodes. Upon receiving an adjustment instruction and its hash value, each node independently re-hashes the instruction content and compares the result with the received hash value to verify whether the instruction has been tampered with during transmission. Subsequently, each node votes on the validity of the instruction based on a preset consensus algorithm (e.g., a proof-of-authority mechanism). Consensus is considered achieved when more than two-thirds of the authorized nodes in the network reach a consensus. After consensus is reached, the master node of the current round digitally signs the original instruction and returns the signed instruction to the control unit. If consensus is not reached, the instruction is considered invalid and discarded, and the system simultaneously records an anomaly log.

[0073] In step 104, the real-time control unit drives the silicon carbide MOSFET in the bridge resonant topology unit according to the adjustment command, and adjusts its switching frequency and resonant network parameters to achieve dynamic response to load changes.

[0074] In one embodiment, after receiving a digitally signed adjustment command, the dynamic tuning control module first verifies the validity of the signature. If the verification is successful, the target resonant frequency value is parsed from the command. The frequency tracking controller reads the current system's actual operating frequency and calculates the deviation e(t) between the target frequency and the actual frequency.

[0075] In one embodiment, the frequency deviation e(t) is input to a proportional-integral-derivative (PID) controller. The controller calculates the control quantity u(t) according to a PID control algorithm based on pre-tuned proportional, integral, and derivative coefficients. This control quantity corresponds to the adjustment step size of the switching frequency. The controller converts the adjustment step size into specific switching timing parameters and outputs them to the drive circuit.

[0076] In one embodiment, the driving circuit generates driving pulses with corresponding frequencies and duty cycles based on the received switching timing parameters, which are then applied to the gates of each MOSFET in the silicon carbide-based composite resonant converter unit. By precisely controlling the turn-on and turn-off times of the switching transistors, the operating frequency of the entire resonant converter unit gradually approaches and eventually stabilizes at the target resonant frequency. This closed-loop adjustment process continues, ensuring that the system tracks the optimal frequency point predicted by the neural network in real time, thereby achieving dynamic matching between the resonant state and the load characteristics.

[0077] Step S105: Real-time monitoring and simulation of the system's operating status are performed through a multi-level digital twin monitoring module.

[0078] This disclosure provides a composite resonant power supply and distribution control method based on silicon carbide (SiC) devices. It employs SiC MOSFETs to construct the resonant topology and combines it with a dynamic tuning algorithm to achieve a conversion efficiency of 98.2% ± 0.3% over a wide load range (10%-100%), significantly higher than traditional IGBT solutions (typically ≤95%). It is suitable for high-energy-consumption scenarios (such as data centers and 5G base stations), saving tens of thousands of kWh of electricity per site annually, directly reducing operating costs. The magnetically integrated transformer (PITv2) supports operating frequencies ≥500kHz, reduces volume by more than 40%, and achieves core losses as low as 15mW / cm³ (traditional solutions >50mW / cm³). It meets the needs of space-sensitive scenarios such as edge computing devices and on-board charging piles, improving device integration and deployment flexibility.

[0079] In a third aspect, embodiments of this disclosure provide an electronic device including a memory and a processor, wherein the memory stores a program that runs on the processor, and the processor executes the steps of the composite resonant power supply and distribution control method based on silicon carbide devices as described in the first aspect when running the program.

[0080] In a fourth aspect, embodiments of this disclosure provide a computer-readable storage medium having computer instructions stored thereon, which, when executed, perform the steps of the composite resonant power supply and distribution control method based on silicon carbide devices as described in the first aspect.

Claims

1. A composite resonant power supply and distribution system based on silicon carbide devices, characterized in that, include: The bridge resonant topology module, composed of silicon carbide MOSFETs, is used to perform high-frequency power conversion; A dynamic parameter tuning module, connected to the bridge resonant topology module, includes an IoT sensor array and a neural network model. The IoT sensor array is used to collect load characteristic data, and the neural network model is used to predict and generate adjustment commands for the optimal resonant frequency based on the load characteristic data. The blockchain verification module is communicatively connected to the dynamic parameter tuning module and is used to perform consensus verification on the adjustment command and output the verified adjustment command after the verification is successful. A magnetically integrated composite insulated transformer, operating at high frequency, is coupled to the bridge resonant topology module for voltage transformation and electrical isolation; and The multi-level digital twin monitoring module is communicatively connected to the bridge resonant topology module, dynamic parameter tuning module, blockchain consensus verification system, and magnetically integrated composite insulation transformer, and is used for real-time monitoring and simulation analysis of the system's operating status.

2. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 1, characterized in that, The bridge resonant topology module includes: A full-bridge circuit composed of silicon carbide MOSFET switches; The series resonant circuit located on the primary side includes a primary-side leakage inductance and a primary-side resonant capacitor connected in series. The voltage divider resonant circuit located on the secondary side includes a secondary-side resonant inductor and a secondary-side resonant capacitor connected in series. The synchronous rectification circuit connected to the full-bridge circuit is used to perform synchronous rectification on the secondary side.

3. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 2, characterized in that, The bridge resonant topology module also includes a real-time control unit, which is connected to the control terminal of the blockchain verification module and the bridge resonant topology module. It is used to generate driving pulses according to the verified adjustment instructions to control the turn-on and turn-off timing of the silicon carbide MOSFET switch, thereby adjusting the resonant frequency and switching timing.

4. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 3, characterized in that, The real-time control unit employs a proportional-integral-derivative (PID) control algorithm to perform closed-loop adjustment of the resonant frequency according to the verified adjustment command.

5. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 3 or 4, characterized in that, The real-time control unit is configured to execute a timing control strategy for the resonant converter, the timing control strategy comprising sequentially looping: In the first conduction stage, the first pair of switching transistors of the full-bridge circuit are turned on, so that energy is stored in the transformer core. During the dead time phase, all primary-side switches are turned off, and the synchronous rectifier circuit is turned on, so that the energy stored in the transformer core is released to the load. In the second conduction phase, the second pair of switching transistors of the full-bridge circuit are turned on, so that energy is stored in the transformer core again. During the secondary synchronous rectification stage, all primary-side switches are turned off, and the synchronous rectification circuit is turned on, so that the energy stored in the transformer core is released to the load again.

6. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 1, characterized in that, The dynamic parameter tuning module includes: An Internet of Things (IoT) sensor array is used to collect load characteristic data, including at least current ripple rate, total harmonic distortion rate, and power factor. A neural network prediction unit, connected to the sensor array, has a built-in long short-term memory network model for predicting the optimal resonant frequency based on the load characteristic data. The instruction generation unit, connected to the neural network prediction unit, is used to generate adjustment instructions containing the target frequency based on the prediction results.

7. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 1, characterized in that, The dynamic parameter tuning module also includes a data preprocessing unit, which is used to clean, normalize and reduce the dimensionality of the collected load feature data, and input the processed data into the neural network model.

8. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 1 or 6, characterized in that, The blockchain verification module adopts an authoritative proof consensus mechanism, which uses multiple authorized nodes to perform hash consistency verification on the content of the adjustment instruction, and signs and confirms the instruction after consensus is reached.

9. The composite resonant power supply and distribution system based on silicon carbide devices according to claim 1, characterized in that, Furthermore, the multi-level digital twin monitoring module is connected to the bridge resonant topology module, the dynamic parameter tuning module, and the blockchain verification module to collect system operating status data in real time, construct a digital twin model, and perform status simulation and fault early warning.

10. A composite resonant power supply and distribution control method based on silicon carbide devices, applied to the composite resonant power supply and distribution system based on silicon carbide devices as described in claims 1-9, characterized in that, Includes the following steps: S1: Collect load characteristic data through an IoT sensor array; S2: The dynamic parameter tuning module processes the load characteristic data based on a neural network model to generate an adjustment command for the optimal resonant frequency; S3: The adjustment command is verified through consensus by the blockchain verification module. After the verification is successful, the adjustment command is sent to the real-time control unit. S4: The real-time control unit drives the silicon carbide MOSFET in the bridge resonant topology unit according to the adjustment command, and adjusts its switching frequency and resonant network parameters to achieve dynamic response to load changes. S5: Real-time monitoring and simulation of system operation status through multi-level digital twin monitoring modules.

Images