A modular liquid-cooled supercharged stack design method supporting power module plug-and-play and intelligent temperature control

Abstract

The application belongs to the technical field of super-charging of electric vehicles. A design method of a modular liquid-cooled super-charging stack supporting power modules plug-and-play and intelligent temperature control, characterized in that it comprises the following steps: Step 1, overall modular structure design of the super-charging stack: the overall super-charging stack adopts layered modular design, and is divided into a main control layer, a power layer, a liquid cooling layer and a basic support layer from top to bottom, and the layers are connected through standardized interfaces; Step 2, power module plug-and-play design: the power module plug-and-play design includes four parts of mechanical interface standardization, electrical interface standardization, liquid cooling interface standardization and intelligent identification and configuration mechanism; Step 3, intelligent temperature control method based on Transformer-GNN: combining the time sequence feature extraction advantage of the Transformer model and the spatial correlation capture advantage of the GNN model. The application realizes quick replacement and expansion of the power module, precise adaptive adjustment of the temperature control system, and improves the operation stability, operation and maintenance efficiency and energy consumption economy of the super-charging stack.

Description

Technical Field

[0001] This invention belongs to the field of electric vehicle supercharging technology, specifically relating to a modular liquid-cooled supercharging stack design method that supports plug-and-play power modules and intelligent temperature control. Background Technology

[0002] With the rapid development of the electric vehicle industry, users' demand for charging speed is increasing. Modular liquid-cooled supercharger stacks have become the mainstream technology in the supercharging field due to their advantages such as high power density, excellent heat dissipation efficiency, and flexible expansion. However, existing modular liquid-cooled supercharger stacks still have two major pain points: First, the replacement and expansion of power modules rely on professional personnel for disassembly, wiring, and debugging, which cannot achieve true plug-and-play functionality, resulting in low operation and maintenance efficiency, long downtime, and difficulty in adapting to the needs of large-scale operation of supercharging stations. Second, temperature control systems mostly adopt traditional PID control and single time-series predictive models (such as LSTM), which can only passively adjust based on the real-time temperature of a single module. They cannot fully capture the thermal coupling relationship (spatial characteristics) and dynamic temperature change pattern (time-series characteristics) between multiple power modules, resulting in low temperature control accuracy, slow response speed, and easy occurrence of local overheating or energy waste, which seriously affects the operational stability, service life, and energy economy of the supercharger stack.

[0003] In existing technologies, temperature control in liquid-cooled supercharged reactors largely relies on Kalman filtering for temperature prediction or a single algorithm for simple adjustment. This results in inaccurate dynamic model construction and an inability to adapt to collaborative heat dissipation across multiple modules. Furthermore, the modular design of power modules often emphasizes structural assembly, lacking standardized plug-and-play interfaces and intelligent identification mechanisms, hindering rapid module replacement and flexible expansion. In addition, while Transformer models excel in temporal feature extraction and GNNs (Graph Neural Networks) are outstanding in handling spatial correlation features, there is currently no technical solution to apply a Transformer-GNN fusion model to the temperature control of modular liquid-cooled supercharged reactors. This fails to balance the temporal dynamics of temperature with the spatial coupling of multiple modules, making it difficult to meet the precise temperature control requirements of high power density supercharged reactors. Therefore, developing a modular liquid-cooled supercharged reactor design method that supports plug-and-play power modules and achieves high-precision intelligent temperature control has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a modular liquid-cooled supercharger design method that supports plug-and-play power modules and intelligent temperature control, solving the problems of existing supercharger power modules not being plug-and-play, low temperature control accuracy, slow response, and insufficient heat coupling processing of multiple modules. This enables rapid replacement and expansion of power modules, precise adaptive adjustment of the temperature control system, and improves the operational stability, maintenance efficiency, and energy economy of the supercharger.

[0005] To achieve the above objectives, the technical solution adopted by this invention is a modular liquid-cooled supercharged reactor design method that supports plug-and-play power modules and intelligent temperature control. The method is characterized by three core components: an overall modular structure design for the supercharged reactor, a plug-and-play power module design, and an intelligent temperature control method based on Transformer-GNN. The method includes the following steps: Step 1, Overall modular structure design of the supercharged reactor: The supercharger adopts a layered modular design, which is divided into a main control layer, a power layer, a liquid cooling layer and a basic support layer from top to bottom. Each layer is connected through a standardized interface to achieve independent operation and maintenance and flexible expansion. 1.1) Main Control Layer: Integrates the main controller, communication module, and data acquisition module. The main controller adopts an industrial-grade MCU, which is responsible for receiving the operating data of each power module and the status data of the liquid cooling system, and executing power distribution, temperature control command issuance, and plug-and-play identification control. The communication module adopts Ethernet + CAN bus dual-mode communication to ensure the real-time performance and reliability of data transmission. The data acquisition module is responsible for collecting the temperature, current, and voltage data of each power module and the coolant temperature, flow rate, and pressure data of the liquid cooling system. 1.2) Power layer: It consists of several standardized power modules (2-20 in total). Each power module uses silicon carbide (SiC) MOSFET devices. The power modules are connected in parallel to achieve flexible power distribution. The power layer has reserved standardized mounting slots to adapt to power modules of different specifications and meet different charging power requirements. 1.3) Liquid Cooling Layer: A distributed liquid cooling loop design is adopted, including a main coolant tank, a variable frequency circulating pump, a plate heat exchanger, a distribution manifold, and several modular liquid cooling interfaces (2-20 in total). The main coolant tank is filled with nanofluid coolant (the nanofluid coolant contains silica nanoparticles, with the amount of silica nanoparticles accounting for 0.5% of the total mass (balancing thermal conductivity gain, circulation energy consumption, and cost, resulting in the highest cost-effectiveness), the freezing point is reduced to -40℃, and the thermal conductivity is increased by 40% compared to traditional ethylene glycol solutions). The variable frequency circulating pump is responsible for driving the coolant circulation, the plate heat exchanger realizes the heat exchange between the coolant and the external environment, and the distribution manifold evenly distributes the coolant to the liquid cooling channels of each power module to achieve synchronous heat dissipation of multiple modules. 1.4) Basic support layer: A steel structure frame is adopted to provide installation support for each layer of modules. The frame integrates heat dissipation and ventilation ports and shock absorption and buffer structure to reduce vibration and noise during the operation of the supercharged reactor and improve the stability of equipment operation. Step 2, Plug and Play Design for Power Modules: The plug-and-play design of the power modules comprises four main parts: standardized mechanical interfaces, standardized electrical interfaces, standardized liquid cooling interfaces, and an intelligent identification and configuration mechanism, enabling rapid, non-stop replacement and capacity expansion of power modules. 2.1) Standardized mechanical interface: Each power module uses a uniform mechanical housing. Standardized buckles and guide grooves are set on both sides of the housing to precisely match the installation slots of the power layer. The buckles adopt a quick-locking structure, which can complete the installation and removal of the module without tools. The guide grooves ensure that the interface is aligned during the module insertion and removal process to avoid damage to the interface. 2.2) Standardized electrical interface: The electrical interface of the power module adopts a standardized hot-swappable connector, which integrates the power interface, signal interface and power supply interface. The power interface is responsible for transmitting charging power, the signal interface is responsible for data communication with the main control layer, and the power supply interface provides power to the control circuit of the power module. The connector adopts an anti-misinsertion design to ensure that the positive and negative terminals are not reversed. It also has waterproof and dustproof functions, and the protection level reaches IP67. 2.3) Standardization of liquid cooling interface: The liquid cooling interface of the power module adopts a quick-connect sealing connector, which matches the modular liquid cooling interface of the liquid cooling layer. The connector has a built-in sealing gasket, which automatically seals during insertion and removal to prevent coolant leakage. At the same time, the interface adopts a self-locking structure to ensure a firm connection and adapt to the coolant circulation pressure requirements. 2.4) Intelligent Identification and Configuration Mechanism: Each power module has a built-in unique identification (ID) chip that stores information such as the module's power specifications, factory parameters, and operating history. When a power module is inserted into the installation slot, the main control layer's data acquisition module reads the module ID information through the signal interface. The main controller automatically identifies the module specifications based on the ID information and completes the adaptive configuration of power allocation parameters and communication parameters without manual debugging, achieving plug-and-play functionality for the power module. If the inserted module is faulty or mismatched in specifications, the main control layer will issue an alarm signal and prohibit the module from being put into operation, ensuring the overall safe operation of the supercharger stack. Step 3, Intelligent temperature control method based on Transformer-GNN: The intelligent temperature control method based on Transformer-GNN combines the temporal feature extraction advantages of the Transformer model with the spatial correlation capture advantages of the GNN model to achieve precise adaptive temperature control of the liquid cooling system. Specifically, it includes five stages: data acquisition, feature fusion, temperature prediction, temperature control adjustment, and closed-loop feedback. 3.1) Data Acquisition: The supercharged reactor's operating data is acquired in real time through the main control layer's data acquisition module, with an acquisition cycle of 100ms. The acquired data includes: real-time temperature of each power module, real-time output current and voltage of each power module, coolant inlet and outlet temperatures, circulation flow rate and system pressure of the liquid cooling system, and ambient temperature and humidity. The acquired data is preprocessed to remove outliers (using the 3σ criterion to remove data deviating from the normal range) and normalized. The min-max normalization method is used to map the data to the [0,1] interval to eliminate the influence of dimensions. The normalization formula is as follows: In the formula: The raw data collected; This is the historical minimum value for this type of data; This represents the historical maximum value for this type of data; The normalized data eventually yields a standardized input dataset. 3.2) Feature Fusion: Construct a Transformer-GNN fusion model to extract temporal and spatial features from the collected standardized data. 3.2.1) Temporal Feature Extraction: The Transformer encoder is used to extract the temperature time-series data and heat generation power time-series data of each power module; the historical temperature data and heat generation power data of each power module are arranged in time series to form a temporal feature vector, which is input into the self-attention mechanism of the Transformer encoder. The self-attention mechanism automatically captures the correlation between temperature and heat generation power at different time steps and extracts the temporal trend features of temperature changes. 3.2.2) Spatial Feature Extraction: A GNN model is used to extract the thermal coupling features between multiple power modules and the spatial correlation features of the liquid cooling circuit. Each power module is used as a node in the GNN model, and the heat conduction relationship between modules and the connection relationship of the liquid cooling circuit are used as edges between nodes to construct a graph structure. The GNN model uses graph convolutional layers to aggregate the features of each node with the features of adjacent nodes to capture the spatial thermal coupling features between multiple power modules and the spatial correlation features between coolant flow rate and the temperature of each module. The feature aggregation formula of the graph convolutional layer is as follows: In the formula: For the first Nodes after layer graph convolution (Eigenvector corresponding to a single power module); , For nodes , The set of adjacent nodes (corresponding to other modules that have thermal coupling with this power module or share the same liquid cooling branch); N is the number of training samples; , They are nodes , The number of adjacent nodes; For the first The weight matrix of layer graph convolution; For the first Layer nodes eigenvectors; For the first Layer bias terms; For the activation function, this invention uses the ReLU function; 3.2.3) Feature Fusion: The temporal feature vector extracted by the Transformer encoder and the spatial feature vector extracted by the GNN model are fused through a fully connected layer. An attention weight allocation mechanism is used to assign different weights to the temporal and spatial features. The attention weight allocation formula and the feature fusion formula are as follows: In the formula: For time-series feature weights; Spatial feature weights; Assess the importance of time-series features; Score the importance of spatial features; Temporal feature vectors extracted by the Transformer encoder; Spatial feature vectors extracted for the GNN model; The fused feature vector serves as the input for temperature prediction. The natural index value is used to score the importance of time-series features and is used to normalize the weight allocation through the Softmax mechanism. The natural index value for scoring the importance of spatial features is used to normalize the weight allocation through the Softmax mechanism; 3.3) Temperature Prediction: The fused feature vector is input into the prediction layer of the Transformer-GNN fusion model. The prediction layer uses a fully connected neural network to output the predicted temperature values ​​of each power module and the predicted temperature value of the coolant outlet of the liquid cooling system within the next 5 seconds. At the same time, the model predicts the temperature change trend under different temperature control strategies based on the changes in ambient temperature and module heat generation power, providing a basis for temperature control adjustment. The training process of the Transformer-GNN fusion model is as follows: Historical operating data of the supercharged heap under different operating conditions are collected to construct training and testing datasets. The training dataset is used for model parameter training, and the testing dataset is used for model performance verification. Mean squared error (MSE) is used as the loss function, and the model parameters are iteratively optimized using the gradient descent algorithm until the model's prediction error is less than a preset threshold (≤0.5℃), thus completing model training. The loss function formula is as follows: In the formula: This represents the mean squared error loss function value; This represents the number of training samples; For the first The true temperature value of each sample; For the first Temperature prediction values ​​for each sample; the trained model is deployed to the main controller of the main control layer, supporting online fine-tuning and continuously optimizing prediction accuracy based on long-term operating data of the supercharged reactor; 3.4) Temperature Control: Based on the predicted temperature value and the preset temperature threshold (the normal operating temperature range of the power module is 40℃~65℃, and the coolant outlet temperature threshold is 55℃), the main controller formulates an adaptive temperature control strategy and sends control commands to the actuators of the liquid cooling layer. The specific adjustment method is as follows: 3.4.1) When the temperature of a certain power module is predicted to exceed 65℃, the main controller controls the variable frequency circulating pump to increase its speed and increase the coolant circulation flow rate. At the same time, it adjusts the corresponding liquid cooling branch valve of the module to increase the coolant distribution of the module and accelerate heat dissipation. If the temperature prediction value exceeds 70℃ (overheat warning threshold), the output power of the module is reduced to avoid overheating and damage to the module. 3.4.2) When the predicted temperature of each power module is below 45℃ and the coolant outlet temperature is below 40℃, the main controller controls the variable frequency circulating pump to reduce the speed, reduce the coolant circulation flow, reduce the energy consumption of the liquid cooling system, and achieve energy-saving operation. 3.4.3) When the temperature difference between multiple power modules is predicted to exceed 5℃, the main controller adjusts the valve opening of the distributor manifold to adjust the coolant distribution of each module, balance the temperature of each module, and avoid local overheating or insufficient heat dissipation. 3.4.4) When the power module achieves plug-and-play functionality (insertion or removal), the Transformer-GNN fusion model automatically updates the graph structure (the relationship between the number of nodes and edges), quickly re-extracts spatial and temporal features, predicts new temperature distributions, and the main controller adjusts the operating parameters of the liquid cooling system in a timely manner to ensure that the temperature control system can quickly adapt to changes in the number of modules and maintain the stability of the supercharged reactor temperature. 3.5) Closed-loop feedback: The data acquisition module collects the supercharged reactor operating data (temperature of each module, coolant parameters, etc.) in real time after temperature control adjustment, and feeds the data back to the Transformer-GNN fusion model. The model corrects the temperature prediction value based on the feedback data, and the main controller adjusts the temperature control adjustment strategy based on the corrected prediction value, forming a closed-loop temperature control system of "acquisition-fusion-prediction-adjustment-feedback" to ensure long-term stable temperature control accuracy and adapt to the dynamic operating conditions of the supercharged reactor. Step 4, System Debugging and Operation Monitoring: 4.1) System Debugging: After the supercharger stack is assembled, plug-and-play and temperature control functions are debugged. During plug-and-play debugging, power modules of different specifications are inserted in sequence to verify the accuracy of module identification, the automation of parameter configuration, and the stability of operation. During temperature control function debugging, different operating conditions (full power output, partial module operation, and ambient temperature changes) are simulated to verify the temperature prediction accuracy and the effectiveness of temperature control adjustment of the Transformer-GNN fusion model, ensuring that all indicators meet the design requirements. 4.2) Operation Monitoring: During normal operation of the supercharged reactor, the main control layer monitors the operating status of each module, the working status of the liquid cooling system, and the temperature control effect in real time. The operating data is stored in the local storage module and uploaded to the cloud monitoring platform through the communication module. If a module failure, liquid cooling system abnormality, or temperature exceeding the standard occurs, the main control layer immediately issues an audible and visual alarm signal and uploads the alarm information to the cloud. At the same time, emergency measures are taken (such as disconnecting the faulty module and reducing the system power) to ensure the safe and stable operation of the supercharged reactor.

[0006] The beneficial effects of this invention are: 1. It solves the problems of existing supercharged reactor power modules being unable to be plug-and-play, having low temperature control accuracy, slow response, and insufficient multi-module thermal coupling processing, enabling rapid replacement and expansion of power modules, and precise adaptive adjustment of the temperature control system, thereby improving the operational stability, maintenance efficiency, and energy economy of the supercharged reactor. 2. By combining the temporal feature extraction capability of Transformer and the spatial correlation capture capability of GNN, it accurately predicts and adaptively adjusts the temperature and cooling system parameters of multiple modules, forming a closed-loop control system of "acquisition, fusion, prediction, adjustment, and feedback". This invention achieves flexible expansion and efficient operation and maintenance of power modules, while significantly improving temperature control accuracy, response speed, and system energy efficiency, ensuring the long-term stable operation of high-power-density supercharged reactors. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the overall modular structure of the supercharged reactor of the present invention.

[0008] Figure 2 This is a schematic diagram of the plug-and-play interface structure of the power module of the present invention.

[0009] Figure 3 This is a flowchart of the intelligent temperature control method based on Transformer-GNN of the present invention.

[0010] Figure 4 This is a schematic diagram of the Transformer-GNN fusion model structure of the present invention. Detailed Implementation

[0011] The present invention will be further described in detail below with reference to specific embodiments.

[0012] Example 1 This embodiment provides a modular liquid-cooled supercharged reactor design method that supports plug-and-play power modules and intelligent temperature control. The specific implementation steps are as follows: Step 1, Overall modular structure design of the supercharged reactor: The supercharger stack adopts a layered modular design. The main control layer uses an STM32H743VIT6 industrial-grade MCU as the main controller, and the communication module uses a DP83848 Ethernet chip and a TJA1050 CAN bus chip to achieve dual-mode communication. The power layer has 10 standardized mounting slots, and each power module uses SiC MOSFET devices with a rated power of 60kW. The modules are connected in parallel, and the maximum output power can reach 600kW. The liquid cooling layer uses a main coolant tank (50L volume), a TD32-160B variable frequency circulating pump, and a BR0.3 plate heat exchanger. The water distribution manifold adopts a 1-in-10-out structure. Each power module is equipped with an independent liquid cooling channel, and the coolant is a nanofluid coolant (with 0.5% silica nanoparticles added). The foundation support layer uses a Q235 steel structure frame with a height of 1.8m and a width of 1.2m, integrating ventilation openings and shock-absorbing rubber pads.

[0013] Step 2, Plug and Play Design for Power Modules: The mechanical interface adopts a standard 3U chassis specification, with quick-locking spring clips and a dovetail groove structure for precise insertion and removal. The electrical interface uses MX34012 series hot-swappable connectors, integrating 3 power interfaces (rated current 300A), 2 signal interfaces (CAN bus), and 1 power supply interface (12V). The connector features an anti-misinsertion boss design with an IP67 protection rating. The liquid cooling interface uses LLH series quick-connect sealing connectors with built-in fluororubber gaskets, a rated pressure of 1.6MPa, and ≥1000 insertion / removal cycles. Each power module has a built-in AT24C02 ID chip that stores information such as module ID, rated power, and manufacturing date. After insertion, the main control layer completes identification and parameter configuration within 100ms.

[0014] Step 3, Intelligent temperature control method based on Transformer-GNN: 3.1 Data Acquisition: The temperature of each power module was acquired using a DS18B20 temperature sensor (accuracy ±0.5℃), the module current and voltage were acquired using an ACS712 current sensor and an ACS758 voltage sensor, the coolant temperature was acquired using a PT100 sensor, and the coolant flow rate was acquired using an LWGY turbine flow meter. The acquisition period was 100ms. Abnormal data was removed using the 3σ criterion, and the data was mapped to the [0,1] interval using the min-max normalization method. 3.2 Feature Fusion: The Transformer encoder has 6 layers, 8 self-attention heads, and 128 hidden layer dimensions; the GNN model uses GCN (Graph Convolutional Network), with 3 graph convolutional layers, 64 node feature dimensions, and 32 edge feature dimensions; the fusion layer uses a fully connected layer, with initial attention weights set to 0.6 for temporal features and 0.4 for spatial features, dynamically adjusted according to the running status; 3.3 Temperature Prediction: The prediction layer uses a two-layer fully connected neural network with a sigmoid activation function for the output layer, predicting the temperature of each module and the coolant outlet temperature within the next 5 seconds. The model is trained using 10,000 sets of historical operating data (covering ambient temperatures of 30℃~60℃ and output power conditions of 0~600kW). The loss function is the mean squared error (MSE), the training iterations are 1000, and the learning rate is 0.001. After training, the prediction error is ≤0.5℃. The loss function formula and learning rate update formula are as follows: In the formula: This represents the number of training samples; For the first The true temperature value of each sample; For the first Predicted temperature values ​​for each sample; For the first The learning rate for each iteration; The initial learning rate ( Take 0.001). The learning rate decay coefficient (in this embodiment) (Take 0.98). This represents the number of iterations.

[0015] 3.4 Temperature Control Adjustment: The preset normal operating temperature range of the power module is 40℃~65℃, and the coolant outlet temperature threshold is 55℃; when the predicted module temperature is ≥65℃, the variable frequency circulating pump speed is increased from 1500r / min to 2500r / min, and the coolant flow rate is increased from 10L / min to 20L / min; when the module temperature is ≤45℃ and the coolant outlet temperature is ≤40℃, the pump speed is reduced to 1000r / min, and the flow rate is reduced to 5L / min; when the temperature difference between multiple modules is ≥5℃, the opening of the distributor manifold valve is adjusted to adjust the coolant distribution difference between each module to ≤2L / min. 3.5 Closed-loop feedback: The adjusted data is collected every 100ms and fed back to the model to correct the predicted value, dynamically adjusting the temperature control strategy to ensure that the temperature is stable within the preset range.

[0016] Step 4, System Debugging and Operation Monitoring: Plug and play debugging: When 5, 8, and 10 60kW power modules are inserted in sequence, the main control layer can complete the identification within 100ms, automatically configure the power allocation parameters, and the modules operate normally. There are no problems such as coolant leakage or poor electrical contact during the insertion and removal process. Temperature control function debugging: Simulating an ambient temperature of 35℃ and a supercharger at full power (600kW) output, the temperature of each module is stable at 58℃~62℃, the coolant outlet temperature is stable at 52℃~54℃, and the temperature control accuracy is ±0.5℃; Simulating module insertion and removal (reducing from 10 to 5), the temperature control system completes parameter adjustment within 500ms, and the temperature fluctuation is ≤1℃; Operational monitoring: During normal operation of the supercharged reactor, the cloud monitoring platform displays the operating status of each module, temperature data, and liquid cooling system parameters in real time. When a module failure occurs, the main control layer immediately issues an audible and visual alarm and disconnects the faulty module. The system continues to operate normally without any downtime impact.

Claims

1. A modular liquid-cooled supercharged reactor design method supporting plug-and-play power modules and intelligent temperature control, characterized in that... Includes the following steps: Step 1, Overall modular structure design of the supercharged reactor: The supercharger adopts a layered modular design, which is divided into a main control layer, a power layer, a liquid cooling layer and a basic support layer from top to bottom. Each layer is connected through a standardized interface to achieve independent operation and maintenance and flexible expansion. 1.1) Main Control Layer: Integrates the main controller, communication module, and data acquisition module. The main controller adopts an industrial-grade MCU, which is responsible for receiving the operating data of each power module and the status data of the liquid cooling system, and executing power distribution, temperature control command issuance, and plug-and-play identification control. The communication module adopts Ethernet + CAN bus dual-mode communication to ensure the real-time performance and reliability of data transmission. The data acquisition module is responsible for collecting the temperature, current, and voltage data of each power module and the coolant temperature, flow rate, and pressure data of the liquid cooling system. 1.2) Power layer: It consists of several standardized power modules. Each power module uses silicon carbide MOSFET devices. The power modules are connected in parallel to achieve flexible power distribution. The power layer has reserved standardized mounting slots to adapt to power modules of different specifications. 1.3) Liquid Cooling Layer: A distributed liquid cooling loop design is adopted, including a main coolant tank, a variable frequency circulating pump, a plate heat exchanger, a distribution manifold, and several modular liquid cooling interfaces. The main coolant tank is filled with nanofluid coolant. The variable frequency circulating pump is responsible for driving the coolant circulation. The plate heat exchanger realizes the heat exchange between the coolant and the external environment. The distribution manifold evenly distributes the coolant to the liquid cooling channel of each power module to realize synchronous heat dissipation of multiple modules. 1.4) Basic support layer: A steel structure frame is used to provide installation support for each layer of modules. The frame integrates heat dissipation and ventilation openings and shock absorption and buffer structures. Step 2, Plug and Play Design for Power Modules: The plug-and-play design of the power modules comprises four main parts: standardized mechanical interfaces, standardized electrical interfaces, standardized liquid cooling interfaces, and an intelligent identification and configuration mechanism, enabling rapid, non-stop replacement and capacity expansion of power modules. 2.1) Standardized mechanical interface: Each power module uses a uniform mechanical housing with standardized buckles and guide grooves on both sides of the housing to precisely match the mounting slots of the power layer. The buckles adopt a quick-locking structure. 2.2) Standardized electrical interface: The electrical interface of the power module adopts a standardized hot-swappable connector, which integrates the power interface, signal interface and power supply interface. The power interface is responsible for transmitting charging power, the signal interface is responsible for data communication with the main control layer, and the power supply interface provides power to the control circuit of the power module. The connector adopts an anti-misinsertion design to ensure that the positive and negative terminals are not reversed. 2.3) Standardization of liquid cooling interface: The liquid cooling interface of the power module adopts a quick-connect sealing connector, which matches the modular liquid cooling interface of the liquid cooling layer. The connector has a built-in sealing gasket, which automatically seals during insertion and removal to prevent coolant leakage. At the same time, the interface adopts a self-locking structure. 2.4) Intelligent Identification and Configuration Mechanism: Each power module has a built-in unique identification chip that stores the module's power specifications, factory parameters, and operating history information. When a power module is inserted into the installation slot, the main control layer's data acquisition module reads the module ID information through the signal interface. The main controller automatically identifies the module specifications based on the ID information and completes the adaptive configuration of power allocation parameters and communication parameters, enabling the power module to be plug-and-play. If the inserted module is faulty or its specifications are incompatible, the main control layer will issue an alarm signal and prohibit the module from being put into operation, ensuring the overall safe operation of the supercharger stack. Step 3, Intelligent temperature control method based on Transformer-GNN: The intelligent temperature control method based on Transformer-GNN combines the temporal feature extraction advantages of the Transformer model with the spatial correlation capture advantages of the GNN model to achieve precise adaptive temperature control of the liquid cooling system. Specifically, it includes five stages: data acquisition, feature fusion, temperature prediction, temperature control adjustment, and closed-loop feedback. 3.1) Data Acquisition: The supercharged reactor's operating data is acquired in real time through the main control layer's data acquisition module, with an acquisition cycle of 100ms. The acquired data includes: real-time temperature of each power module, real-time output current and voltage of each power module, coolant inlet and outlet temperatures, circulation flow rate and system pressure of the liquid cooling system, and ambient temperature and humidity. The acquired data is preprocessed to remove outliers and normalized. The min-max normalization method is used to map the data to the [0,1] interval to eliminate the influence of dimensions. The normalization formula is as follows: In the formula: The raw data collected; This is the historical minimum value for this type of data; This represents the historical maximum value for this type of data; The normalized data eventually yields a standardized input dataset. 3.2) Feature Fusion: Construct a Transformer-GNN fusion model to extract temporal and spatial features from the collected standardized data. 3.2.1) Temporal Feature Extraction: The Transformer encoder is used to extract the temperature time-series data and heat generation power time-series data of each power module; the historical temperature data and heat generation power data of each power module are arranged in time series to form a temporal feature vector, which is input into the self-attention mechanism of the Transformer encoder. The self-attention mechanism automatically captures the correlation between temperature and heat generation power at different time steps and extracts the temporal trend features of temperature changes. 3.2.2) Spatial Feature Extraction: A GNN model is used to extract the thermal coupling features between multiple power modules and the spatial correlation features of the liquid cooling circuit. Each power module is used as a node in the GNN model, and the heat conduction relationship between modules and the connection relationship of the liquid cooling circuit are used as edges between nodes to construct a graph structure. The GNN model uses graph convolutional layers to aggregate the features of each node with the features of adjacent nodes to capture the spatial thermal coupling features between multiple power modules and the spatial correlation features between coolant flow rate and the temperature of each module. The feature aggregation formula of the graph convolutional layer is as follows: In the formula: For the first Nodes after layer graph convolution eigenvectors; , For nodes , The set of neighboring nodes; N is the number of training samples; , They are nodes , The number of adjacent nodes; For the first The weight matrix of layer graph convolution; For the first Layer nodes eigenvectors; For the first Layer bias terms; The ReLU function is used as the activation function. 3.2.3) Feature Fusion: The temporal feature vector extracted by the Transformer encoder and the spatial feature vector extracted by the GNN model are fused through a fully connected layer. An attention weight allocation mechanism is used to assign different weights to the temporal and spatial features. The attention weight allocation formula and the feature fusion formula are as follows: In the formula: For time-series feature weights; Spatial feature weights; Assess the importance of time-series features; Score the importance of spatial features; Temporal feature vectors extracted by the Transformer encoder; Spatial feature vectors extracted for the GNN model; The fused feature vector serves as the input for temperature prediction. The natural index value is used to score the importance of time-series features and is used to normalize the weight allocation through the Softmax mechanism. The natural index value for scoring the importance of spatial features is used to normalize the weight allocation through the Softmax mechanism; 3.3) Temperature Prediction: The fused feature vector is input into the prediction layer of the Transformer-GNN fusion model. The prediction layer uses a fully connected neural network to output the predicted temperature values ​​of each power module and the predicted temperature value of the coolant outlet of the liquid cooling system within the next 5 seconds. At the same time, the model predicts the temperature change trend under different temperature control strategies based on the changes in ambient temperature and module heat generation power, providing a basis for temperature control adjustment. The training process of the Transformer-GNN fusion model is as follows: Historical operating data of the supercharged heap under different operating conditions are collected to construct training and testing datasets. The training dataset is used for model parameter training, and the testing dataset is used for model performance verification. Mean squared error is used as the loss function, and the model parameters are iteratively optimized using the gradient descent algorithm until the model's prediction error is less than a preset threshold, thus completing model training. The loss function formula is as follows: In the formula: This represents the mean squared error loss function value; This represents the number of training samples; For the first The true temperature value of each sample; For the first Temperature prediction values ​​for each sample; the trained model is deployed to the main controller of the main control layer, supporting online fine-tuning and continuously optimizing prediction accuracy based on long-term operating data of the supercharged reactor; 3.4) Temperature Control: Based on the predicted temperature value and the preset temperature threshold, the main controller formulates an adaptive temperature control strategy and sends control commands to the actuators of the liquid cooling layer. The specific adjustment method is as follows: 3.4.1) When the temperature of a certain power module is predicted to exceed 65℃, the main controller controls the variable frequency circulating pump to increase its speed and increase the coolant circulation flow rate. At the same time, it adjusts the corresponding liquid cooling branch valve of the module to increase the coolant distribution of the module and accelerate heat dissipation. If the temperature prediction value exceeds 70℃, the output power of the module is reduced to avoid overheating and damage to the module. 3.4.2) When the predicted temperature of each power module is below 45℃ and the coolant outlet temperature is below 40℃, the main controller controls the variable frequency circulating pump to reduce the speed, reduce the coolant circulation flow, reduce the energy consumption of the liquid cooling system, and achieve energy-saving operation. 3.4.3) When the temperature difference between multiple power modules is predicted to exceed 5℃, the main controller adjusts the valve opening of the distributor manifold to adjust the coolant distribution of each module, balance the temperature of each module, and avoid local overheating or insufficient heat dissipation. 3.4.4) When the power module achieves plug-and-play functionality, the Transformer-GNN fusion model automatically updates the graph structure, quickly re-extracts spatial and temporal features, predicts new temperature distributions, and the main controller promptly adjusts the operating parameters of the liquid cooling system to ensure that the temperature control system quickly adapts to changes in the number of modules and maintains stable supercharged reactor temperature. 3.5) Closed-loop feedback: The data acquisition module collects the operating data of the supercharged reactor after temperature control adjustment in real time and feeds the data back to the Transformer-GNN fusion model. The model corrects the temperature prediction value based on the feedback data. The main controller adjusts the temperature control adjustment strategy based on the corrected prediction value, forming a closed-loop temperature control system of "acquisition-fusion-prediction-adjustment-feedback" to ensure long-term stable temperature control accuracy and adapt to the dynamic operating conditions of the supercharged reactor. Step 4, System Debugging and Operation Monitoring: 4.1) System Debugging: After the supercharged reactor is assembled, plug-and-play and temperature control functions are debugged. During plug-and-play debugging, power modules of different specifications are inserted in sequence to verify the accuracy of module identification, the automation of parameter configuration, and the stability of operation. During temperature control function debugging, different operating conditions are simulated to verify the temperature prediction accuracy and the effectiveness of temperature control adjustment of the Transformer-GNN fusion model, ensuring that all indicators meet the design requirements. 4.2) Operation Monitoring: During normal operation of the supercharged reactor, the main control layer monitors the operating status of each module, the working status of the liquid cooling system, and the temperature control effect in real time. The operating data is stored in the local storage module and uploaded to the cloud monitoring platform through the communication module. If a module failure, liquid cooling system abnormality, or temperature exceeding the standard occurs, the main control layer immediately issues an audible and visual alarm signal and uploads the alarm information to the cloud. At the same time, emergency handling measures are taken to ensure the safe and stable operation of the supercharged reactor.

2. The modular liquid-cooled supercharged reactor design method supporting plug-and-play power modules and intelligent temperature control as described in claim 1, characterized in that: In step 1.3), silica nanoparticles are added to the nanofluid coolant, and the amount of silica nanoparticles added accounts for 0.5% of the total mass.

3. The modular liquid-cooled supercharged reactor design method supporting plug-and-play power modules and intelligent temperature control according to claim 1, characterized in that: In step 3.1), outliers are removed using the 3σ criterion, which removes data that deviates from the normal range. In step 3.3), the prediction error of the model continues until it is less than the preset threshold of ≤0.5℃; In step 3.4), the main controller determines the normal operating temperature range of the power module from 40℃ to 65℃ and the coolant outlet temperature threshold from 55℃ based on the predicted temperature value and the preset temperature threshold. In step 3.5), the data acquisition module collects real-time operating data of the supercharged reactor after temperature control adjustment: temperature of each module and coolant parameters.

4. The modular liquid-cooled supercharged reactor design method supporting plug-and-play power modules and intelligent temperature control according to claim 1, characterized in that: In step 4.1), the simulated operating conditions are: full power output, partial module operation, and ambient temperature change; In step 4.2), the emergency response measures are: disconnecting the faulty module and reducing the system power.

5. A modular liquid-cooled supercharged reactor design method supporting plug-and-play power modules and intelligent temperature control according to claim 1, characterized in that... The specific steps are as follows: Step 1, Overall modular structure design of the supercharged reactor: The supercharger stack adopts a layered modular design. The main control layer uses an STM32H743VIT6 industrial-grade MCU as the main controller, and the communication module uses a DP83848 Ethernet chip and a TJA1050 CAN bus chip to achieve dual-mode communication. The power layer has 10 standardized mounting slots, and each power module uses SiC MOSFET devices with a rated power of 60kW. The modules are connected in parallel, and the maximum output power can reach 600kW. The liquid cooling layer uses a main coolant tank, a TD32-160B variable frequency circulating pump, and a BR0.3 plate heat exchanger. The water distribution manifold adopts a 1-in-10-out structure. Each power module is equipped with an independent liquid cooling channel, and the coolant is a nanofluid coolant. The foundation support layer uses a Q235 steel structure frame with a height of 1.8m and a width of 1.2m, integrating ventilation openings and shock-absorbing rubber pads. Step 2, Plug and Play Design for Power Modules: The mechanical interface adopts a standard 3U chassis specification, with quick-locking spring clips and a dovetail groove structure to ensure precise insertion and removal. The electrical interface uses MX34012 series hot-swappable connectors, integrating 3 power interfaces, 2 signal interfaces, and 1 power supply interface. The connector features an anti-misinsertion boss design with an IP67 protection rating. The liquid cooling interface uses LLH series quick-connect sealing connectors with built-in fluororubber gaskets, a rated pressure of 1.6MPa, and a insertion / removal cycle of ≥1000 times. Each power module has a built-in AT24C02 ID chip that stores information such as module ID, rated power, and manufacturing date. After insertion, the main control layer completes identification and parameter configuration within 100ms. Step 3, Intelligent temperature control method based on Transformer-GNN: 3.1 Data Acquisition: The temperature of each power module was acquired using a DS18B20 temperature sensor, the module current and voltage were acquired using an ACS712 current sensor and an ACS758 voltage sensor, the coolant temperature was acquired using a PT100 sensor, and the coolant flow rate was acquired using an LWGY turbine flow meter. The acquisition period was 100ms. Abnormal data was removed using the 3σ criterion, and the data was mapped to the [0,1] interval using the min-max normalization method. 3.2 Feature Fusion: The Transformer encoder is set with 6 layers, 8 self-attention heads, and 128 hidden layer dimensions; the GNN model uses GCN, with 3 graph convolutional layers, 64 node feature dimensions, and 32 edge feature dimensions; the fusion layer uses a fully connected layer, with the initial attention weights set to 0.6 for temporal features and 0.4 for spatial features, and dynamically adjusted according to the running status. 3.3 Temperature Prediction: The prediction layer uses a two-layer fully connected neural network with a sigmoid activation function for the output layer, predicting the temperature of each module and the coolant outlet temperature within the next 5 seconds. The model is trained using 10,000 sets of historical operating data, covering ambient temperatures from 30℃ to 60℃ and output power conditions from 0 to 600kW. The loss function is the mean squared error, with 1000 training iterations and a learning rate of 0.

001. After training, the prediction error is ≤0.5℃. The loss function formula and learning rate update formula are as follows: In the formula: This represents the number of training samples; For the first The true temperature value of each sample; For the first Predicted temperature values ​​for each sample; For the first The learning rate for each iteration; The initial learning rate, Take 0.001; This is the learning rate decay coefficient. Take 0.98; This represents the number of iterations. 3.4 Temperature Control Adjustment: The preset normal operating temperature range of the power module is 40℃~65℃, and the coolant outlet temperature threshold is 55℃; when the predicted module temperature is ≥65℃, the variable frequency circulating pump speed is increased from 1500r / min to 2500r / min, and the coolant flow rate is increased from 10L / min to 20L / min; when the module temperature is ≤45℃ and the coolant outlet temperature is ≤40℃, the pump speed is reduced to 1000r / min, and the flow rate is reduced to 5L / min; when the temperature difference between multiple modules is ≥5℃, the opening of the distributor manifold valve is adjusted to adjust the coolant distribution difference between each module to ≤2L / min. 3.5 Closed-loop feedback: The adjusted data is collected every 100ms and fed back to the model to correct the predicted value, dynamically adjusting the temperature control strategy to ensure that the temperature is stable within the preset range; Step 4, System Debugging and Operation Monitoring: Plug and play debugging: When 5, 8, and 10 60kW power modules are inserted in sequence, the main control layer can complete the identification within 100ms, automatically configure the power allocation parameters, and the modules operate normally. There are no problems such as coolant leakage or poor electrical contact during the insertion and removal process. Temperature control function debugging: Simulating an ambient temperature of 35℃ and the overcharge full power output condition, the temperature of each module is stable at 58℃~62℃, the coolant outlet temperature is stable at 52℃~54℃, and the temperature control accuracy is ±0.5℃; simulating module plugging and unplugging, the temperature control system completes parameter adjustment within 500ms, and the temperature fluctuation is ≤1℃; Operational monitoring: During normal operation of the supercharged reactor, the cloud monitoring platform displays the operating status of each module, temperature data, and liquid cooling system parameters in real time. When a module failure occurs, the main control layer immediately issues an audible and visual alarm and disconnects the faulty module. The system continues to operate normally without any downtime impact.

Images