A junction temperature prediction and liquid metal adaptive cooling method and system for power semiconductor devices

Abstract

The application discloses a junction temperature prediction and liquid metal adaptive cooling method and system of a power semiconductor device, and belongs to the technical field of power semiconductor device thermal management and intelligent control. The method comprises the following steps: constructing a digital twin thermal model of the power semiconductor device based on a device data manual or experimental test data, and introducing a liquid metal cooling channel to establish a thermal flow coupling multi-physical field simulation model; constructing a multivariate time sequence sample by using simulation data generated under random working conditions and experimental test data, training a GRU prediction model combined with a physical constraint loss, and realizing future junction temperature sequence prediction; and according to the deviation between the predicted junction temperature and a target junction temperature, and in combination with the upper and lower limits of the flow rate and the change rate constraint, the liquid metal flow rate is prospectively adjusted, so that high-precision prediction and closed-loop adaptive cooling control of the junction temperature of the power semiconductor device are realized, and the problems of insufficient junction temperature prediction accuracy, control response lag and difficulty in realizing prospective thermal control in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of thermal management and intelligent control technology for power semiconductor devices, and in particular to a method and system for predicting junction temperature and adaptive cooling with liquid metal for power semiconductor devices. Background Technology

[0002] With the development of new energy vehicles, rail transit, aviation electric propulsion, and high-power-density power systems, power semiconductor devices are playing an increasingly crucial role in high-end power electronic systems. Wide-bandgap power devices such as SiC MOSFETs are widely used in high-power-density applications due to their advantages such as high voltage withstand capability, fast switching speed, and low conduction losses. However, when these power semiconductor devices operate in high-frequency, high-power, and high-temperature environments, their internal junction temperature rises rapidly. Excessive junction temperature can lead to device parameter drift, increased switching losses, thermal stress accumulation, and accelerated aging and failure. Therefore, accurate prediction and real-time control of junction temperature has become an important research direction in device thermal management.

[0003] Current power semiconductor device thermal management primarily employs traditional heat dissipation methods such as air cooling and water cooling. Air cooling structures are low-cost but have limited heat exchange capacity; water cooling systems have relatively strong heat dissipation capacity, but still suffer from problems such as response lag and large system size under high heat flux density. Liquid metals, due to their high thermal conductivity and strong heat dissipation capabilities, are increasingly being used to cool high heat flux density devices. However, the flow rate regulation of existing liquid metal cooling systems usually still relies on empirical settings or simple feedback control, lacking prediction of future temperature evolution trends. Therefore, when power surges or ambient temperature fluctuates, problems such as response lag, junction temperature overshoot, or unstable flow rate regulation can easily occur.

[0004] From a methodological perspective, existing junction temperature prediction methods mainly include mechanistic models based on thermal resistance and thermal capacity networks, numerical simulation models based on finite element or computational fluid dynamics, and data-driven models based on neural networks. Thermal resistance and thermal capacity models are simple in structure and computationally efficient, but rely on many simplifying assumptions, making it difficult to accurately describe the complex coupling relationship between the device and the cooling system. Numerical simulation models have high physical accuracy, but require large computational resources and are difficult to meet real-time control requirements. Pure data-driven models, while possessing strong nonlinear fitting capabilities, are prone to prediction distortion under unseen operating conditions due to the lack of physical constraints, and are also difficult to directly use for stable and reliable closed-loop control.

[0005] Existing research has mostly focused on offline modeling or single-time temperature prediction, lacking a linkage control mechanism with cooling actuators. For liquid metal cooling scenarios of power semiconductor devices, temperature prediction alone is insufficient to meet engineering requirements. The real key is to obtain a reliable prediction of the future junction temperature and then adjust the liquid metal flow rate in real time based on the prediction results to achieve forward-looking closed-loop regulation. This allows for increased heat dissipation capacity before the temperature rises, reducing junction temperature peaks and thermal shocks. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for predicting junction temperature and adaptive cooling with liquid metal for power semiconductor devices, which can reduce junction temperature overshoot and control lag, improve the stability of the cooling system and the reliability of device operation, thereby solving the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal includes the following steps:

[0009] S1: Construct a digital twin thermal model to characterize the thermal behavior of power semiconductor devices, so that the digital twin thermal model can reflect the junction temperature variation characteristics of the device under different operating conditions;

[0010] S2: Introduce a liquid metal cooling channel model into the digital twin thermal model to establish the coupling relationship between device heat conduction and liquid metal flow and heat transfer process;

[0011] S3: Acquire operating data under multiple operating conditions, including device power, ambient temperature, liquid metal flow rate, and junction temperature information;

[0012] S4: Construct a multivariate time series sample with a sliding time window based on the running data, and train a time series prediction model with physical information constraints. The time series prediction model outputs a multi-step junction temperature prediction sequence within a future preset time range. Physical constraints related to the thermal behavior of the device are introduced during the training process.

[0013] S5: During operation, the current operating condition is input into the time-series prediction model to obtain the junction temperature prediction result within a preset time range in the future;

[0014] S6: Adjust the liquid metal cooling system according to the junction temperature prediction results to make the device junction temperature tend to the target range, thereby realizing forward-looking closed-loop control.

[0015] Preferably, the liquid metal is a gallium-based liquid metal, using a gallium indium tin alloy as the cooling medium, and combined with an electromagnetic pump to achieve circulating flow heat exchange. A temperature sensor is used to collect the temperature of the liquid metal, and a flow meter is used to collect the flow rate of the liquid metal, thereby providing real-time data for subsequent model prediction and control adjustment.

[0016] Preferably, the gallium-based liquid metal is an alloy formed of gallium and at least one other metal.

[0017] Preferably, the gallium-based liquid metal is at least one of gallium indium tin alloy, gallium indium alloy, and gallium tin alloy.

[0018] Preferably, the power semiconductor device is a semiconductor device with a power loss and junction temperature coupling relationship, and the power semiconductor device is at least one of SiC MOSFET, IGBT, and silicon-based MOSFET.

[0019] Preferably, the digital twin thermal model is constructed based on at least one of device parameter data, experimental measurement data, and numerical simulation data.

[0020] Preferably, the digital twin thermal model is used to ensure that both the steady-state thermal response and the transient thermal response are consistent with the actual thermal characteristics of the device.

[0021] Preferably, the running data includes simulation-generated data and actual measurement data. The time series prediction model is trained based on the simulation-generated data and the parameters are calibrated or fine-tuned in conjunction with the actual measurement data.

[0022] Preferably, the time-series prediction model is a gated cyclic unit neural network model, which is used to characterize the dynamic relationship between historical operating conditions and junction temperature.

[0023] Preferably, the physical constraints are established by introducing the relationship between the junction temperature change rate and device power, ambient temperature, liquid metal temperature at the cold plate inlet, and liquid metal flow rate.

[0024] Preferably, in the time-series prediction model, the input of the multivariate time-series sample includes device power, ambient temperature, liquid metal temperature at the cold plate inlet, liquid metal flow rate, historical junction temperature and temperature rise information at several historical moments, and the output is a junction temperature sequence within a preset future time range.

[0025] Preferably, the adjustment amount of the liquid metal cooling system is determined based on the deviation between the predicted junction temperature and the target junction temperature and the peak junction temperature within the future prediction time window, and the adjustment amount is subject to amplitude constraints and rate of change constraints.

[0026] According to another aspect of the present invention, a junction temperature prediction and liquid metal adaptive cooling system for a power semiconductor device is provided, for implementing the junction temperature prediction and liquid metal adaptive cooling method for a power semiconductor device as described above, comprising:

[0027] The digital twin modeling module is used to build digital twin thermal models to characterize the thermal behavior of power semiconductor devices.

[0028] The data acquisition module is used to acquire operational data under multiple operating conditions;

[0029] The time-series prediction module is used to train a time-series prediction model constrained by physical information based on the running data and output the junction temperature prediction result within a preset time range in the future.

[0030] The control execution module is used to adjust the liquid metal cooling system according to the junction temperature prediction result, so that the device junction temperature tends to the target range.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] This invention establishes a digital twin thermal model of a power semiconductor device and employs a training method that combines simulation and experimental data. This enables the model to balance physical consistency with actual operating conditions, effectively reducing the distribution discrepancy between simulation and reality, and improving the engineering applicability and robustness of junction temperature prediction. Furthermore, by introducing thermal balance differential equations as physical constraints into the model training, it is possible to achieve junction temperature prediction that combines data-driven and mechanism-constrained approaches. Specifically, the liquid metal flow rate is pre-adjusted based on the deviation between the predicted junction temperature and the target junction temperature. Compared with traditional fixed flow rate control or simple feedback control methods, this effectively reduces control lag and junction temperature overshoot, improves cooling efficiency and system stability, thereby enhancing device reliability and extending service life. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall structure of the junction temperature prediction and liquid metal adaptive cooling system of the present invention;

[0034] Figure 2 This is a model structure diagram of the junction temperature prediction and liquid metal adaptive cooling system of the present invention;

[0035] Figure 3 This is a random operating condition data graph for the junction temperature prediction and liquid metal adaptive cooling method of the present invention;

[0036] Figure 4 This is a COMSOL multiphysics coupling simulation diagram of the present invention;

[0037] Figure 5 This is a curve comparing the predicted and actual values ​​of the settling temperature for verification purposes in this invention.

[0038] Figure 6 This is a diagram showing the junction temperature and flow rate control response of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] To address the problems of insufficient junction temperature prediction accuracy, lag in control response, and difficulty in achieving forward-looking thermal control in existing technologies, especially under complex operating conditions such as sudden power changes, fluctuating ambient temperatures, and significant variations in cooling conditions, traditional fixed flow rate control or simple feedback control based on current errors often only allow for passive adjustment after the temperature rises, making it difficult to suppress junction temperature overshoot in a timely manner. This, in turn, affects the thermal stability, operational reliability, and lifespan of power semiconductor devices. This embodiment provides the following technical solution:

[0041] A method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal includes the following steps:

[0042] S1: Based on device datasheets or experimental test data, establish a geometric model and a thermal conduction model of the power semiconductor device in multiphysics simulation software, and calibrate the device thermal resistance parameters, material thermal conductivity parameters and boundary conditions to ensure that the steady-state thermal response and transient thermal response of the simulation model are consistent with the actual characteristics of the device, thereby constructing a digital twin thermal model of the power semiconductor device with engineering consistency.

[0043] S2: Based on the digital twin thermal model of power semiconductor devices, a liquid metal cooling channel model is introduced. The liquid metal temperature, flow rate and boundary heat transfer conditions at the cold plate inlet are set to establish a multiphysics simulation model that couples device heat conduction with liquid metal flow. The liquid metal is gallium-based liquid metal.

[0044] It should be noted that the multiphysics simulation model is built based on COMSOL or ANSYS.

[0045] S3: Use programming tools to generate a sequence of random operating conditions with a preset duration. The random operating conditions include information such as power of power semiconductor devices, ambient temperature and flow rate changes. Import the random operating conditions into a multiphysics simulation model for transient solution to obtain junction temperature response data under multiple operating conditions. Also, collect experimental test data of power semiconductor devices during actual operation. Combine the experimental test data with the simulation data to form a model training dataset, so that the established model can cover a wider range of thermal behavior scenarios.

[0046] S4: Organize the multiphysics simulation data and experimental test data into multivariate time series samples based on a sliding time window for training the model. First, pre-train the model based on the simulation data, and then combine the experimental test data to calibrate or fine-tune the model parameters so that the model can gradually adapt to the real operating conditions.

[0047] During model training, the input samples include device power, ambient temperature, liquid metal flow rate, and historical junction temperature at several historical moments. The output samples are multi-step junction temperature prediction sequences within a preset future time range. During training, physical constraints based on the junction temperature change rate are introduced, and a physical constraint loss function is constructed by combining device power input, ambient heat dissipation, and convective heat transfer caused by liquid metal flow. By jointly optimizing the data loss term and the physical loss term, the model can satisfy thermodynamic laws while learning the mapping relationship between operating parameters and junction temperature evolution, thereby achieving high-precision prediction of future junction temperature.

[0048] S5: During the closed-loop control phase, the current operating condition sequence is acquired in real time and input into the trained junction temperature prediction model to obtain the predicted junction temperature of the power semiconductor device within a preset time range in the future.

[0049] It should be noted that, unlike the traditional method of feedback adjustment based solely on the current junction temperature, in this embodiment, by predicting the future junction temperature change trend, the control system has the ability to make forward-looking judgments and can adjust the flow rate in advance before the temperature rise becomes significant. The prediction result is a sequence of junction temperatures at multiple future moments, which is used for subsequent flow rate optimization calculations.

[0050] S6: Based on the predicted junction temperature sequence and target junction temperature threshold within the future preset time window, combined with the upper and lower limits of liquid metal flow rate and the constraint of change rate, calculate the flow rate adjustment amount and output it to the electromagnetic pump control module to realize closed-loop control of the junction temperature of power semiconductor devices.

[0051] like Figure 1 As shown, in order to better realize the junction temperature prediction and liquid metal adaptive cooling process of power semiconductor devices, this embodiment provides a junction temperature prediction and liquid metal adaptive cooling system for power semiconductor devices, which is used to realize the junction temperature prediction and liquid metal adaptive cooling method of the above-mentioned power semiconductor devices. The system includes a physical system, a digital twin augmentation system, and a control execution module, and the various parts work together to form a closed-loop control architecture that combines virtual and physical elements.

[0052] like Figure 1 As shown above, the physical system includes components such as power semiconductor devices, liquid metal cold plates, electromagnetic pumps, flow meters, heat sinks, and temperature sensors, forming a closed-loop liquid metal flow channel.

[0053] In this embodiment, the liquid metal is gallium-based liquid metal, and gallium indium tin alloy is used as the cooling medium. Combined with an electromagnetic pump drive, it realizes circulating flow heat exchange. A temperature sensor is used to collect the temperature of the liquid metal, and a flow meter is used to collect the flow rate of the liquid metal, thereby providing real-time data for subsequent model prediction and control adjustment.

[0054] like Figure 1 As shown below, the digital twin enhancement system includes a multiphysics simulation model, a random operating condition generation module, and a digital twin enhanced prediction model. The random operating condition generation module constructs simulation inputs under different power, ambient temperature, and flow rate conditions, and uses the multiphysics simulation model to generate corresponding junction temperature response data, forming a simulation dataset covering multiple operating scenarios. This simulation dataset, together with the real-time data collected by the physical system, is used for model training, thereby achieving digital twin enhancement of the thermal behavior of power semiconductor devices.

[0055] like Figure 2 As shown, the model includes operating condition input, time-series feature extraction, junction temperature prediction, and control feedback. The operating condition input consists of multivariate time-series data, including parameters such as power semiconductor device power, ambient temperature, liquid metal inlet temperature, flow rate, and junction temperature. An input sequence is constructed using a sliding time window. The time-series feature extraction part employs a gated recurrent unit (GRU) structure to extract time-related features from historical operating conditions. Based on this, a fully connected network outputs the junction temperature prediction results within a preset future time range. During model training, physical constraints based on device thermal equilibrium are introduced and, together with the data error term, constitute a joint optimization objective to improve the model's prediction accuracy and generalization ability. In the closed-loop control process, the liquid metal flow rate is adjusted according to the deviation between the predicted junction temperature and the target junction temperature, and feedback is formed by combining real-time measurement data, thereby achieving dynamic control of the power semiconductor device junction temperature.

[0056] In this embodiment, a three-dimensional thermal model of the power semiconductor device and a liquid metal flow model are established based on multiphysics simulation software. The preferred multiphysics simulation software is COMSOL, but other simulation platforms with thermal-fluid coupling solution capabilities, such as ANSYS, can also be used. The power semiconductor device is a SiC MOSFET, and the device parameters can be calibrated with reference to the device datasheet. The parameters can be set according to the data provided by Wolfspeed. By adjusting the thermal conductivity of each layer of material inside the device and the boundary heat transfer conditions, the steady-state and transient junction temperature response of the model under a given power condition is made consistent with the actual thermal characteristics of the device, thereby constructing a digital twin thermal model with engineering consistency.

[0057] It should be noted that the liquid metal cooling medium is not limited to gallium-based liquid metals; other liquid metal materials with high thermal conductivity can also be used. However, in specific experimental and simulation embodiments, gallium-based liquid metals are preferred to ensure the consistency between simulation parameters and experimental conditions.

[0058] like Figure 3 As shown, a sequence of random operating conditions is generated using programming tools, forming simulation input data with multiple time points and variables. These random operating conditions include variables such as device power, ambient temperature, and liquid metal flow rate, and can employ a combination of step changes, slow changes, and random fluctuations to cover as many operating scenarios as possible. The time-series samples obtained in this way can effectively improve the model's generalization ability to complex, abrupt, and unseen operating conditions.

[0059] In this embodiment, the duration of the random operating condition sequence can be set according to training requirements, specifically 500 s, to ensure that the data can cover a sufficient amount of thermal dynamic processes.

[0060] The generated random operating conditions are input into the multiphysics simulation model for transient solution to obtain the junction temperature response, flow velocity and other relevant data at each time point.

[0061] like Figure 4 As shown, the established multiphysics coupling model can simultaneously reflect the flow behavior of liquid metal and the heat conduction process of power semiconductor devices. The flow state of liquid metal in the channel determines the heat transfer capacity of the device, while the change in device junction temperature in turn affects the outlet temperature and return temperature, thus forming a heat-flow coupling relationship. Through this simulation, the device temperature field, flow field and junction temperature response data over time can be obtained, providing a basis for constructing a prediction model that combines data-driven and mechanism-constrained approaches.

[0062] Multiphysics simulation data and experimental test data are organized into multivariate time series samples based on a sliding time window to train the junction temperature prediction model. The time series samples consist of the input of the operating condition sequence at several historical moments and the output is the junction temperature sequence within a preset future time range, so as to realize the modeling of the dynamic change process of junction temperature.

[0063] In this embodiment, the time window length and prediction step size can be set according to specific application requirements.

[0064] In this embodiment, the junction temperature prediction model adopts a temporal neural network structure and combines it with a physical information neural network method to construct a temporal prediction model constrained by physical information.

[0065] Specifically, the temporal neural network adopts a gated recurrent unit (GRU) structure to extract features from historical operating condition sequences to characterize the temporal correlation and dynamic response characteristics during junction temperature evolution. The physical information module introduces junction temperature time derivative information and establishes a physical constraint loss function by combining the thermal balance relationship of power semiconductor devices. By jointly optimizing the data loss term and the physical loss term, the model can satisfy thermodynamic laws while learning the mapping relationship between operating condition parameters and junction temperature evolution, thereby improving the physical consistency and generalization ability of the prediction results.

[0066] In this embodiment, the model can be pre-trained based on multiphysics simulation data, and then the parameters can be calibrated or fine-tuned by combining experimental test data, so that the model can adapt to actual operating conditions and improve prediction accuracy and stability.

[0067] like Figure 5 As shown, the model's prediction results on the validation dataset are in good agreement with the reference junction temperature data, indicating that the model can effectively characterize the relationship between operating parameters and junction temperature changes, and maintain good prediction performance under complex perturbation conditions.

[0068] Once the model training is complete, it is deployed into the control program, which can run on a general computing platform or an embedded control system. In each control cycle, the system reads information such as the current junction temperature, power, ambient temperature, flow rate, and inlet temperature, constructs a historical input sequence, and then calls the prediction model to predict the junction temperature within a preset time range in the future. Unlike the traditional method of feedback adjustment based solely on the current junction temperature, this embodiment predicts the future junction temperature change trend, enabling the control system to have forward-looking judgment capabilities and adjust the flow rate in advance before the temperature rise becomes significant.

[0069] like Figure 6 As shown, during the closed-loop control process, the system dynamically adjusts the liquid metal flow rate based on the predicted future junction temperature change trend, so that the junction temperature is stabilized near the target range. Compared with feedback control based only on the current temperature error, this control method can achieve forward-looking adjustment, thereby reducing temperature overshoot and improving system stability.

[0070] The above method can be used to adjust the liquid metal flow rate in advance based on the predicted junction temperature, reduce response lag and junction temperature overshoot in the thermal management process, improve the thermal stability of power semiconductor devices during operation, and enhance the overall reliability of the system.

[0071] In summary, the junction temperature prediction and liquid metal adaptive cooling method and system for power semiconductor devices proposed in this invention address the problems of insufficient junction temperature prediction accuracy, lag in control response, and difficulty in achieving forward-looking thermal control in existing technologies. Especially under complex operating conditions such as sudden power changes, ambient temperature fluctuations, and significant changes in cooling conditions, traditional fixed flow rate control or simple feedback control based on current errors can often only passively adjust after the temperature rises, making it difficult to suppress junction temperature overshoot in a timely manner, thereby affecting the thermal stability, operational reliability, and service life of power semiconductor devices.

[0072] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal, characterized in that, Includes the following steps: S1: Construct a digital twin thermal model to characterize the thermal behavior of power semiconductor devices, so that the digital twin thermal model can reflect the junction temperature variation characteristics of the device under different operating conditions; S2: Introduce a liquid metal cooling channel model into the digital twin thermal model to establish the coupling relationship between device heat conduction and liquid metal flow and heat transfer process; S3: Acquire operating data under multiple operating conditions, including device power, ambient temperature, liquid metal flow rate, and junction temperature information; S4: Construct a multivariate time series sample with a sliding time window based on the running data, and train a time series prediction model with physical information constraints. The time series prediction model outputs a multi-step junction temperature prediction sequence within a future preset time range. Physical constraints related to the thermal behavior of the device are introduced during the training process. S5: During operation, the current operating condition is input into the time-series prediction model to obtain the junction temperature prediction result within a preset time range in the future; S6: Adjust the liquid metal cooling system according to the junction temperature prediction results to make the device junction temperature tend to the target range, thereby realizing forward-looking closed-loop control.

2. The method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal according to claim 1, characterized in that, The liquid metal is a gallium-based liquid metal.

3. The method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal according to claim 2, characterized in that, The gallium-based liquid metal is an alloy formed of gallium and at least one other metal.

4. The method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal according to claim 2, characterized in that, The gallium-based liquid metal is at least one of gallium indium tin alloy, gallium indium alloy, and gallium tin alloy.

5. The method for predicting junction temperature and adaptive liquid metal cooling of a power semiconductor device according to claim 1, characterized in that, The power semiconductor device is a semiconductor device with a power loss and junction temperature coupling relationship, and the power semiconductor device is at least one of SiC MOSFET, IGBT, and silicon-based MOSFET.

6. The method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal according to claim 1, characterized in that, The digital twin thermal model is constructed based on at least one of device parameter data, experimental measurement data, and numerical simulation data. The digital twin thermal model is used to ensure that both steady-state thermal response and transient thermal response are consistent with the actual thermal characteristics of the device.

7. The method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal according to claim 1, characterized in that, The operational data includes simulation-generated data and actual measurement data. The timing prediction model is trained based on the simulation-generated data and calibrated or fine-tuned using actual measurement data. The timing prediction model is a gated cyclic unit neural network model, which is used to characterize the dynamic relationship between historical operating conditions and junction temperature. The physical constraints are established by introducing the relationship between the junction temperature change rate and device power, ambient temperature, liquid metal temperature at the cold plate inlet, and liquid metal flow rate.

8. The method for predicting the junction temperature of a power semiconductor device and adaptively cooling it with liquid metal according to claim 1, characterized in that, In the time-series prediction model, the input of the multivariate time-series sample includes device power, ambient temperature, liquid metal temperature at the cold plate inlet, liquid metal flow rate, historical junction temperature and temperature rise information at several historical moments, and the output is a junction temperature sequence within a preset future time range.

9. The method for predicting the junction temperature of a power semiconductor device and adaptive cooling with liquid metal according to claim 1, characterized in that, The adjustment amount of the liquid metal cooling system is determined based on the deviation between the predicted junction temperature and the target junction temperature, as well as the peak junction temperature within the future prediction time window, and the adjustment amount is subject to amplitude constraints and rate of change constraints.

10. A junction temperature prediction and liquid metal adaptive cooling system for a power semiconductor device, used to implement the junction temperature prediction and liquid metal adaptive cooling method for a power semiconductor device as described in any one of claims 1-9, characterized in that, include: The digital twin modeling module is used to build digital twin thermal models to characterize the thermal behavior of power semiconductor devices. The data acquisition module is used to acquire operational data under multiple operating conditions; The time-series prediction module is used to train a time-series prediction model constrained by physical information based on the running data and output the junction temperature prediction result within a preset time range in the future. The control execution module is used to adjust the liquid metal cooling system according to the junction temperature prediction result, so that the device junction temperature tends to the target range.

Images