A temperature management system for a gallium nitride power adapter

Abstract

The invention relates to the technical field of power adapter temperature management, and discloses a temperature management system for a gallium nitride power adapter, comprising: a dynamic thermal model module, for accurately predicting the temperature change trend of each key component; a multi-level temperature monitoring network module, for forming a complete temperature distribution map; a multi-device collaborative perception module, for generating a perception result of the overall heat dissipation environment; an adaptive air duct structure module, for optimizing the airflow path; a heat dissipation path module based on swarm intelligence, for calculating and generating an optimal multi-device collaborative heat dissipation path using a swarm intelligence algorithm; and an adaptive heat dissipation control strategy module, for achieving multi-objective balanced optimization of temperature, noise and energy consumption. The invention solves the heat dissipation difficulty problem when multiple adapters are used in a concentrated manner, and achieves improved heat dissipation efficiency, reduced energy consumption and reduced noise.

Description

Technical Field

[0001] The present invention relates to the technical field of power adapter temperature management, and more specifically, it relates to a temperature management system for a gallium nitride power adapter. Background Art

[0002] Due to its high power density characteristics, a gallium nitride power adapter generates a large amount of heat during operation. Especially in portable multi-device charging scenarios such as conference rooms and shared workspaces, when multiple adapters are placed together, it is easy to form a local high-temperature area.

[0003] In the prior art, each adapter dissipates heat independently, lacking cooperation between devices; the air duct structure is fixed and cannot cope with dynamically changing hot spot positions; the fan control strategy is simple and it is difficult to balance noise, energy consumption, and heat dissipation effect. This results in inefficient utilization of heat dissipation resources, affecting the performance and service life of the power adapter. When multiple gallium nitride power adapters are placed together, the heat generated by each adapter affects each other, forming a complex thermal environment. Traditional single-device independent heat dissipation control is difficult to handle this cluster environment and cannot achieve globally optimal temperature management. Eventually, the temperature in the hot spot area is too high, while there is waste of heat dissipation resources in other areas. Summary of the Invention

[0004] The present invention provides a temperature management system for a gallium nitride power adapter to solve the technical problem of low heat dissipation efficiency when multiple adapters are placed together in related technologies.

[0005] The present invention provides a temperature management system for a gallium nitride power adapter, including:

[0006] A dynamic thermal model module for real-time constructing a three-dimensional thermal distribution model inside the gallium nitride power adapter, and accurately predicting the temperature change trend of each key component based on the heat conduction equation and the finite element analysis method;

[0007] A multi-level temperature monitoring network module for real-time collecting temperature data of each key point inside the power adapter through a distributed micro temperature sensor array to form a complete temperature distribution map;

[0008] A multi-device collaborative perception module for collecting and sharing temperature distribution data, fan status data, and airflow information between multiple gallium nitride power adapters to generate a perception result of the overall heat dissipation environment;

[0009] An adaptive air duct structure module for dynamically adjusting the internal air duct structure through shape memory materials and microelectromechanical actuators according to real-time thermal gradient distribution data to optimize the airflow path;

[0010] The heat dissipation path module based on swarm intelligence calculates and generates an optimal multi-device collaborative heat dissipation path based on the device topology network and the thermal state information of each device by applying swarm intelligence algorithms.

[0011] The adaptive heat dissipation control strategy module, based on environmental perception data and the heat dissipation path, uses deep reinforcement learning algorithms to intelligently regulate the operating parameters of the fans, achieving multi-objective balance optimization of temperature, noise, and energy consumption.

[0012] In a preferred embodiment, the multi-device collaborative perception module includes:

[0013] Collect the temperature distribution data of each key point of itself through the built-in temperature sensor network, and the sampling frequency of each temperature sensor is dynamically adjusted according to the temperature change gradient;

[0014] Dynamically adjust the information exchange frequency between devices according to the temperature change rate; use the proximity device discovery algorithm based on signal strength to identify adjacent adapter devices in the physical space through near-field communication technology;

[0015] Apply the weighted data fusion algorithm to generate a comprehensive state representation of the local environment.

[0016] In a preferred embodiment, the information exchange frequency between devices is calculated according to the following formula:

[0017] ;

[0018] Where, represents the information exchange frequency between devices; represents the lowest communication frequency of the system in the stable state; represents the adjustment coefficient, which is used to control the influence degree of temperature change on the communication frequency; represents the absolute magnitude of temperature change per unit time;

[0019] When the temperature changes violently, the communication frequency will increase accordingly to ensure that the system can respond to the changes in the thermal environment in a timely manner.

[0020] In a preferred embodiment, the adaptive air duct structure module includes:

[0021] Construct a micro air duct network made of shape memory polymer materials. In the initial state, the air ducts are in a uniformly distributed grid structure;

[0022] Based on the temperature distribution data, calculate the thermal gradient vector field in the space;

[0023] Based on the thermal response equation, calculate the deformation state of the air duct structure;

[0024] Convert the calculated air duct structure deformation command into an electronic control signal to precisely control the local deformation of the shape memory polymer material through the micro actuator network.

[0025] In a preferred embodiment, the swarm particle cooperative optimization algorithm adopted in the heat dissipation path module based on swarm intelligence includes the following steps:

[0026] Construct an objective function for overall heat dissipation optimization according to the temperature distribution and heat dissipation requirements of multiple adapters;

[0027] Initialize a heat dissipation path vector for each adapter, and the heat dissipation path vector represents the direction and intensity of the device's heat dissipation air flow;

[0028] Adopt an improved particle swarm optimization algorithm to iteratively update the heat dissipation path vector of each device;

[0029] Continuously update the record of the optimal position of each device and the record of the global optimal position during the iteration process.

[0030] In a preferred embodiment, the improved particle swarm optimization algorithm updates the heat dissipation path vector through the following formula:

[0031] ;

[0032] Where, represents the heat dissipation path vector of device at time ; represents the updated heat dissipation path vector of device at time ; represents the historical optimal path of device itself; represents the global optimal path; , , , respectively represent the weight coefficients of inertia, cognitive learning, social learning, and neighbor cooperation; , , respectively represent the acceleration coefficients of cognitive learning, social learning, and neighbor cooperation; represents the set of neighbor devices of device ; represents the total difference between the heat dissipation paths of neighbor devices and the current device's heat dissipation path, which is used to achieve heat dissipation cooperation between adjacent devices.

[0033] In a preferred embodiment, the adaptive heat dissipation control strategy module includes the following steps:

[0034] Define the state space and action space of fan control;

[0035] Design a reward function that comprehensively considers temperature control accuracy, noise level, energy consumption, and group collaboration effect;

[0036] Adopt an improved double deep Q-network structure, including an online network and a target network;

[0037] Based on the learned Q-network, regularly execute fan control decisions.

[0038] In a preferred embodiment, the reward function is calculated by the following formula:

[0039] ;

[0040] Wherein, represents the comprehensive evaluation score of the fan control decision; represents the temperature control reward; represents the noise control reward; represents the energy consumption control reward; represents the group collaboration reward; 、 、 、 respectively represent the importance degrees of temperature, noise, energy consumption, and group collaboration in the total reward.

[0041] In a preferred embodiment, the adaptive heat dissipation control strategy module further includes the following steps:

[0042] Continuously record the performance data during the operation process;

[0043] Based on the collected historical data, use the clustering algorithm to divide the operation environment into multiple typical scenarios;

[0044] Combined with the simulated annealing algorithm, improve the search process of the genetic algorithm.

[0045] In a preferred embodiment, a computer-readable storage medium is used to store computer-readable instructions, and when the computer-readable instructions are read by a computer, a temperature management system of a gallium nitride power adapter can be run.

[0046] The beneficial effects of the present invention are as follows:

[0047] Through the collaborative perception and information sharing among devices, the accurate perception of the overall heat dissipation environment is realized;

[0048] Through the adaptive air duct structure, the targeted heat dissipation is realized;

[0049] Through the optimization of the swarm intelligence heat dissipation path, the efficient diversion of heat is realized;

[0050] Through reinforcement learning for fan control, a multi-objective balance among temperature, noise, and energy consumption is achieved;

[0051] Through rule evolution, continuous self-optimization of the system is achieved. Brief Description of the Drawings

[0052] Figure 1 It is a module diagram of a temperature management system for a gallium nitride power adapter according to the present invention. Detailed Description of the Embodiments

[0053] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0054] In at least one embodiment of the present invention, a temperature management system for a gallium nitride power adapter is disclosed, as Figure 1 shown, including:

[0055] A dynamic thermal model module, which is used to construct a three-dimensional thermal distribution model inside the gallium nitride power adapter in real time, and based on the heat conduction equation and the finite element analysis method, accurately predict the temperature change trends of each key component;

[0056] Including the following:

[0057] Establish a dynamic thermal model including a multi-level thermal resistance and heat capacity network, which accurately describes the thermal characteristics of each component inside the gallium nitride power adapter and their interactions. The dynamic thermal model can be expressed as:

[0058] ;

[0059] Wherein, represents the heat capacity of each component of the system; represents the temperature at each point of the system; is the rate of change of temperature with time; is the power loss vector; represents the heat conduction coefficient between each point inside the system and between the system and the environment; is the ambient temperature.

[0060] This dynamic thermal model can be further enhanced to a time-varying adaptive model by updating the thermal resistance and heat capacity parameters in real time through an online parameter estimation method. The system can use the Recursive Least Squares (RLS) or Kalman filter to estimate the model parameters in real time, expressed as:

[0061] ;

[0062] where, 、 represent the parameter estimation values at time and time respectively; is the gain matrix, controlling the step size and direction of parameter update; is the actual measured value; is the regression vector, containing input variables related to parameter estimation; is the model prediction output based on the previous parameter estimation; is the prediction error, used to correct parameter estimation.

[0063] During the model construction process, the system can adopt various heat conduction models according to the different characteristics of the heat conduction paths. For example, for the inside of electronic components, a three-dimensional heat diffusion equation can be used for fine modeling; for the heat exchange between the radiator and the environment, a convective heat transfer model can be used; for the thermal radiation between components, the Stefan-Boltzmann law can be used for description. This multi-physics field coupling modeling method can more accurately reflect the thermal behavior of the actual system.

[0064] In practical applications, this dynamic thermal model shows strong environmental adaptability. For example, in a high-power gallium nitride charging pile used in an electric vehicle fast charging station, the system needs to adapt to a wide environmental temperature range from -20°C to 45°C.

[0065] Traditional static models often show large deviations under extreme temperature conditions, while the adaptive dynamic thermal model of this system can adjust parameters in real time, accurately capture the changes in material thermal properties and the temperature transient process during the startup phase in a low-temperature environment, and accurately predict the thermal saturation phenomenon of electronic components in a high-temperature environment.

[0066] This model reduces the prediction error in the full temperature range, improves the prediction accuracy, and effectively avoids the risks of overprotection or thermal runaway caused by inaccurate temperature prediction.

[0067] The multi-level temperature monitoring network module is used to collect the temperature data of each key point inside the power adapter in real time through a distributed micro temperature sensor array to form a complete temperature distribution map;

[0068] Specifically, it includes the following content:

[0069] The system designs a multi - level temperature monitoring network, including surface temperature sensors, internal temperature sensors, thermal imagers, etc., to comprehensively monitor the thermal distribution of the gallium nitride power adapter. The multi - level temperature monitoring network can provide accurate thermal distribution data, providing a scientific basis for the formulation of thermal management strategies.

[0070] The temperature monitoring network can adopt a hierarchical architecture, consisting of three levels: the core component layer, the power heat dissipation layer, and the shell surface layer. Different types and precisions of temperature sensors are deployed at each level to meet different monitoring requirements.

[0071] The core component layer uses micro - thermocouples or integrated MEMS temperature sensors, with high precision (±0.1°C) and fast response characteristics (response time < 20ms), capable of accurately capturing the transient temperature changes of key components such as GaN power devices and driver chips;

[0072] The power heat dissipation layer uses an infrared array sensor, which can realize the temperature distribution scanning of heat dissipation components such as radiators and heat pipes, with a resolution of up to 0.5mm;

[0073] The shell surface layer uses a combination of thermal resistors and thermistors, reducing the system cost while ensuring measurement stability.

[0074] The system can also implement sensor redundancy design and dynamic calibration mechanisms. Multiple sensors are set at key temperature points for redundant monitoring, and by comparing the data of different sensors, abnormal readings can be identified and excluded.

[0075] At the same time, the system periodically executes a self - calibration program, using known temperature points as references to dynamically adjust the calibration coefficients of each sensor, ensuring long - term monitoring accuracy.

[0076] In practical applications, this multi - level temperature monitoring network performs excellently. For example, in a gallium nitride server power system deployed in a high - density server room, traditional single - point temperature monitoring schemes often fail to detect local hotspots in a timely manner. However, the multi - level monitoring network of this system can accurately locate the hotspot position and visually present the temperature distribution through the real - time heat map display function.

[0077] In a test, the system successfully identified a local overheating phenomenon caused by a blocked radiator. The warning mechanism was triggered within only 7 seconds after the temperature anomaly occurred, sounding the alarm 43 seconds earlier than the traditional monitoring scheme, effectively avoiding the potential risk of equipment damage.

[0078] The multi - device collaborative perception module is used to collect and share the temperature distribution data, fan status data, and airflow information among multiple gallium nitride power adapters, generating a perception result of the overall heat dissipation environment;

[0079] Specifically, it includes the following content:

[0080] This system uses wireless communication technologies such as Bluetooth Low Energy (BLE), ZigBee, or WiFi to achieve communication and information sharing among multiple gallium nitride power adapters. The adapters share temperature data, power load information, and the status of heat dissipation resources, thus forming a collaborative sensing network.

[0081] The system can adopt a hierarchical communication architecture, including a device-level communication layer and a regional coordination layer.

[0082] The device-level communication layer is responsible for direct data exchange between nearby devices and can use the BLE Mesh network to achieve point-to-point and broadcast communication with low latency (<10ms);

[0083] The regional coordination layer is connected to the central management unit through WiFi or Ethernet to achieve information aggregation and decision-making distribution over a wider range.

[0084] The system applies lightweight compression algorithms and incremental transmission mechanisms during communication, reducing the transmission bandwidth requirements for temperature monitoring data while keeping the data accuracy loss within an acceptable range (<0.2°C).

[0085] The system can also implement an adaptive communication strategy, dynamically adjusting the communication frequency and data priority according to network congestion conditions and data importance. For example, when a sharp rise in temperature is detected, the relevant devices will automatically increase the data transmission frequency and priority to ensure that critical information can be conveyed in a timely manner; while during the temperature stable stage, the communication frequency is reduced to save energy.

[0086] In practical applications, this collaborative sensing and information sharing mechanism performs excellently. In an enterprise conference room scenario, when 12 laptops are using gallium nitride chargers simultaneously, the system identifies the hotspot distribution pattern through information sharing among devices and coordinates the heat dissipation resources of each charger.

[0087] Compared with the traditional independent operation scheme, the collaborative system reduces the highest temperature in the hotspot area and also reduces the overall energy consumption. User feedback shows that the intelligent collaborative features of the system significantly improve the charging experience in high-density usage environments and reduce the risk of device overheating shutdown.

[0088] An adaptive air duct structure module, used to dynamically adjust the internal air duct structure according to real-time thermal gradient distribution data through shape memory materials and microelectromechanical actuators to optimize the air flow path;

[0089] Specifically, it includes the following content:

[0090] The system adopts a deformable air duct structure, which can dynamically adjust the air duct shape and air volume distribution according to the real-time monitored temperature distribution to achieve precise heat dissipation. The adaptive air duct structure includes multiple independently controllable micro fans, adjustable air guide plates, and an electromagnetic-driven air duct switching mechanism, etc.

[0091] The adaptive air duct structure can adopt flexible air guide blades driven by shape memory alloy (SMA), which can undergo controllable deformation under current excitation to change the air flow direction and distribution.

[0092] The system uses a PWM control method to precisely adjust the SMA drive current to achieve continuous adjustment of the air guide blade angle within the range of 0 - 45°.

[0093] At the same time, the air duct structure also includes a main air duct switching valve driven by a micro servo, which can quickly switch between 3 preset air duct modes (switching time < 500ms) to adapt to different heat dissipation demand scenarios.

[0094] The system can also implement the dynamic optimization function of the air duct structure. Through computational fluid dynamics (CFD) simulation and digital twin technology, the system can real-time predict the air flow distribution and heat dissipation effect under different air duct configurations and select the optimal air duct structure.

[0095] The optimization algorithm uses an improved particle swarm optimization method to complete the air duct configuration optimization calculation within an average of 3.2 seconds to meet the real-time adjustment requirements.

[0096] In practical applications, this adaptive air duct structure performs excellently. In the gallium nitride power supply system of a portable high-performance computing workstation, the traditional fixed air duct design either has insufficient heat dissipation or energy waste under different load conditions.

[0097] And the adaptive air duct of this system can adjust the heat dissipation strategy in real-time according to the change of the computing load of the workstation:

[0098] When in light load, the system automatically switches to the low-noise mode, reducing the fan noise from 38dB to 26dB;

[0099] When in heavy load, the system quickly switches to the high-efficiency heat dissipation mode, controlling the temperature of key components within a safe range, and saving energy compared with the traditional fixed air duct design. This intelligent adaptive characteristic significantly improves user satisfaction and system reliability.

[0100] Step 5, generate a heat dissipation path based on swarm intelligence;

[0101] Based on an embodiment of the present application, this step applies the swarm particle collaborative optimization algorithm. Based on the device topology network and the thermal state information of each device, it calculates and generates an optimal multi-device collaborative heat dissipation path to achieve efficient heat dissipation. It should be understood that the specific implementation process of this step is as follows:

[0102] The heat dissipation path module based on swarm intelligence calculates and generates an optimal multi-device collaborative heat dissipation path based on the device topology network and the thermal state information of each device by applying the swarm intelligence algorithm;

[0103] Specifically, it includes the following content:

[0104] The system constructs an objective function for overall heat dissipation optimization according to the temperature distribution and heat dissipation requirements of multiple adapters:

[0105] ;

[0106] Wherein, represents the objective function of overall heat dissipation optimization, represents the set of heat dissipation path vectors of all devices, and each vector describes the heat dissipation direction and intensity of the corresponding device; is the average temperature of device ; is the target temperature; is the weight coefficient of device , which is determined according to its heat load and heat dissipation importance. The larger the value, the more important the temperature control of this device; is the collaborative consistency weight, which controls the consistency requirement of the heat dissipation paths of adjacent devices. The larger the value, the higher the requirement for the coordination of the heat dissipation paths of adjacent devices; is the total number of devices in the system; represents the set of neighbor devices of device ; represents the square difference between the actual temperature and the target temperature of device , which is used to quantify the degree of temperature deviation; represents the square difference between the heat dissipation path vector of device and its neighbor device , which is used to quantify the degree of inconsistency of the heat dissipation paths; is the first term of the objective function, which represents the weighted sum of the squares of the deviations between the temperatures of all devices and the target temperature, and promotes each device to reach the target temperature; is the second term of the objective function, which represents the sum of the squares of the differences in the heat dissipation paths between all adjacent devices, and promotes the heat dissipation paths of adjacent devices to be coordinated and consistent to avoid air flow conflicts.

[0107] Initialization of particle swarm position and velocity:

[0108] The system initializes a heat dissipation path vector for each adapter , which represents the direction and intensity of the device's heat dissipation air flow. The initialization process takes into account the physical location and current thermal state of the device:

[0109] ;

[0110] wherein, represents the initial heat dissipation path vector of the device ; represents the negative gradient direction of the local temperature field of the device , that is, the natural heat flow direction, pointing to the direction where the temperature drops fastest; is a small random perturbation vector, used to increase the diversity of the initial solution and prevent the algorithm from falling into a local optimal solution.

[0111] Meanwhile, the system initializes the optimal position record of each device (the initial value is equal to ), representing the best heat dissipation path found by the device historically; and the global optimal position record (the initial value is the one that makes the objective function the smallest among all ), representing the best heat dissipation path configuration found among all devices.

[0112] In some embodiments, various strategies can be adopted for the initialization of the heat dissipation path vector. For example, a memory-based initialization method can be used according to the historical heat dissipation effect data of the device. That is, when the current environment is more similar to a certain historical environment than a threshold, the optimal heat dissipation path in that historical environment is directly used as the initial solution to accelerate the convergence process. Another optional method is to adopt an initialization strategy based on the device layout. Analyze the geometric arrangement of the devices in advance, calculate the ideal heat dissipation direction as the dominant direction of the initial vector, and then add a small random perturbation.

[0113] The random perturbation term can be generated using different distributions.

[0114] In the early exploration stage, a uniform distribution can be used, where is the perturbation amplitude parameter, representing the maximum value of the perturbation, set to 10%-20% of the temperature gradient magnitude, used to control the range of the random perturbation;

[0115] In the later fine-tuning stage, a normal distribution can be used, where is the standard deviation parameter, representing the degree of dispersion of the random perturbation, set to 5%-10% of the temperature gradient magnitude, used to control the concentration of the perturbation.

[0116] Uniform distribution Indicates that to the probability of occurrence of random values within the range is equal, while the normal distribution indicates that the random values are bell-shaped distributed around the mean of 0, where is the variance, which determines the width of the distribution.

[0117] This adaptive perturbation generation strategy helps to maintain a large exploration range in the early stage of the algorithm, enabling the system to widely search for possible heat dissipation path solutions. In the later stage, it focuses on the fine optimization of the solution, fine-tuning the found better heat dissipation paths to improve the accuracy of the final solution.

[0118] In scenarios with high-density electronic devices such as multimedia classrooms, this particle swarm position initialization system exhibits excellent adaptability. For example, in a classroom containing 30 computers and 8 gallium nitride power adapters, the system can quickly identify the temperature distribution characteristics in different areas (such as near the projector, near the window or door), and accordingly initialize the heat dissipation paths of the adapters.

[0119] For the high-temperature area in the center of the classroom, the heat dissipation path initialized by the system will point towards the cooler window or door direction;

[0120] For the adapters located at the edge, the system will initialize heat dissipation paths towards the outside of the classroom to prevent heat from accumulating inside the classroom. Actual measurements show that this intelligent initialization strategy can shorten the convergence time compared to simple gradient reverse initialization and achieve an improvement in heat dissipation efficiency in the final solution.

[0121] Iterative optimization of heat dissipation paths:

[0122] The system uses an improved particle swarm optimization algorithm to iteratively update the heat dissipation path vector of each device:

[0123] ;

[0124] where, represents the heat dissipation path vector of device after the th iteration; , respectively represent the heat dissipation path vectors of device and device at the th iteration; represents the historical optimal heat dissipation path vector of device ; represents the set of neighbor devices of device ; is the inertia weight, which controls the tendency of the path to maintain the original direction; and respectively represent the individual cognitive weight and the acceleration coefficient, and control the degree of approaching the historical optimal path of the device itself; and respectively represent the social learning weight and the acceleration coefficient, and control the degree of approaching the global optimal path; and respectively represent the neighbor cooperation weight and the acceleration coefficient, and control the degree of path coordination with adjacent devices. Through this multi-level optimization mechanism, the system can achieve collaborative optimization of the group while maintaining local heat dissipation efficiency.

[0125] In a specific embodiment of the present application, the parameter settings of the group particle collaborative optimization algorithm are as follows:

[0126] Inertia weight The initial value is 0.8 and linearly decreases to 0.4 with the number of iterations;

[0127] Individual cognitive weight Set to 0.6, social learning weight Set to 0.4;

[0128] Neighbor cooperation weight Dynamically adjusted according to the device density, set to 0.2 in the sparse device scenario and increased to 0.5 in the dense device scenario;

[0129] Acceleration coefficient , and Are set to 1.5, 1.2, and 0.8 respectively.

[0130] The improved particle swarm algorithm also includes an adaptive congestion control mechanism. When it is detected that there is an excessive concentration of devices in a local area, the repulsive force between the devices in this area is automatically increased to avoid air flow interference caused by over-concentration of heat dissipation paths. The calculation formula is:

[0131] ;

[0132] Where, represents the repulsive force vector of device on device to avoid over-concentration of heat dissipation paths, is the physical distance between device and device , is the unit vector from device to device , is the repulsive force coefficient, with an initial value of 0.15 and linearly increasing to 0.4 when the local device density exceeds the preset threshold; It represents the characteristic that the repulsive force decays inversely with the square of the distance. The closer the distance, the greater the repulsive force. Indicates the device Points to the device The unit vector of determines the direction of the repulsive force. This repulsive force term is added to the particle update formula to form a more balanced heat dissipation path distribution.

[0133] In practical applications, the system quantifies the heat dissipation paths into 8 basic directions and 3 wind force intensity levels, forming a total of 24 discrete action options, enabling the control instructions to be effectively understood and implemented by the fan execution system. During the iterative process of the algorithm, the physical layout constraints of the devices are considered to avoid generating infeasible heat dissipation paths.

[0134] Path Convergence and Implementation:

[0135] The system continuously updates the optimal position records of each device and the global optimal position record :

[0136] ;

[0137] Among them, Indicates the historical optimal heat dissipation path vector of the device ; Indicates the device At the The heat dissipation path vector after the iteration; Indicates that the heat dissipation path of the device Is updated to The objective function value after; Indicates the use of the device The objective function value when the historical optimal heat dissipation path is used. This formula means that if the new heat dissipation path vector can reduce the objective function value (i.e., the heat dissipation effect is better), the historical optimal record is updated, otherwise it remains unchanged.

[0138] ;

[0139] Among them, Indicates the global optimal heat dissipation path configuration, that is, the heat dissipation path configuration that makes the objective function Obtain the minimum value among all devices; Indicates finding the parameter that minimizes the objective function The minimum value of . This formula means that the global optimal position record is the one that selects the minimum objective function value among the historical optimal positions of all devices.

[0140] When the iteration converges (the improvement amplitude of the objective function in consecutive iterations is less than the preset threshold) or reaches the maximum number of iterations, the system distributes the final heat dissipation path vector to each device to guide fan control and duct adjustment. Here, the "preset threshold" refers to a small positive number preset by the system to determine whether the algorithm has converged to a stable solution;

[0141] The "maximum number of iterations" refers to the maximum number of calculation cycles allowed by the algorithm, usually set to 100 to 200 times, to prevent the algorithm from looping infinitely.

[0142] Therefore, through this collaborative optimization based on swarm intelligence, the system can achieve efficient collaborative guidance of heat flow in a group consisting of 3 to 8 devices, significantly improving the overall heat dissipation efficiency.

[0143] In practical applications, this collaborative heat dissipation technology performs excellently. For example, in an edge computing node of a data center, 6 servers equipped with gallium nitride power adapters are deployed. In the traditional scheme, each device manages heat dissipation independently, which easily leads to heat accumulation in certain areas.

[0144] However, the swarm intelligence heat dissipation path optimization function of this system can form a "heat wave pushing effect", synergistically pushing heat from high-temperature areas to areas with better heat dissipation conditions.

[0145] In a high-load test, the collaborative system reduced the highest temperature point in the cabinet, while narrowing the temperature difference, significantly reducing the damage to devices caused by thermal cycling. More importantly, the system reduced the average fan energy consumption of the edge computing node while maintaining the same computing performance, improving the overall energy efficiency level.

[0146] The adaptive heat dissipation control strategy module, based on environmental perception data and heat dissipation paths, uses a deep reinforcement learning algorithm to intelligently adjust the fan operation parameters to achieve multi-objective balance optimization of temperature, noise, and energy consumption.

[0147] Specifically, it includes the following:

[0148] Environmental state representation:

[0149] Define the environmental state space, including the following key features:

[0150] Multi-point temperature sequence: , where The temperature values of , , respectively represent the temperature values of the , , th key monitoring points; represents the number of key monitoring points;

[0151] Power load status: , indicating load information of , , respectively represent the load information of the , , th power levels; represents the number of power levels;

[0152] Environmental conditions: , where , , respectively represent the environmental temperature, humidity and air flow conditions;

[0153] Equipment running time: , indicating the current continuous running time;

[0154] Historical control effect: , indicating the actions and rewards of the past time steps, where , , respectively represent the heat dissipation control actions executed by the system at the , , th time steps; , , respectively represent the reward values obtained by the system after executing the heat dissipation control actions at the , , th time steps; represents the length of the historical record;

[0155] Feature engineering is performed on the state features, including time series feature extraction (such as temperature change rate, maximum temperature, temperature difference, etc.) and spatial distribution feature extraction (such as hot spot clustering, temperature gradient, etc.). The processed features are normalized so that the range of feature values in each dimension is unified within the interval, which is beneficial to the training of the neural network.

[0156] The system can also apply an autoencoder to the state space for dimensionality reduction, compressing the original high-dimensional state into a low-dimensional latent space to reduce the learning difficulty.

[0157] The autoencoder adopts a stacked fully connected layer structure. The encoder part compresses from the original 128-dimensional features to 32-dimensional latent representations, and the decoder part reconstructs back to the original dimension.

[0158] After pre-training, the autoencoder improves the reconstruction accuracy in the state reconstruction task, reduces the feature space, and effectively reduces the complexity of reinforcement learning.

[0159] Thermal management action space:

[0160] The system defines the thermal management action space, including:

[0161] Fan speed control: , representing the rotation speed settings of , , representing the rotation speed values of the , , th, represents the number of fans;

[0162] Air duct switching decision: , representing the switch states of , , representing the state values of the , , th air duct components, with values of 0 (closed) or 1 (open); represents the number of air duct components;

[0163] Power limit strategy: , representing the limit levels of , , representing the limit levels of the , , th power modules, with a value range of 0 - 3, corresponding to no limit, mild limit, moderate limit, and severe limit respectively; represents the number of power modules;

[0164] To simplify the control complexity, the system discretizes the above continuous action space into a finite number of action combinations, forming a total of 128 thermal management strategy options. These options are pre-designed through expert knowledge to cover various thermal management scenario requirements.

[0165] The system can also implement a hierarchical design of the action space.

[0166] The top-level control strategy selector is responsible for switching among the three modes of "energy-saving mode", "balanced mode", and "high-performance mode";

[0167] The middle - layer controller is responsible for determining the cooperative control mode of the fan group;

[0168] The bottom - layer actuator is responsible for precisely adjusting the rotational speed of each fan and the position of the air duct components.

[0169] This hierarchical design enables the system to reduce the complexity of the decision - making space while maintaining sufficient flexibility.

[0170] Reward function design:

[0171] The multi - objective reward function of the system design considers three aspects: heat dissipation effect, energy consumption, and noise:

[0172] ;

[0173] Among them, is the total reward function, indicating the comprehensive reward value obtained by executing action under state ; represents the current environmental state of the system, including information such as temperature distribution and power load; represents the heat dissipation control action taken by the system, such as fan speed adjustment, air duct switching, etc.; is the temperature control reward, indicating the system's evaluation of the temperature control effect; is the energy consumption control reward, indicating the system's evaluation of the energy consumption control effect; is the noise control reward, indicating the system's evaluation of the noise control effect; 、 and represent the importance weights of temperature control, energy consumption control, and noise control in the overall reward respectively, which are dynamically adjusted according to different usage scenarios and user preferences, and satisfy , to ensure the normalization of the weights of the three aspects.

[0174] The temperature control reward adopts the following formula:

[0175] ;

[0176] Among them, represents the temperature control reward value obtained by executing action under state ; is the current highest temperature, indicating the value of the highest temperature point monitored in the system; is the temperature threshold, indicating the highest safe temperature allowed by the system; represents the square penalty for the highest temperature exceeding the threshold, which is 0 when the temperature does not exceed the threshold; is the standard deviation of the temperature distribution, representing the temperature uniformity. The smaller the value, the more uniform the temperature distribution; and represent the weight coefficients of the highest temperature over-limit penalty and the temperature non-uniformity penalty respectively; indicates that these factors reduce the overall reward value as penalty terms. The better the temperature control, the smaller the penalty and the higher the reward value.

[0177] The calculation formula for the energy consumption control reward is:

[0178] ;

[0179] where represents the energy consumption control reward value obtained by executing the action in the state ; represents the power consumption of the fan at the rotational speed ; is the energy consumption weight coefficient, controlling the intensity of the energy consumption penalty. The larger the value, the higher the degree of attention of the system to energy consumption; represents the total number of fans in the system; represents the fan number, ranging from 1 to ; represents the summation of the power consumption of all fans; indicates that the energy consumption reduces the overall reward value as a penalty term. The lower the energy consumption, the smaller the penalty and the higher the reward value.

[0180] The calculation formula for the noise control reward is:

[0181] ;

[0182] where represents the noise control reward value obtained by executing the action in the state ; represents the noise decibel value generated by the fan at the rotational speed ; represents the rotational speed value of the th fan; represents the total number of fans in the system; is the noise weight coefficient, controlling the intensity of the noise penalty. The larger the value, the higher the degree of attention of the system to noise control; represents the operation of taking the maximum noise value among all fans, because human ear perception is usually affected by the maximum noise source; indicates that the noise reduces the overall reward value as a penalty term. The smaller the noise, the smaller the penalty and the higher the reward value.

[0183] In practical applications, the reward function also includes additional smoothing terms and sparsity control terms to avoid frequent jumps in the control strategy:

[0184] ;

[0185] where denotes the smooth control reward value obtained by executing action under state ; denotes the rotational speed of fan at the current time step ; denotes the rotational speed of fan at the previous time step ; denotes the absolute value of the change in fan rotational speed, which is used to quantify the degree of rotational speed fluctuation; is the smoothing penalty coefficient, which controls the intensity of the rotational speed change penalty. The larger the value, the higher the system's requirement for rotational speed stability; denotes the total number of fans in the system; denotes the fan number, ranging from 1 to ; denotes the summation of the rotational speed changes of all fans; denotes that the rotational speed change reduces the overall reward value as a penalty term. The smaller the rotational speed change, the smaller the penalty and the higher the reward value, thus encouraging the system to maintain a stable control strategy.

[0186] Through this multi-objective reward design, the system can balance energy efficiency and user experience while meeting the heat dissipation requirements.

[0187] Implementation of the deep reinforcement learning algorithm:

[0188] The system uses an improved deep reinforcement learning Q-network (DeepQ-Network, DQN) algorithm to implement heat dissipation control strategy learning. The network architecture consists of 3 layers of MLP, including an input layer with 256 nodes, hidden layers with 128 nodes and 64 nodes, and an output layer with the number of actions. The network is optimized using the Adam optimizer, with the learning rate set to 0.001 and the discount factor set to 0.95.

[0189] To improve learning efficiency and policy stability, the system adopts the following improvement techniques:

[0190] Dual network structure: including an evaluation network and a target network, and the target network is updated every 100 training steps;

[0191] Priority experience replay: Prioritize experience sampling based on TD error size to improve learning efficiency;

[0192] Noise Injection Exploration: Adding Ornstein-Uhlenbeck noise during action selection to facilitate exploration of policy space.

[0193] In addition, the system also implements the transfer learning function, which migrates the pre-trained strategies in the simulation environment to the actual device, accelerating the learning process in practical applications.

[0194] During the migration process, a layered fine-tuning strategy was adopted, first fixing the underlying feature extraction part of the network and only fine-tuning the decision layer parameters; then gradually unlocking more layers for fine-tuning based on performance feedback. This approach can quickly adapt to the characteristics of specific devices while retaining general heat dissipation knowledge.

[0195] This adaptive heat dissipation control strategy performs well in the scenario of electric vehicle charging stations, where high-power GaN charging piles need to maintain stable operation under outdoor environments and changing charging loads.

[0196] Traditional rule-based cooling control solutions cannot adapt to complex environmental changes. The reinforcement learning control strategy of this system can automatically adjust the optimal cooling strategy according to different seasons and charging power:

[0197] In the high temperature environment in summer, the system will start heat dissipation and cooling in advance;

[0198] In winter, when the temperature is low, the system uses a more energy-efficient intermittent cooling mode.

[0199] Actual measurements show that compared with traditional control schemes, the reinforcement learning control system reduces the average temperature fluctuation of the charging piles, while saving heat dissipation energy consumption during the annual operation cycle, significantly improving the reliability and service life of the charging equipment.

[0200] The technical effects of this implementation are as follows:

[0201] The temperature management system of a gallium nitride power adapter provided in this embodiment has achieved remarkable technical effects in the temperature management of gallium nitride power adapters by adopting technical means such as dynamic thermal model, multi-level temperature monitoring network, multi-device collaborative perception and information sharing, adaptive air duct structure, heat dissipation path based on swarm intelligence and adaptive heat dissipation control strategy:

[0202] Improved heat dissipation efficiency: Compared with traditional static heat dissipation systems, the system's dynamic thermal model and adaptive air duct structure can accurately identify hot spots and dissipate heat in a targeted manner, thereby improving heat dissipation efficiency.

[0203] Significant reduction in energy consumption: The adaptive control strategy based on deep reinforcement learning flexibly adjusts the cooling intensity according to actual demands, minimizing energy consumption while meeting the cooling requirements.

[0204] Substantial reduction in noise: The intelligent air duct structure and the control algorithm based on machine learning can minimize noise while meeting the cooling requirements.

[0205] Enhanced system adaptability: The dynamic thermal model combined with the multi-sensor monitoring network enables the system to adapt to different environmental conditions and usage scenarios. Within the environmental temperature range of -10°C to 45°C, the system can maintain the best cooling effect, and its adaptability is improved compared with traditional systems.

[0206] Improved device reliability: Through precise temperature control and thermal distribution equalization, the system can significantly reduce thermal stress and minimize the damage to components caused by thermal cycling. Long-term tests show that this system extends the expected lifespan of gallium nitride power devices and greatly reduces the failure rate caused by overheating.

[0207] Expanded application scenarios: The multi-device collaborative perception and the cooling path optimization based on swarm intelligence enable the system to adapt to high-density device deployment scenarios. In environments such as data centers and server rooms, the collaborative cooling effect between adjacent devices is improved compared with independent cooling, providing the possibility for the application of gallium nitride power adapters in more professional scenarios.

[0208] Increased system integration: The intelligent control of the adaptive temperature management system enables the gallium nitride power adapter to dissipate heat safely in a smaller volume, supporting a higher power density. This results in an increase in power density in the same volume, providing technical support for the development of portable high-power adapters.

[0209] Optimized user experience: The system automatically adjusts the cooling strategy based on the device usage pattern, maintaining the best performance and user experience in different load scenarios. For example, it automatically switches to the low-noise mode in quiet environments such as conference room presentations, and gives priority to ensuring the cooling effect in high-density computing environments, intelligently balancing various indicators.

[0210] Through the above technical effects, the temperature management system provided by this embodiment comprehensively improves the performance, reliability, and user experience of the gallium nitride power adapter, providing key technical support for the development and application of high-efficiency, small-size, and high-reliability power adapters.

[0211] The above describes the embodiments of the present invention. However, these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A temperature management system for a gallium nitride power adapter, characterized in that, Including: A dynamic thermal model module for constructing a three-dimensional thermal distribution model inside a gallium nitride power adapter in real time, and accurately predicting the temperature change trends of key components based on the heat conduction equation and the finite element analysis method; A multi-level temperature monitoring network module for collecting temperature data at key points inside the power adapter in real time through a distributed micro temperature sensor array to form a complete temperature distribution map; A multi-device collaborative perception module for collecting and sharing temperature distribution data, fan status data, and airflow information among multiple gallium nitride power adapters, and generating a perception result of the overall heat dissipation environment, including: Collecting temperature distribution data at key points of itself through a built-in temperature sensor network, and dynamically adjusting the sampling frequency of each temperature sensor according to the temperature change gradient; Dynamically adjusting the information exchange frequency between devices according to the temperature change rate; Using a proximity device discovery algorithm based on signal strength to identify adjacent adapter devices in the physical space through near-field communication technology; Applying a weighted data fusion algorithm to generate a comprehensive state representation of the local environment; An adaptive air duct structure module for dynamically adjusting the internal air duct structure according to real-time thermal gradient distribution data through shape memory materials and microelectromechanical actuators to optimize the airflow path, including: Constructing a micro air duct network made of shape memory polymer materials, and the air ducts are in a uniformly distributed grid structure in the initial state; Calculating the thermal gradient vector field in the space based on the temperature distribution data; Calculating the deformation state of the air duct structure based on the heat response equation; Converting the calculated air duct structure deformation instruction into an electric control signal, and precisely controlling the local deformation of the shape memory polymer material through a micro actuator network; A heat dissipation path module based on swarm intelligence, which calculates and generates an optimal multi-device collaborative heat dissipation path based on the device topology network and the thermal state information of each device by applying swarm intelligence algorithms; An adaptive heat dissipation control strategy module, which intelligently regulates the fan operation parameters by using a deep reinforcement learning algorithm based on the environmental perception data and the heat dissipation path to achieve multi-objective balance optimization of temperature, noise, and energy consumption.

2. The temperature management system of a gallium nitride power adapter according to claim 1, characterized in that, The information exchange frequency between devices is calculated according to the following formula: ； Among them, represents the frequency of information exchange between devices; represents the lowest communication frequency of the system in a stable state; represents the adjustment coefficient, which is used to control the influence degree of temperature change on the communication frequency; represents the absolute magnitude of temperature change per unit time; When the temperature changes violently, the communication frequency will increase accordingly to ensure that the system can respond to the changes in the thermal environment in a timely manner.

3. The temperature management system of a gallium nitride power adapter according to claim 1, characterized in that, The swarm particle collaborative optimization algorithm adopted in the heat dissipation path module based on swarm intelligence includes the following steps: Constructing an objective function for overall heat dissipation optimization according to the temperature distribution and heat dissipation requirements of multiple adapters; Initializing a heat dissipation path vector for each adapter, and the heat dissipation path vector represents the direction and intensity of the device heat dissipation airflow; Adopting an improved particle swarm optimization algorithm to iteratively update the heat dissipation path vector of each device; Continuously updating the optimal position record of each device and the global optimal position record during the iteration process.

4. The temperature management system of a gallium nitride power adapter according to claim 3, characterized in that, The improved particle swarm optimization algorithm updates the heat dissipation path vector through the following formula: ； Among them, represents the heat dissipation path vector of the device at time ; represents the updated heat dissipation path vector of the device at time ; represents the device's own historical optimal path; represents the global optimal path; , , , respectively represent the weight coefficients of inertia, cognitive learning, social learning, and neighbor cooperation; , , respectively represent the acceleration coefficients of cognitive learning, social learning, and neighbor cooperation; represents the set of neighbor devices of the device ; represents the total difference between the heat dissipation paths of neighbor devices and the heat dissipation path of the current device, which is used to achieve heat dissipation cooperation between adjacent devices.

5. The temperature management system of a gallium nitride power adapter according to claim 1, characterized in that, The adaptive heat dissipation control strategy module includes the following steps: Defining the state space and action space of fan control; Designing a reward function that comprehensively considers temperature control accuracy, noise level, energy consumption, and group collaboration effect; Adopting an improved double deep Q network structure, including an online network and a target network; Based on the learned Q-network, the fan control decision is periodically executed.

6. The temperature management system of a gallium nitride power adapter according to claim 5, characterized in that, The reward function is calculated by the following formula: ； Among them, represents the comprehensive evaluation score of the fan control decision; represents the current environmental state of the system; represents the heat dissipation control action taken by the system; represents the temperature control reward; represents the noise control reward; represents the energy consumption control reward; represents the group collaboration reward; 、 、 、 respectively represent the importance degrees of temperature, noise, energy consumption, and group collaboration in the total reward.

7. The temperature management system of a gallium nitride power adapter according to claim 1, wherein, The adaptive heat dissipation control strategy module further includes the following steps: Continuously record the performance data during operation; Based on the collected historical data, use the clustering algorithm to divide the operating environment into multiple typical scenarios; Combined with the simulated annealing algorithm, improve the search process of the genetic algorithm.

8. A computer-readable storage medium, characterized in that, It is used to store computer-readable instructions, and when the computer-readable instructions are read by a computer, it can run a temperature management system of a gallium nitride power adapter as described in any one of claims 1-7.

Images