Flight case liquid cooling heat dissipation temperature control optimization method, system and equipment

Abstract

The invention relates to the technical field of temperature control optimization, and discloses an aviation case liquid cooling heat dissipation temperature control optimization method, system and equipment, and the method comprises the steps: collecting instantaneous temperature data, temperature change rate data and thermal load data through a temperature sensor array on the surface of an aviation case; calculating to obtain an instantaneous temperature deviation value, a temperature change rate deviation value and a thermal load deviation value; generating a temperature tracking prediction value based on the instantaneous temperature deviation value, the temperature change rate deviation value and the thermal load deviation value; and liquid cooling heat dissipation parameter collaborative optimization is executed based on the temperature tracking predicted value, an optimal parameter control strategy is determined, and the flight box liquid cooling pump rotating speed, the adjusting valve opening degree and the heat dissipation fan frequency are synchronously adjusted according to the optimal parameter control strategy, so that intelligent collaborative optimization of the liquid cooling pump rotating speed, the adjusting valve opening degree and the heat dissipation fan frequency is realized; and the stability and the reliability of the aviation case liquid cooling heat dissipation temperature control system are improved.

Description

Technical Field

[0001] This invention relates to the field of temperature control optimization technology, and in particular to a method, system and equipment for optimizing temperature control of liquid cooling heat dissipation in aircraft cases. Background Technology

[0002] Traditional liquid cooling temperature control systems for aircraft cases typically employ single-point temperature monitoring and fixed-parameter control strategies, which cannot accurately identify multi-source temperature deviations under complex operating conditions. This leads to insufficient temperature control accuracy and lag in response. Existing technologies lack the comprehensive ability to identify instantaneous temperature deviations, temperature change rate deviations, and heat load deviations when facing dynamic changes in the moving conditions of aircraft cases. This prevents the temperature control system from responding promptly to temperature changes, severely impacting heat dissipation efficiency and system stability. Particularly in low-temperature environments, the lack of an effective preheating tracking mechanism results in long system startup times and delayed temperature control tracking. Summary of the Invention

[0003] This invention provides a method, system, and device for optimizing liquid cooling temperature control in aircraft cases. This invention achieves intelligent collaborative optimization of liquid cooling pump speed, regulating valve opening, and cooling fan frequency, thereby improving the stability and reliability of the liquid cooling temperature control system for aircraft cases.

[0004] In a first aspect, the present invention provides a method for optimizing the liquid cooling heat dissipation and temperature control of an aircraft case, the method comprising: Instantaneous temperature data, temperature change rate data, and heat load data are collected by an array of temperature sensors on the surface of the flight case, and instantaneous temperature deviation value, temperature change rate deviation value, and heat load deviation value are calculated. An error comprehensive evaluation vector is generated based on the instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value; The error comprehensive evaluation vector is input into the predictive temperature control tracking algorithm, and combined with the characteristics of the flight case movement conditions, a temperature tracking prediction value is generated. Based on the temperature tracking and prediction values, perform collaborative optimization of liquid cooling parameters to determine the optimal parameter control strategy, and synchronously adjust the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy.

[0005] In conjunction with the first aspect, in a first implementation of the first aspect of the present invention, the step of collecting instantaneous temperature data, temperature change rate data, and heat load data through a temperature sensor array on the surface of the aircraft case, and calculating the instantaneous temperature deviation value, temperature change rate deviation value, and heat load deviation value, includes: By arranging a temperature sensor array on the surface of the flight case to collect instantaneous temperature data of each monitoring point in real time, the weighting coefficient is determined according to the positional relationship between each sensor and the liquid cooling pipeline, and the weighted average temperature value is calculated. The difference between the weighted average temperature value and the target temperature value is calculated to obtain the instantaneous temperature deviation value. Collect temperature change data and calculate the temperature change rate data through a preset sliding window. Calculate the difference between the temperature change rate data and the reference temperature change rate to obtain the temperature change rate deviation value. Temperature difference data and flow data of liquid cooling medium inlet and outlet are collected by temperature sensors and flow sensors, respectively. Heat load data are calculated based on the density and specific heat capacity of liquid cooling medium. The difference between the heat load data and the theoretically calculated heat dissipation load value is calculated to obtain the heat load deviation value.

[0006] In conjunction with the first aspect, in a second implementation of the first aspect of the present invention, the step of generating a comprehensive error evaluation vector based on the instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value includes: The instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value are vector-combined to obtain the initial error vector; The current ambient temperature and working status data of the flight case are obtained through the GPS positioning module, and a dynamic weighting coefficient combination is determined based on the current ambient temperature and the working status data. The initial error vector is squared and multiplied with the corresponding dynamic weight coefficient in the dynamic weight coefficient combination to obtain multiple calculation results; The multiple calculation results are input into the softmax function for dynamic weight allocation to obtain normalized weight allocation results. The normalized weight allocation results are then vectorized through the AI ​​edge computing gateway to obtain the comprehensive error evaluation vector.

[0007] In conjunction with the first aspect, in a third implementation of the first aspect of the present invention, the step of inputting the error comprehensive evaluation vector into the predictive temperature control tracking algorithm and generating a temperature tracking prediction value by combining the flight case movement condition characteristics includes: The error comprehensive evaluation vector is used as the temperature prediction correction parameter input into the predictive temperature control tracking algorithm. The algorithm is then fitted with historical temperature data using the least squares method to obtain the basic data of temperature change trend. Based on the aforementioned temperature change trend data, a mobile operating condition adaptive parameter combination is calculated using a mobile operating condition learning algorithm. When the ambient temperature is between the first temperature value and the second temperature value, the current temperature value and the error comprehensive evaluation vector and its rate of change are calculated according to the temperature change trend prediction function to obtain the temperature tracking prediction value under normal temperature conditions. When the ambient temperature is lower than the first temperature value, the preheating tracking algorithm is activated and the preheating power is adjusted based on the heat load deviation value in the error comprehensive evaluation vector. At the same time, ethylene glycol antifreeze is injected into the liquid cooling pipeline, and pulse heating is used for rapid ice breaking to obtain the temperature tracking prediction value for low-temperature start-up. When the ambient temperature is higher than the second temperature value, the temperature tracking prediction value for high-temperature conditions is calculated based on the instantaneous temperature deviation value and heat load deviation value in the error comprehensive evaluation vector.

[0008] In conjunction with the first aspect, in the fourth implementation of the first aspect of the present invention, the step of activating the preheating tracking algorithm and adjusting the preheating power based on the heat load deviation value in the error comprehensive evaluation vector when the ambient temperature is lower than the first temperature value, while simultaneously injecting ethylene glycol antifreeze into the liquid cooling pipeline and using pulse heating for rapid ice breaking to obtain the temperature tracking prediction value for low-temperature start-up, includes: When the ambient temperature is lower than the first temperature value, the preheating tracking algorithm is started and the ambient temperature and the preheating start time are used as initial parameters. The preheating temperature is calculated based on the preheating tracking function to obtain the basic value of the preheating temperature. The preheating power is dynamically adjusted based on the heat load deviation value in the error comprehensive evaluation vector to obtain the dynamically adjusted preheating power value. Based on the dynamically adjusted preheating power value, ethylene glycol antifreeze is injected into the liquid cooling pipeline through a solenoid valve, while the internal temperature change of the liquid cooling pipeline is monitored to obtain pipeline temperature data after antifreeze injection. Based on the pipeline temperature data after the antifreeze injection, a pulse heating method is used for rapid ice breaking, and the ice breaking heating process is executed by the pulse heating controller according to the dynamically adjusted preheating power value to obtain the temperature change data after ice breaking heating. The preheating temperature baseline value is fused with the temperature change data after ice breaking and heating, and the temperature prediction is corrected by combining the preheating tracking function to obtain the temperature tracking prediction value for low-temperature start-up.

[0009] In conjunction with the first aspect, in a fifth implementation of the first aspect of the present invention, the step of performing collaborative optimization of liquid cooling parameters based on the temperature tracking prediction value, determining the optimal parameter control strategy, and synchronously adjusting the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy includes: Based on the temperature tracking prediction value, combined with the liquid cooling flow deviation measured by the electromagnetic flowmeter and the pressure fluctuation data measured by the pressure sensor, a three-dimensional optimization state space for the coordinated optimization of liquid cooling heat dissipation parameters is constructed. The performance of the three-dimensional optimization state space is evaluated by optimizing the multi-objective reward function to obtain a comprehensive performance evaluation value. The comprehensive performance evaluation value is used as a reward signal and input into the SAC reinforcement learning decision network for processing, and the optimal parameter control strategy is output. The optimal parameter control strategy is synchronously sent to each sub-controller via the CAN bus. The liquid cooling pump controller adjusts the speed of the liquid cooling pump in the aircraft case, the regulating valve controller adjusts the valve opening, and the fan controller adjusts the frequency of the cooling fan.

[0010] In conjunction with the first aspect, in a sixth implementation of the first aspect of the present invention, the step of inputting the comprehensive performance evaluation value as a reward signal into the SAC reinforcement learning decision network for processing and outputting an optimal parameter control strategy includes: The three-dimensional optimized state space is used as the state input, and the comprehensive performance evaluation value is used as the reward signal input to the Actor network of the SAC reinforcement learning decision network. Forward propagation calculation is performed through the three-layer neural network structure of the Actor network to generate the action probability distribution. Based on the action probability distribution, candidate action vectors are generated through a Gaussian distribution sampling mechanism. At the same time, the three-dimensional optimized state space and the candidate action vectors are input into the dual Critic network of the SAC reinforcement learning decision network to calculate the state-action value function and obtain the target Q-value evaluation result. Based on the target Q-value evaluation result and the comprehensive performance evaluation value, the strategy loss function and the value loss function are calculated. The parameters of the Actor network and the dual Critic network are updated respectively using the gradient descent algorithm to obtain the optimized network parameters. Based on the optimized network parameters, the Actor network outputs the original control actions, including the liquid cooling pump speed adjustment, the electric regulating valve opening increment, and the cooling fan frequency change. The original control actions are then transformed by action space mapping to obtain the optimal parameter control strategy.

[0011] In conjunction with the first aspect, in the seventh implementation of the first aspect of the present invention, the step of generating candidate action vectors based on the action probability distribution through a Gaussian distribution sampling mechanism, and simultaneously inputting the three-dimensional optimized state space and the candidate action vectors into the dual Critic network of the SAC reinforcement learning decision network to calculate the state-action value function and obtain the target Q-value evaluation result, includes: Based on the mean vector and variance vector in the action probability distribution, Gaussian distribution random sampling is performed using the reparameterization technique to generate candidate action vectors. The temperature tracking prediction, liquid cooling flow deviation, and pressure fluctuation data in the three-dimensional optimized state space are concatenated and combined with the candidate action vector to form a state-action input vector; The state-action input vector is input into the first Critic network of the dual Critic network in the SAC reinforcement learning decision network, and the first Q-value evaluation result is obtained by calculating the first Critic network. The state-action input vector is simultaneously input into the second Critic network of the dual Critic network in the SAC reinforcement learning decision network, and the second Q-value evaluation result is obtained by calculating the second Critic network. The minimum value selection operation is performed on the first Q-value evaluation result and the second Q-value evaluation result to obtain the target Q-value evaluation result.

[0012] Secondly, the present invention provides a liquid cooling heat dissipation and temperature control optimization system for aircraft cases, the liquid cooling heat dissipation and temperature control optimization system for aircraft cases comprising: The calculation module is used to collect instantaneous temperature data, temperature change rate data, and heat load data through the temperature sensor array on the surface of the flight case, and calculate the instantaneous temperature deviation value, temperature change rate deviation value, and heat load deviation value. The generation module is used to generate a comprehensive error evaluation vector based on the instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value; The prediction module is used to input the error comprehensive evaluation vector into the predictive temperature control tracking algorithm and generate temperature tracking prediction values ​​by combining the characteristics of the flight case movement conditions. The collaborative optimization module is used to perform collaborative optimization of liquid cooling parameters based on the temperature tracking prediction value, determine the optimal parameter control strategy, and synchronously adjust the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy.

[0013] A third aspect of the present invention provides a liquid cooling heat dissipation and temperature control optimization device for an aircraft case, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the liquid cooling heat dissipation and temperature control optimization device for the aircraft case to execute the above-described liquid cooling heat dissipation and temperature control optimization method for the aircraft case.

[0014] The technical solution provided by this invention establishes a three-level error identification model to achieve comprehensive and accurate perception of instantaneous temperature deviation, temperature change rate deviation, and heat load deviation, significantly improving the accuracy of temperature status identification compared to traditional single-point monitoring methods. Employing a dynamic weight allocation mechanism based on GPS positioning and ambient temperature perception, combined with an AI edge computing gateway, the importance of each error component can be intelligently adjusted according to actual operating conditions, avoiding the limitations of fixed weight methods. By using the comprehensive error evaluation vector as the temperature prediction correction parameter, combined with a moving operating condition learning algorithm, a shift from passive response to active prediction is achieved, particularly in preheating tracking and heat dissipation prediction compensation mechanisms under extreme temperature environments. Using a SAC reinforcement learning decision network and a dual-Critic architecture, intelligent collaborative optimization of liquid cooling pump speed, regulating valve opening, and cooling fan frequency is achieved, exhibiting stronger adaptability compared to traditional PID control. This invention, through a CAN bus distributed collaborative control architecture, ensures synchronous response of each sub-controller, improving the stability and reliability of the temperature control system.

[0015] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of one embodiment of the liquid cooling heat dissipation and temperature control optimization method for aircraft cases in this invention; Figure 2 This is a schematic diagram of one embodiment of the liquid cooling heat dissipation and temperature control optimization system for aircraft cases in this invention; Figure 3 This is a schematic diagram of one embodiment of the liquid cooling heat dissipation and temperature control optimization device for aircraft cases in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions 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, 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.

[0019] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0020] To facilitate understanding of this embodiment, a detailed description of the liquid cooling heat dissipation and temperature control optimization method for aircraft cases disclosed in this embodiment of the invention will be provided first. For example... Figure 1 As shown, this method includes the following steps: 101. Instantaneous temperature data, temperature change rate data, and heat load data are collected through a temperature sensor array on the surface of the flight case, and instantaneous temperature deviation value, temperature change rate deviation value, and heat load deviation value are calculated. It is understood that the executing entity of this invention can be an aviation case liquid cooling heat dissipation and temperature control optimization system, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0021] Specifically, a temperature sensor array is uniformly deployed on the surface of the flight case shell to collect instantaneous temperature data in different areas in real time. The temperature sensor array forms an 8×6 monitoring grid with a fixed 50 mm spacing, constructing a high-density temperature sensing network covering the entire case. This array can reflect the temperature distribution on the flight case surface in real time and the spatial thermal coupling relationship with the liquid cooling pipeline. To improve the accuracy of error modeling, the measured temperature values ​​of each sensor are weighted according to their spatial distance from the main liquid cooling channel. Sensors closer to the source are given higher weights. After weighted summation, a weighted average temperature value representing the overall thermal state is obtained. The difference between this value and the system's set target temperature value is then calculated to obtain the instantaneous temperature deviation value under the current system state, which is used to measure the degree of deviation of the overall heat dissipation performance from the target thermal control requirements. Based on this, in order to capture the dynamic temperature response of the flight case when the environment changes rapidly or the operating conditions change abruptly, a sliding window mechanism is introduced in the temperature data acquisition. This window updates the temperature change rate of each monitoring point in real time with a period of 0.5 seconds and a sampling frequency of 10Hz. The temperature change rate data is formed by calculating the differential slope between consecutive sampling points and taking the average. At the same time, the system has a reference temperature change rate model to describe the dynamic response behavior of the temperature control system under ideal operating conditions. By performing difference analysis between the real-time temperature change rate data and the reference model, the system can identify whether there is response hysteresis, overshoot or abnormal fluctuation in the temperature control system, thereby obtaining the temperature change rate deviation value. Considering the actual heat exchange behavior of the liquid cooling medium during operation, temperature sensors are installed at the inlet and outlet of the liquid cooling system to obtain the temperature difference data between the liquid inlet and outlet. Combined with an electromagnetic flow sensor to measure the liquid cooling flow rate in real time, and given the known density and specific heat capacity of the liquid cooling medium (e.g., ethylene glycol solution), the actual heat load value of the current system is calculated according to the thermodynamic calculation formula. This value is then compared with the theoretical heat dissipation load value calculated based on preset operating conditions. The numerical difference between the two is used as the quantitative assessment result of the heat load deviation. This deviation value is used to determine whether there are hidden problems such as insufficient heat exchange capacity, unstable flow rate, or decreased heat transfer efficiency in the liquid cooling system.

[0022] 102. Generate a comprehensive error evaluation vector based on instantaneous temperature deviation, temperature change rate deviation, and heat load deviation; Specifically, the instantaneous temperature deviation, temperature change rate deviation, and heat load deviation are standardized and formatted, then arranged in a predetermined order to form a three-dimensional initial error vector, corresponding to the discrete error descriptions of steady-state temperature control capability, dynamic response capability, and heat exchange capability, respectively. To enhance the error assessment mechanism's ability to perceive environmental changes, an integrated GPS positioning module acquires the current geographical location of the flight case in real time and matches it with the ambient temperature at that location. Simultaneously, it combines the operating speed and attitude information captured by sensors to identify the flight case's operating condition category, such as stationary, normal movement, or high-speed transport. Based on this, a set of corresponding dynamic weight coefficients are derived from the system's environment-condition mapping rule table. This weight combination reflects the importance ranking of the three error indicators under the current operating conditions; for example, instantaneous temperature control is prioritized under extremely cold conditions, while the heat load dimension is given priority under high heat load conditions. Subsequently, each dimension of the initial error vector is squared to highlight the weighted impact of outliers. The squared values ​​are then multiplied by the corresponding weight coefficients in the acquired dynamic weight coefficient combination, forming a new intermediate error weight vector. Each component of this vector reflects the coupling result between the deviation intensity and operating condition sensitivity of the current temperature control system in different dimensions. The intermediate error weight vector is normalized using the softmax function. Through exponential operations and denominator normalization constraints, the relative deviation weights of each dimension are kept between 0 and 1, with a total sum of 1, outputting a set of normalized weight values. Based on the normalization results, the AI ​​edge computing gateway re-weights and synthesizes the initial error vector, amplifying or suppressing each error value according to its normalized weight before vector summation and integration, thereby generating a comprehensive error evaluation vector.

[0023] 103. Input the error comprehensive evaluation vector into the predictive temperature control tracking algorithm, and generate temperature tracking prediction values ​​by combining the characteristics of the flight case movement conditions; Specifically, the error comprehensive evaluation vector is used as the temperature prediction correction parameter input to the predictive temperature control tracking algorithm. The instantaneous temperature deviation, temperature change rate deviation, and heat load deviation within this vector are used as adjustment factors for the temperature evolution trend. Simultaneously, combined with historically collected temperature change data, a basic model of temperature change over time is established using a least squares fitting algorithm. This model can describe the temporal distribution characteristics of the temperature response under current operating conditions. Through a mobile operating condition identification module, the current operating state of the flight case is classified based on GPS positioning and velocity vector information. An adaptive parameter combination for the current state is extracted using a mobile operating condition learning algorithm. These parameters are used to correct the fitting coefficients in the basic temperature trend model, enabling personalized adjustments to the dynamic temperature characteristics under different motion states. When the system identifies that the ambient temperature is between the preset first and second temperature values, i.e., the flight case is in the normal operating range, it inputs the current temperature value, the error comprehensive evaluation vector, and its time derivative (i.e., the rate of change of each deviation dimension) into the temperature change trend prediction function. The function structure contains multiple weighted terms and dynamic response terms, and outputs a temperature tracking prediction value through combined calculation. This prediction value can predict the development trend of the temperature control state within 2 to 5 seconds without relying on future actual measurement data. If the ambient temperature is lower than the first temperature value, such as in extreme environmental conditions below -40℃, the preheating tracking algorithm is activated, and the preheating power is dynamically adjusted by identifying the magnitude of the heat load deviation value in the error comprehensive evaluation vector. The system simultaneously activates the antifreeze mechanism during the adjustment and preheating process, injecting a 30% ethylene glycol solution as an antifreeze into the liquid cooling pipeline. A pulse heating method with a 2Hz frequency and 60% duty cycle is used to rapidly increase the temperature of the liquid cooling circuit, achieving rapid ice breaking and system reset, and outputting a temperature tracking prediction value under low-temperature conditions. When the ambient temperature exceeds a second temperature value, such as above 60℃, the system enters a high-temperature operating condition response mode. The system focuses on processing the instantaneous temperature deviation and heat load deviation values ​​in the error comprehensive evaluation vector, enhancing the prediction sensitivity by increasing the weight of the high-frequency response term in the model, and performing nonlinear correction by combining the heat transfer inertia parameters of the heat dissipation path, thereby generating a temperature tracking prediction value suitable for high-temperature operating conditions.

[0024] 104. Based on the temperature tracking prediction value, perform collaborative optimization of liquid cooling heat dissipation parameters, determine the optimal parameter control strategy, and synchronously adjust the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy.

[0025] Specifically, the predicted temperature tracking value is used as the main control input signal, and together with the liquid cooling flow deviation data collected in real time from the electromagnetic flowmeter and the system pressure fluctuation data measured by the pressure sensor, a three-dimensional optimization state space is formed. This state space reflects the current temperature control target trend of the system and integrates the dynamic characteristics of the internal medium transport and mechanical response of the liquid cooling system, thus fully expressing the multi-source coupling factors affecting heat dissipation efficiency. On this basis, in order to achieve the linkage optimization among the liquid cooling pump speed, electric regulating valve opening, and cooling fan frequency, a multi-objective optimization reward function is introduced. This function comprehensively considers multiple performance indicators such as temperature deviation suppression capability, energy consumption efficiency, response speed, and temperature control stability. By assigning differentiated weights to the instantaneous deviation term, power consumption normalization term, temperature overshoot term, and settling time deviation term, and combining them with the error normalization coefficient for exponential decay calculation, a comprehensive performance evaluation value with monotonicity and continuity is obtained. This evaluation value can quantify the overall performance of the current control strategy under multi-dimensional objectives. The system uses this comprehensive performance evaluation value as a reward signal to input into a reinforcement learning decision network built on the SAC algorithm. This network adopts a three-layer structure: the input layer maps 9-dimensional state space data, the hidden layer extracts nonlinear features through 64 neurons and maintains the continuity and stability of the policy, and the output layer generates control combinations corresponding to the liquid cooling pump speed increment, valve opening adjustment range, and fan frequency change commands. During offline pre-training and online fine-tuning, this decision network has established an optimal policy mapping with typical operating condition samples, thus enabling it to quickly output an adaptive optimal parameter control policy within 5 milliseconds during runtime. The main controller synchronously sends the control strategy to three sub-modules—the liquid cooling pump controller, the regulating valve controller, and the fan controller—via the CAN bus. The liquid cooling pump controller receives and parses the speed adjustment command and then drives the actuator motor to respond via a frequency converter. The adjustment range is between 500 and 3000 rpm with an accuracy of 10 rpm. The regulating valve controller uses a stepper motor to achieve adjustment within the range of 0 to 100%, with a minimum adjustment accuracy of 0.5%. The fan controller uses PWM technology to regulate the fan speed in the frequency range of 20 Hz to 100 Hz with a resolution of 1 Hz. The three sub-modules maintain system delays of 100 milliseconds, 200 milliseconds, and 50 milliseconds respectively during the command response process. Through synchronous adjustment between the controllers, the liquid cooling system can quickly converge to the target temperature trajectory under different load environments, thereby achieving a high-precision, low-energy-consumption, and oscillation-free temperature control process.

[0026] In one specific embodiment, the process of performing step 101 may specifically include the following steps: By arranging a temperature sensor array on the surface of the flight case to collect instantaneous temperature data at each monitoring point in real time, the weighting coefficient is determined according to the positional relationship between each sensor and the liquid cooling pipeline, and the weighted average temperature value is calculated. The difference between the weighted average temperature value and the target temperature value is calculated to obtain the instantaneous temperature deviation value. Collect temperature change data and calculate the temperature change rate data through a preset sliding window. Calculate the difference between the temperature change rate data and the reference temperature change rate to obtain the temperature change rate deviation value. Temperature difference data and flow data of liquid cooling medium inlet and outlet are collected by temperature sensors and flow sensors, respectively. Heat load data are calculated based on the density and specific heat capacity of liquid cooling medium. The difference between heat load data and theoretical heat dissipation load value is calculated to obtain heat load deviation value.

[0027] Specifically, an 8×6 array of temperature sensors is arranged on the surface of the flight case shell at 50mm intervals. This array covers most of the flight case's heat dissipation surface. By evenly distributing the sensors in key structural areas and placing them close to the liquid cooling channels, the system can detect changes in heat distribution on the case surface in real time and capture the thermal coupling state with the liquid cooling pipes. During data acquisition, each temperature sensor samples the instantaneous temperature at its location at a frequency of 10Hz. The system internally assigns weighting coefficients based on the spatial distance between each sensor and the main liquid cooling circulation pipes; sensors closer to the liquid cooling pipes have higher weights, thus constructing a weighted temperature model for the heat conduction path. By summing the instantaneous temperature data from all sensors with their respective weighting coefficients, a weighted average temperature value representing the overall thermal response of the flight case is obtained. Subsequently, the difference between the weighted average temperature value and the system's set temperature control target value is calculated to obtain the instantaneous temperature deviation value at the current moment. Simultaneously, to capture the dynamic changes in the flight case's thermal state over a short period, the system introduces a time-oriented data analysis mechanism, i.e., constructing a sliding time window model based on continuous temperature measurement data. Each temperature sensor continuously records sampling data for 0.5 seconds, totaling 5 data points. The rate of temperature change over time is calculated using the first-order difference method to obtain the temperature change rate data for each monitoring point. The change rate data from all sensors are then averaged to obtain the overall temperature change trend. This trend data represents the response rate of the aircraft case surface temperature and is related to the degree of matching with actual environmental disturbances, heat load changes, and control actions. Simultaneously, the system embeds a reference temperature change rate model. This model is derived from long-term statistical fitting results of the temperature change rate under ideal temperature control conditions in an experimental environment, possessing typical response curves under different operating conditions, thus serving as the reference surface for the dynamic response of temperature control. When there is a significant deviation between the temperature change rate data collected by the system and this reference model, it indicates problems such as hysteresis, abnormal heating, or cooling in the current temperature control response. Therefore, the system performs a difference calculation between the collected temperature change rate data and the reference value to obtain the temperature change rate deviation value. To evaluate the operating status of the liquid cooling system from the perspective of energy conservation in the heat exchange process, high-precision temperature sensors are installed at the inlet and outlet of the liquid cooling pipeline to obtain the temperature difference of the liquid cooling medium in real time during the process of entering and leaving the aviation box. At the same time, an electromagnetic flow sensor is installed in the liquid cooling circuit to collect the current liquid flow rate. These two physical quantities together determine the amount of heat carried away by the liquid cooling medium per unit time.In this system, the liquid cooling medium is an aqueous solution of ethylene glycol, whose density and specific heat capacity are known constants. The temperature difference and flow rate are converted into actual heat load values ​​using a heat power conversion formula, thus calculating the current actual heat dissipation load of the flight case. Simultaneously, the system establishes a theoretical heat load model based on the current equipment power consumption level, heat dissipation structure characteristics, and boundary conditions. This model provides the heat dissipation capacity level that should be achieved under ideal conditions. The difference between the two models represents the loss, imbalance, or nonlinear deviation of heat power in the heat transfer path. By calculating the difference between the actual heat dissipation load and the theoretical heat load model and quantifying this value, the heat load deviation value is obtained.

[0028] In one specific embodiment, the process of performing step 102 may specifically include the following steps: The instantaneous temperature deviation, temperature change rate deviation, and heat load deviation are vectorized to obtain the initial error vector. The current ambient temperature and working status data of the flight case are obtained through the GPS positioning module, and the dynamic weighting coefficient combination is determined based on the current ambient temperature and working status data. The initial error vector is squared and multiplied with the corresponding dynamic weight coefficient in the dynamic weight coefficient combination to obtain multiple calculation results; Multiple calculation results are input into the softmax function for dynamic weight allocation to obtain normalized weight allocation results. The normalized weight allocation results are then vectorized through the AI ​​edge computing gateway to obtain the comprehensive error evaluation vector.

[0029] Specifically, the instantaneous temperature deviation, temperature change rate deviation, and heat load deviation are structured and organized in a fixed order to form an initial error vector with three-dimensional spatial characteristics. The system's built-in GPS positioning module initiates the environmental status identification process. By acquiring the real-time geographical location information of the flight case and combining it with a geographic environment database, it automatically analyzes the ambient temperature at the current location. Then, in conjunction with a triaxial accelerometer and speed sensor, it determines the current operating state of the flight case, including different operating modes such as stationary, low-speed movement, high-speed transportation, impact conditions, and variable load processes. The ambient temperature and operating state are input into the system's preset weight adjustment rule matrix. This matrix has been subdivided into several sets of dynamic weight coefficient combinations according to conditional logic. For example, when the ambient temperature is below -40℃ and the flight case is in low-speed movement... In dynamic conditions, the system prioritizes instantaneous temperature control, correspondingly increasing the weighting coefficient of instantaneous temperature deviation to 0.6, while compressing the weighting coefficient of heat load deviation to 0.1. If in a neutral temperature range and the system operates in steady-state transport mode, a relatively balanced weighting configuration is adopted to ensure all three deviation indicators are evaluated equivalently. Under high temperature and high load conditions, the weighting of heat load deviation is increased to 0.4 to highlight the risk of insufficient heat exchange capacity. Therefore, through this weight matrix query process, the system dynamically selects the most suitable weighting configuration under different operating scenarios, generating a set of dynamic weighting coefficient combinations in the form of a three-dimensional vector. Each component in the initial error vector is squared to enhance the weight of outliers in the deviation values, and the influence of large outliers on the weighted result is strengthened through nonlinear gain. The squared three-dimensional error vector is then multiplied by each weight value in the dynamic weighting coefficient combination to form a weighted deviation intermediate vector, where each component represents a composite result of the significance of a certain type of deviation under the current operating condition and the system's level of concern. To improve the comparability, consistency, and balanced distribution among the three deviation indicators, the system introduces the softmax function to normalize the aforementioned weighted deviation intermediate vector. This function, acting as a multinomial exponential normalization mapping tool, transforms any set of positive numbers into a probability distribution with a sum of 1, while preserving the relative differences between the original values. The normalized weight vector obtained after processing by this function represents the final dynamic allocation result of the three types of deviation indicators under the current environment. The system synthesizes the normalized weight vector and the original initial error vector through component-wise multiplication. This vector synthesis operation is executed through an AI edge computing gateway, leveraging the high-speed floating-point matrix calculation capabilities of the edge processor to complete the product superposition calculation between the three-dimensional vectors, generating a comprehensive error evaluation vector.

[0030] In one specific embodiment, the process of performing step 103 may specifically include the following steps: The error comprehensive evaluation vector is used as the temperature prediction correction parameter input into the predictive temperature control tracking algorithm. The algorithm is then fitted with historical temperature data using the least squares method to obtain the basic data of temperature change trend. Based on temperature change trend data, adaptive parameter combinations for mobile operating conditions are calculated using a mobile operating condition learning algorithm. When the ambient temperature is between the first and second temperature values, the current temperature value, the error comprehensive evaluation vector and its rate of change are calculated according to the temperature change trend prediction function to obtain the temperature tracking prediction value under normal operating conditions. When the ambient temperature is lower than the first temperature value, the preheating tracking algorithm is activated and the preheating power is adjusted based on the heat load deviation value in the error comprehensive evaluation vector. At the same time, ethylene glycol antifreeze is injected into the liquid cooling pipeline, and pulse heating is used for rapid ice breaking to obtain the temperature tracking prediction value for low-temperature start-up. When the ambient temperature is higher than the second temperature value, the temperature tracking prediction value for high-temperature conditions is calculated based on the instantaneous temperature deviation value and heat load deviation value in the error comprehensive evaluation vector.

[0031] Specifically, the error comprehensive evaluation vector is input as a temperature prediction correction parameter into the predictive temperature control tracking algorithm. This vector consists of three indicators: instantaneous temperature deviation, temperature change rate deviation, and heat load deviation, which can characterize the coupled deviation behavior of the system's thermal balance, dynamic response, and heat dissipation capacity. Upon receiving this vector, the system simultaneously accesses its built-in historical temperature record database to obtain temperature sequence data for each time point over the past few seconds. Combining this with the corresponding environmental conditions, load changes, and system response information, a basic model of temperature change trend is constructed using a least squares fitting algorithm. This fitting process uses time as the independent variable and historical temperature values ​​as the dependent variable, calculating a set of fitting coefficients to ensure that the predicted curve closely matches the historical data as closely as possible within the mean square error framework, thus constructing the trajectory of the system's thermal state in the time domain. Based on the trend model obtained from the above fitting, the system introduces a moving condition learning algorithm to extract parameter combinations that match the current motion state. This learning algorithm uses GPS sensor-collected data on the flight case's position and velocity gradient as core inputs. By calculating acceleration curves, displacement frequencies, and velocity distribution images, it identifies the typical motion conditions the flight case is currently in, such as uniform transport, intermittent movement, high-speed vibration, or static standby. It then matches historical condition templates with existing training samples and combines this with the temperature inertia and response time lag characteristics exhibited by the system under these conditions to select the best-matching set of adaptive parameters for the current trend model. These parameters include response delay correction terms, prediction window dynamic adjustment coefficients, and nonlinear fitting compensation factors, which are used to dynamically correct the coefficient weights in the trend prediction function to better fit the current condition characteristics and avoid temperature control judgment deviations under different speeds or vibration environments. When the system determines that the current ambient temperature is between the set first and second temperature values, i.e., within the normal operating range, the system inputs the current measured temperature value, the error comprehensive evaluation vector, and its time change rate as variables into the corrected temperature change trend prediction function. This function structure includes a linear time term, a second derivative term, and a working condition coupling weight term. It describes the future temperature change trend through a polynomial fitting function or an exponential recursive function, and uses the volatility of the error vector as an excitation factor to adjust the prediction amplitude, thus obtaining the temperature tracking prediction value under normal operating conditions. When the system detects that the current ambient temperature is lower than the first temperature value (e.g., -40℃), it enters the low-temperature extreme condition processing flow. At this time, due to the risk of solidification, increased viscosity, or pipe icing of the liquid cooling medium, the system first activates the preheating tracking algorithm. This algorithm judges the degree of power conduction lag in the current heat transfer path based on the magnitude of the heat load deviation value in the error comprehensive evaluation vector. If the deviation value is significant, it indicates that the system heat flow is blocked or the cold end cannot effectively absorb heat. The system compensates for this thermal resistance by increasing the preheating power and adjusts the preheating power output proportionally based on the deviation value, thereby achieving customized heating based on actual heat demand.Simultaneously, the system injects a 30% mass fraction ethylene glycol solution, pre-stored in the cryogenic stabilization chamber, into the liquid cooling pipeline. This antifreeze effectively lowers the freezing point of the coolant and improves the initial heat exchange efficiency by adjusting its heat capacity ratio. After the liquid is added, the system activates the pulse heating module, performing rapid heating at a frequency of 2Hz and a duty cycle of 60%. This pulsed heating stimulates thermal convection within the liquid cooling circuit, achieving a rapid transition from a frozen to a thawed state. The entire process is completed within 3 seconds, and a predicted temperature value under cryogenic startup conditions is output. If the system detects that the ambient temperature exceeds a second temperature value (e.g., 60°C), the system enters a high-temperature protection mode, at which point the thermal stress risk to the flight case increases. Under high-temperature conditions, the system prioritizes extracting the instantaneous temperature deviation and heat load deviation from the comprehensive error evaluation vector, as these directly reflect whether the high temperature is caused by control failure or insufficient heat exchange capacity. These values ​​are then input into a modified high-temperature prediction function. This function emphasizes exponential growth and a heat dissipation upper limit term to predict whether there is an overshoot risk in the short term, and uses higher-order derivative analysis to determine whether a secondary temperature rise will occur in the future. Finally, the system integrates the predicted temperature curve and the fluctuation trends of each component to output a temperature tracking prediction value suitable for extreme high-temperature conditions.

[0032] In one specific embodiment, the process of executing the step of starting the preheating tracking algorithm and adjusting the preheating power based on the heat load deviation value in the error comprehensive evaluation vector when the ambient temperature is lower than the first temperature value, while injecting ethylene glycol antifreeze into the liquid cooling pipeline and using pulse heating for rapid ice breaking, to obtain the temperature tracking prediction value for low-temperature start-up can specifically include the following steps: When the ambient temperature is lower than the first temperature value, the preheating tracking algorithm is started and the ambient temperature and the preheating start time are used as initial parameters. The preheating temperature is calculated based on the preheating tracking function to obtain the basic value of the preheating temperature. The preheating power is dynamically adjusted based on the heat load deviation value in the error comprehensive evaluation vector to obtain the dynamically adjusted preheating power value. Based on the dynamically adjusted preheating power value, ethylene glycol antifreeze is injected into the liquid cooling pipeline through a solenoid valve, while the internal temperature change of the liquid cooling pipeline is monitored to obtain pipeline temperature data after antifreeze injection. Based on the pipeline temperature data after antifreeze injection, a pulse heating method is used for rapid ice breaking. The ice breaking heating process is executed by a pulse heating controller according to the dynamically adjusted preheating power value, and the temperature change data after ice breaking heating is obtained. By fusing the preheating temperature baseline value with the temperature change data after ice breaking and heating, and combining it with the preheating tracking function to correct the temperature prediction, the temperature tracking prediction value for low-temperature start-up is obtained.

[0033] Specifically, the measured ambient temperature and the recorded preheating start time are used as initial input parameters to the preheating tracking function built into the system. This function uses a time-power exponential increase model to model thermal behavior and estimate the temperature growth trend over time under ideal conditions during preheating. The coefficients k and n are empirical parameters obtained by fitting the actual system heat capacity characteristics and the preheater power curve during the pre-training stage. The preheating temperature baseline value output by this function represents the theoretical temperature rise baseline that can be achieved under ideal heating conditions without thermal interference or thermal hysteresis. The system retrieves the heat load deviation component from the comprehensive error evaluation vector. This deviation value is determined by the difference between the actual heat dissipation load calculated from the liquid cooling inlet / outlet heat difference and flow rate, and the theoretical heat dissipation model. If this heat load deviation value is large, it indicates that the current liquid cooling system's heat exchange capacity is severely insufficient, leading to a decrease in heat exchange efficiency due to freezing, air resistance, or attenuation of heat diffusion capacity. Therefore, the system inputs this deviation value to the dynamic power adjustment module. This module converts the deviation value into a preheating power correction factor based on a proportional-nonlinear mapping relationship. This factor further affects the PWM duty cycle and modulation frequency control parameters of the heating power supply in the preheating control module, forming a dynamically adjusted preheating power value. Based on the dynamically adjusted preheating power value, the system injects ethylene glycol antifreeze for low-temperature operation into the liquid cooling pipeline through a solenoid valve drive module. This injection action is controlled by real-time closed-loop feedback control logic to ensure that the injection rate matches the current system's heat exchange capacity and preheating rate, and the ethylene glycol concentration is maintained at a mass fraction of 30%. During the injection process, the system continuously monitors the temperature changes of key nodes in the liquid cooling pipeline in real time through a temperature sensor array distributed within the liquid cooling circuit. This generates a set of temperature response curve data after antifreeze injection, reflecting the thermal diffusion rate of the antifreeze and the heat absorption efficiency of the cold-end structure to the liquid cooling medium. The system activates the pulse heating module, which uses a 2Hz square wave pulse signal control method with a 60% duty cycle to regulate the on / off state of the heater. Pulse heating avoids local overheating of the structure caused by continuous power output, while improving heat transfer efficiency and system startup speed. During pulse heating, the system automatically calculates the effective power threshold and pulse duration for each pulse cycle based on the aforementioned dynamically adjusted preheating power value. The pulse heating controller outputs a specific current modulation signal to the heating unit, forming a rapid local heating zone at the interface between the heater surface and the liquid cooling medium. This zone can form a local thawing expansion zone in a very short time, gradually thawing the entire main liquid cooling circuit with each pulse. The system continuously records the temperature change data after ice breaking heating throughout the entire pulse heating process, and dynamically compares and analyzes this data with the pipeline temperature data after antifreeze injection to determine whether the heating process has reached the system's set start-up temperature threshold.The base value of preheating temperature initially output by the preheating tracking function is fused with the temperature change data after the ice-breaking heating is completed. A fused temperature curve is constructed using methods such as weighted averaging, time extrapolation, and trend fitting. This curve is then re-inputted into the corrected preheating tracking function. Combining the latest dynamic power factor, pipeline heat capacity change trend, and medium state change indicators, the prediction function coefficients are adjusted using augmented function modeling to finally output the temperature tracking prediction value under low-temperature start-up conditions.

[0034] In one specific embodiment, the process of performing step 104 may specifically include the following steps: Based on the temperature tracking prediction, combined with the liquid cooling flow deviation measured by the electromagnetic flowmeter and the pressure fluctuation data measured by the pressure sensor, a three-dimensional optimization state space for the coordinated optimization of liquid cooling heat dissipation parameters is constructed. The performance of the three-dimensional optimization state space is evaluated by optimizing the multi-objective reward function to obtain a comprehensive performance evaluation value. The comprehensive performance evaluation value is used as a reward signal and input into the SAC reinforcement learning decision network for processing, and the optimal parameter control strategy is output. The optimal parameter control strategy is synchronously sent to each sub-controller via the CAN bus. The liquid cooling pump controller adjusts the speed of the liquid cooling pump in the aircraft case, the regulating valve controller adjusts the valve opening, and the fan controller adjusts the frequency of the cooling fan.

[0035] Specifically, the predicted temperature value is used as the primary reference, representing the expected change path of the aircraft case's thermal state within the next few seconds. Simultaneously, the system obtains the actual flow rate of the liquid cooling medium through electromagnetic flowmeters deployed in the main liquid cooling circulation pipeline. This actual flow rate is then compared with the ideal flow rate to calculate the liquid cooling flow rate deviation. This parameter reflects the degree of deviation between the current actual cooling capacity and the ideal flow rate required for heat exchange. Simultaneously, the system calls upon the feedback value from the pressure sensor and performs high-speed sampling and filtering to obtain the pressure fluctuation data of the circulation path in the current liquid cooling system. This parameter reflects the fluid stability fluctuations caused by local turbulence, air resistance disturbances, valve disturbances, or abnormal speed regulation in the liquid cooling system. The system integrates the three core parameters—predicted temperature value, liquid cooling flow rate deviation, and system pressure fluctuation—to form a three-dimensional optimization state space for the coordinated optimization of liquid cooling heat dissipation parameters. This state space reflects the coordinated changes between temperature control objectives, execution paths, and system disturbances. Based on this state space, the system introduces a multi-objective optimization performance evaluation mechanism and quantifies the current control state by defining a reward function containing multiple performance objectives. This function comprehensively considers four aspects: temperature control accuracy, energy efficiency, response speed, and system stability. It measures control accuracy by exponentially decaying the deviation between the predicted and target temperatures, energy consumption burden by squaring the ratio of current power to rated power, and overshoot risk by squaring the temperature overshoot amplitude. It also introduces a term for the difference between the actual settling time and the expected response time to measure dynamic response quality. Weight coefficients representing control priorities are assigned to each of these four performance objectives. All results are then uniformly weighted and combined to obtain the comprehensive performance evaluation value of the system operating in the current state space. A higher value indicates that the current temperature control behavior better conforms to the optimization objectives of multi-objective control. The system uses the comprehensive performance evaluation value as a reward signal input to a reinforcement learning policy network constructed based on the maximum entropy strategy optimization idea. A three-layer neural network model is used. This network receives parameters from the three-dimensional optimization state space as input, performs feature combination and nonlinear mapping in the hidden layer, and generates three corresponding control variables in the output layer, used to adjust the speed of the liquid cooling pump, the opening of the regulating valve, and the frequency of the cooling fan, respectively. Before deployment, this control strategy network underwent offline training using environmental simulation data covering various typical operating conditions, from extremely low temperatures to high temperatures and from static to high-speed motion. During operation, it is equipped with an online fine-tuning mechanism. When drastic changes occur in the actual operating environment of the flight case, such as a sudden increase in speed or a rapid rise in heat load, the system can quickly update some parameters of the control strategy network to adapt to the new conditions. Once the optimal control strategy is generated, the system converts it into a standard control instruction set through the bus scheduling module and sends it at a fixed rate through the control area network to multiple sub-controllers in the distributed control architecture, including the liquid cooling pump controller, regulating valve controller, and fan controller.The liquid cooling pump controller, upon receiving a speed adjustment command, adjusts the pump's operating frequency via frequency conversion, covering a range of 500 to 3000 revolutions per minute with a minimum adjustment accuracy of 10 revolutions per minute. The regulating valve controller receives opening commands and drives a stepper motor to perform minute angle adjustments, covering a range of 0% to 100% with a minimum resolution of 0.5%. The fan controller, based on frequency adjustment commands, uses pulse width modulation to control the motor's drive frequency, with a frequency adjustment range of 20 to 100 Hz and a control step size of 1 Hz. The maximum delays of each sub-controller in responding to control commands are controlled to within 100 milliseconds, 200 milliseconds, and 50 milliseconds, respectively.

[0036] In one specific embodiment, the process of inputting the comprehensive performance evaluation value as a reward signal into the SAC reinforcement learning decision network for processing and outputting the optimal parameter control strategy can specifically include the following steps: The Actor network of the SAC reinforcement learning decision network uses the three-dimensional optimized state space as the state input and the comprehensive performance evaluation value as the reward signal input. The forward propagation calculation is performed through the three-layer neural network structure of the Actor network to generate the action probability distribution. Candidate action vectors are generated based on the action probability distribution through a Gaussian distribution sampling mechanism. At the same time, the three-dimensional optimized state space and candidate action vectors are input into the dual Critic network of the SAC reinforcement learning decision network to calculate the state-action value function and obtain the target Q-value evaluation result. Based on the target Q-value evaluation results and the comprehensive performance evaluation results, the strategy loss function and the value loss function are calculated. The parameters of the Actor network and the dual Critic network are updated respectively using the gradient descent algorithm to obtain the optimized network parameters. Based on the optimized network parameters, the Actor network outputs the original control actions, including the liquid cooling pump speed adjustment, the electric regulating valve opening increment, and the cooling fan frequency change. The original control actions are then transformed by action space mapping to obtain the optimal parameter control strategy.

[0037] Specifically, the system injects the 3D optimized state space input as a state vector into the policy generation sub-network, i.e., the Actor network, within a reinforcement learning framework based on a maximum entropy optimization strategy. This state space integrates temperature tracking predictions, liquid cooling flow deviations, and pressure fluctuation data within the liquid cooling loop. Simultaneously, the system injects the comprehensive performance evaluation value as a reward signal for the current control behavior into the policy evaluation sub-network, i.e., the Critic network, forming the learning objective of the state-action value function. The entire training structure employs the Soft Actor-Critic algorithm as the basis for policy updates, ensuring convergence while enhancing the exploratory nature of the policy. Within the Actor network, the system constructs a three-layer neural network structure consisting of an input layer, two hidden layers, and an output layer. The input layer receives and standardizes the numerical input from the 3D optimized state space. The two hidden layers each have 64 neurons and use nonlinear activation functions to perform depth transformation and combination of the input features, generating probabilistic responses to action selection in the current state. The output layer generates the distribution parameters of continuous control actions, including the mean and standard deviation of each control variable, thus defining a complete action probability distribution. This distribution reflects the probability and policy tendency of the system to execute various control actions (such as different pump speeds, valve openings, or fan frequencies) in the current state. After obtaining the action probability distribution, the system uses a Gaussian distribution sampling mechanism to perform a random sampling within this distribution, generating a candidate action vector for the current state. This action vector contains provisional values ​​for three control variables: the adjustment range of the liquid cooling pump speed, the increment of the electric regulating valve opening, and the frequency change of the cooling fan. The system inputs this candidate action vector along with the original three-dimensional optimized state space into a dual-Critic network for the state-action value function evaluation process. The dual-Critic network consists of two independent neural networks, similar in structure to the Actor network, except that its output layer has only one neuron, responsible for outputting the value estimate corresponding to the state and action combination, i.e., the target Q value. During the evaluation process, the two Critic networks independently calculate the same input and output two value estimates. When updating the policy, the system uses the smaller of the two estimates as the upper limit of the current policy value, thereby avoiding the policy from over-relying on overvalued behaviors during the update process. Based on the difference between the actions generated by the Actor network and the target Q-value calculated by the Critic network, the system calculates the policy loss function and the value loss function. The policy loss function measures the deviation between the current action distribution generated by the Actor network and the maximization of the expected value. It adopts a maximum entropy objective function structure that includes an entropy adjustment factor, ensuring that the policy has both high expected rewards and maintains sufficient policy randomness to preserve the system's exploration capability. The value loss function calculates the error between the current output value of the Critic network and the true target Q-value using the mean squared error method, thereby evaluating the estimation accuracy of the value network.The system calculates gradients for both types of loss functions and updates all learnable parameters, including weight matrices and bias vectors, in the Actor and dual Critic networks using gradient descent. The update process iterates incrementally with a small learning rate to ensure stable convergence of network performance during training. After joint optimization of the Actor and Critic network parameters, the system performs forward propagation on the current optimized state space based on the updated Actor network, directly outputting a new set of control action vectors. The three variables in this vector correspond to the speed adjustment of the liquid cooling pump, the opening adjustment increment of the electric regulating valve, and the frequency change amplitude of the cooling fan, respectively. The system uses this vector as the original control action signal. Since the control variables output by the reinforcement learning network are within a standardized continuous space, to ensure that each control variable conforms to the physical boundaries and operating range of the actual actuators, the system introduces an action space mapping and transformation module after the original control action output. This module performs nonlinear range scaling, boundary truncation, and discrete precision adjustment on the actions. For example, the liquid cooling pump speed regulation is mapped from a standard range to 500 to 3000 revolutions per minute, the valve opening increment is mapped to a linear range of 0% to 100%, with a minimum adjustment unit of 0.5%, and the fan frequency variation range is mapped to 20 to 100 Hz, with a minimum resolution of 1 Hz. The mapped control action set is encapsulated into a standard control strategy and synchronously distributed by the main controller to the liquid cooling pump controller, valve controller, and fan controller via the control area network. This achieves a coordinated response at the actuator level, thereby deploying the optimal control strategy driven by reinforcement learning based on the current state, completing a full intelligent closed-loop control cycle of state perception, strategy evaluation, network update, and control execution.

[0038] In one specific embodiment, the process of generating candidate action vectors based on the action probability distribution using a Gaussian distribution sampling mechanism, and simultaneously inputting the three-dimensional optimized state space and candidate action vectors into the dual Critic network of the SAC reinforcement learning decision network to calculate the state-action value function and obtain the target Q-value evaluation result can specifically include the following steps: Based on the mean and variance vectors in the action probability distribution, Gaussian distribution random sampling is performed using reparameterization techniques to generate candidate action vectors. The temperature tracking prediction, liquid cooling flow deviation, and pressure fluctuation data in the three-dimensional optimized state space are concatenated and combined with the candidate action vector to form a state-action input vector; The state-action input vector is input into the first Critic network of the dual Critic network in the SAC reinforcement learning decision network, and the first Q-value evaluation result is obtained by calculating the first Critic network. The state-action input vector is simultaneously input into the second Critic network of the dual Critic network in the SAC reinforcement learning decision network, and the second Q-value evaluation result is obtained by calculating the second Critic network. The minimum value selection operation is performed on the first Q-value evaluation result and the second Q-value evaluation result to obtain the target Q-value evaluation result.

[0039] Specifically, the policy generation sub-network (Actor network) outputs a set of parameters representing the current control action tendency. This set includes the mean vector and variance vector of the action probability distribution. The mean vector represents the center position of the optimal action as perceived by the policy network in the current state, while the variance vector reflects the policy uncertainty and exploration range of action selection. To perform effective sampling while maintaining policy continuity and differentiability, the system uses a reparameterization technique to randomly sample the action probability distribution, converting the original action distribution expression from a random variable form to a differentiable expression. The noise vector generated in the standard normal distribution is multiplied by the square root of the variance vector and then added to the mean vector to generate the actual candidate action vector. The system extracts the current temperature tracking prediction value, liquid cooling flow deviation, and pressure fluctuation data from the three-dimensional optimization state space and concatenates these three state variables with the candidate action vector to form a state-action input vector. The system inputs this state-action input vector into the first Critic network in the reinforcement learning structure. This network is the first state-action value evaluator in the reinforcement learning framework, performing forward propagation calculations on the input combination and outputting the first set of Q-value evaluation results. Within the first Critic network, the state-action input vector undergoes linear transformation and normalization in the input layer and is then passed to two nonlinear computation layers in the hidden layer. These layers, using neurons with activation functions, extract the nonlinear relationship, mutual influence, and combinatorial sensitivity between the state and action, generating a scalar Q-value in the output layer. This Q-value represents the cumulative reward the system expects to obtain after executing a candidate action in the current state. A larger Q-value indicates that the action executed in the current state is more likely to yield high-performance feedback, including higher temperature control accuracy, lower energy consumption, faster response, and greater system stability. The system synchronously inputs the same state-action input vector into the second Critic network. This network has the same structure as the first Critic network but with independent weight parameters and neural network paths. This provides an independent value evaluation path to avoid overfitting or overestimation risks associated with a single evaluation network. In the second Critic network, the input vector is also processed through a combination of input layers, nonlinear hidden layers, and output layers, ultimately outputting a second Q-value evaluation result. After completing the dual-network evaluation, to enhance the stability of policy training and suppress the problem of value function overestimation, a minimum value selection mechanism is introduced. That is, the first Q-value evaluation result of the output of the first Critic network is compared with the second Q-value evaluation result of the output of the second Critic network, and the smaller value is selected as the target Q-value evaluation result of the state-action combination. This processing method is consistent with the minimum value filtering idea in the Soft Actor-Critic algorithm, which conservatively estimates the value of the policy in the output of multiple networks to prevent overly optimistic evaluation from causing the policy network to engage in reckless behavior.

[0040] The above describes the liquid cooling and temperature control optimization method for aircraft cases in embodiments of the present invention. The following describes the liquid cooling and temperature control optimization system for aircraft cases in embodiments of the present invention. Please refer to [link / reference]. Figure 2 One embodiment of the liquid cooling heat dissipation and temperature control optimization system for aircraft cases in this invention includes: The calculation module 201 is used to collect instantaneous temperature data, temperature change rate data and heat load data through the temperature sensor array on the surface of the flight case, and calculate the instantaneous temperature deviation value, temperature change rate deviation value and heat load deviation value. The generation module 202 is used to generate a comprehensive error evaluation vector based on the instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value; Prediction module 203 is used to input the error comprehensive evaluation vector into the predictive temperature control tracking algorithm and generate temperature tracking prediction values ​​by combining the characteristics of the flight case movement conditions. The collaborative optimization module 204 is used to perform collaborative optimization of liquid cooling parameters based on temperature tracking prediction values, determine the optimal parameter control strategy, and synchronously adjust the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy.

[0041] Through the collaborative efforts of the aforementioned components, a three-level error identification model is established to simultaneously monitor instantaneous temperature deviation, temperature change rate deviation, and heat load deviation. Compared to traditional single-point monitoring methods, this achieves comprehensive and accurate perception of the temperature status of the liquid cooling system in the aircraft case. Employing a dynamic weighting coefficient adjustment mechanism based on GPS positioning and ambient temperature perception, combined with a softmax function and an AI edge computing gateway, the importance of each error component can be intelligently adjusted according to actual operating conditions, avoiding the limitations of traditional fixed-weight methods and improving the accuracy and adaptability of comprehensive error assessment. By using the comprehensive error assessment vector as a temperature prediction correction parameter, combined with a moving operating condition learning algorithm and a multi-operating condition prediction mechanism, a shift from passive response to active prediction is achieved. Particularly in the preheating tracking algorithm under low-temperature conditions and the heat dissipation prediction compensation under high-temperature conditions, the temperature control response speed and tracking accuracy are significantly improved. Using a SAC reinforcement learning decision network and a dual-Critic network architecture, through the Actor-Critic structure and reparameterization techniques, intelligent collaborative optimization of liquid cooling pump speed, regulating valve opening, and cooling fan frequency is achieved, exhibiting stronger adaptability and superior control performance compared to traditional PID control. A three-dimensional optimization state space was established based on temperature tracking predictions, liquid cooling flow deviations, and pressure fluctuation data. A multi-objective optimization reward function was used for comprehensive performance evaluation, achieving coordinated optimization of multiple control parameters and avoiding mutual interference issues when adjusting parameters individually. A CAN bus and distributed control architecture were adopted to achieve efficient communication and coordinated control between the main controller and each sub-controller, ensuring synchronous response of the liquid cooling pump controller, regulating valve controller, and fan controller, thus improving the stability and reliability of the entire temperature control system.

[0042] above Figure 2 The liquid cooling and temperature control optimization system for aircraft cases in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The liquid cooling and temperature control optimization device for aircraft cases in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0043] Figure 3This is a schematic diagram of the structure of a liquid cooling and temperature control optimization device for an aircraft case according to an embodiment of the present invention. The liquid cooling and temperature control optimization device 300 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the liquid cooling and temperature control optimization device 300. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the liquid cooling and temperature control optimization device 300 to implement the steps of the aforementioned liquid cooling and temperature control optimization method for an aircraft case.

[0044] The liquid cooling and temperature control optimization device 300 for flight cases may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated structure of the liquid cooling heat dissipation and temperature control optimization device for aircraft cases does not constitute a limitation on the liquid cooling heat dissipation and temperature control optimization device for aircraft cases provided by the present invention. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0045] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0046] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0047] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing temperature control in liquid cooling of an aircraft case, characterized in that, include: Instantaneous temperature data, temperature change rate data, and heat load data are collected by an array of temperature sensors on the surface of the flight case, and instantaneous temperature deviation value, temperature change rate deviation value, and heat load deviation value are calculated. An error comprehensive evaluation vector is generated based on the instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value; The error comprehensive evaluation vector is input into the predictive temperature control tracking algorithm, and combined with the characteristics of the flight case movement conditions, a temperature tracking prediction value is generated. Based on the temperature tracking and prediction values, perform collaborative optimization of liquid cooling parameters to determine the optimal parameter control strategy, and synchronously adjust the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy.

2. The method for optimizing liquid cooling and temperature control of aircraft cases according to claim 1, characterized in that, The method involves collecting instantaneous temperature data, temperature change rate data, and heat load data through a temperature sensor array on the surface of the flight case, and calculating the instantaneous temperature deviation value, temperature change rate deviation value, and heat load deviation value, including: By arranging a temperature sensor array on the surface of the flight case to collect instantaneous temperature data of each monitoring point in real time, the weighting coefficient is determined according to the positional relationship between each sensor and the liquid cooling pipeline, and the weighted average temperature value is calculated. The difference between the weighted average temperature value and the target temperature value is calculated to obtain the instantaneous temperature deviation value. Collect temperature change data and calculate the temperature change rate data through a preset sliding window. Calculate the difference between the temperature change rate data and the reference temperature change rate to obtain the temperature change rate deviation value. Temperature difference data and flow data of liquid cooling medium inlet and outlet are collected by temperature sensors and flow sensors, respectively. Heat load data are calculated based on the density and specific heat capacity of liquid cooling medium. The difference between the heat load data and the theoretically calculated heat dissipation load value is calculated to obtain the heat load deviation value.

3. The method for optimizing liquid cooling and temperature control of aircraft cases according to claim 1, characterized in that, The generation of a comprehensive error evaluation vector based on the instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value includes: The instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value are vectorized to obtain the initial error vector; The current ambient temperature and working status data of the flight case are obtained through the GPS positioning module, and a dynamic weighting coefficient combination is determined based on the current ambient temperature and the working status data. The initial error vector is squared and multiplied with the corresponding dynamic weight coefficient in the dynamic weight coefficient combination to obtain multiple calculation results; The multiple calculation results are input into the softmax function for dynamic weight allocation to obtain normalized weight allocation results. The normalized weight allocation results are then vectorized through the AI ​​edge computing gateway to obtain the comprehensive error evaluation vector.

4. The method for optimizing liquid cooling and temperature control of aircraft cases according to claim 1, characterized in that, The step of inputting the error comprehensive evaluation vector into the predictive temperature control tracking algorithm, and generating a temperature tracking prediction value by combining the characteristics of the flight case's movement conditions, includes: The error comprehensive evaluation vector is used as the temperature prediction correction parameter input into the predictive temperature control tracking algorithm. The algorithm is then fitted with historical temperature data using the least squares method to obtain the basic data of temperature change trend. Based on the aforementioned temperature change trend data, a mobile operating condition adaptive parameter combination is calculated using a mobile operating condition learning algorithm. When the ambient temperature is between the first temperature value and the second temperature value, the current temperature value and the error comprehensive evaluation vector and its rate of change are calculated according to the temperature change trend prediction function to obtain the temperature tracking prediction value under normal temperature conditions. When the ambient temperature is lower than the first temperature value, the preheating tracking algorithm is activated and the preheating power is adjusted based on the heat load deviation value in the error comprehensive evaluation vector. At the same time, ethylene glycol antifreeze is injected into the liquid cooling pipeline, and pulse heating is used for rapid ice breaking to obtain the temperature tracking prediction value for low-temperature start-up. When the ambient temperature is higher than the second temperature value, the temperature tracking prediction value for high-temperature conditions is calculated based on the instantaneous temperature deviation value and heat load deviation value in the error comprehensive evaluation vector.

5. The method for optimizing liquid cooling and temperature control of aircraft cases according to claim 4, characterized in that, When the ambient temperature is lower than the first temperature value, the preheating tracking algorithm is activated and the preheating power is adjusted based on the heat load deviation value in the error comprehensive evaluation vector. Simultaneously, ethylene glycol antifreeze is injected into the liquid cooling pipeline, and rapid ice breaking is performed using pulse heating to obtain the temperature tracking prediction value for low-temperature startup, including: When the ambient temperature is lower than the first temperature value, the preheating tracking algorithm is started and the ambient temperature and the preheating start time are used as initial parameters. The preheating temperature is calculated based on the preheating tracking function to obtain the basic value of the preheating temperature. The preheating power is dynamically adjusted based on the heat load deviation value in the error comprehensive evaluation vector to obtain the dynamically adjusted preheating power value. Based on the dynamically adjusted preheating power value, ethylene glycol antifreeze is injected into the liquid cooling pipeline through a solenoid valve, while the internal temperature change of the liquid cooling pipeline is monitored to obtain pipeline temperature data after antifreeze injection. Based on the pipeline temperature data after the antifreeze injection, a pulse heating method is used for rapid ice breaking, and the ice breaking heating process is executed by the pulse heating controller according to the dynamically adjusted preheating power value to obtain the temperature change data after ice breaking heating. The preheating temperature baseline value is fused with the temperature change data after ice breaking and heating, and the temperature prediction is corrected by combining the preheating tracking function to obtain the temperature tracking prediction value for low-temperature start-up.

6. The method for optimizing liquid cooling and temperature control of aircraft cases according to claim 1, characterized in that, The step of performing collaborative optimization of liquid cooling parameters based on the temperature tracking prediction value, determining the optimal parameter control strategy, and synchronously adjusting the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy includes: Based on the temperature tracking prediction value, combined with the liquid cooling flow deviation measured by the electromagnetic flowmeter and the pressure fluctuation data measured by the pressure sensor, a three-dimensional optimization state space for the coordinated optimization of liquid cooling heat dissipation parameters is constructed. The performance of the three-dimensional optimization state space is evaluated by optimizing the multi-objective reward function to obtain a comprehensive performance evaluation value. The comprehensive performance evaluation value is used as a reward signal and input into the SAC reinforcement learning decision network for processing, and the optimal parameter control strategy is output. The optimal parameter control strategy is synchronously sent to each sub-controller via the CAN bus. The liquid cooling pump controller adjusts the speed of the liquid cooling pump in the aircraft case, the regulating valve controller adjusts the valve opening, and the fan controller adjusts the frequency of the cooling fan.

7. The method for optimizing liquid cooling and temperature control of an aircraft case according to claim 6, characterized in that, The step of inputting the comprehensive performance evaluation value as a reward signal into the SAC reinforcement learning decision network for processing and outputting the optimal parameter control strategy includes: The three-dimensional optimized state space is used as the state input, and the comprehensive performance evaluation value is used as the reward signal input to the Actor network of the SAC reinforcement learning decision network. Forward propagation calculation is performed through the three-layer neural network structure of the Actor network to generate the action probability distribution. Based on the action probability distribution, candidate action vectors are generated through a Gaussian distribution sampling mechanism. At the same time, the three-dimensional optimized state space and the candidate action vectors are input into the dual Critic network of the SAC reinforcement learning decision network to calculate the state-action value function and obtain the target Q-value evaluation result. Based on the target Q-value evaluation result and the comprehensive performance evaluation value, the strategy loss function and the value loss function are calculated. The parameters of the Actor network and the dual Critic network are updated respectively using the gradient descent algorithm to obtain the optimized network parameters. Based on the optimized network parameters, the Actor network outputs the original control actions, including the liquid cooling pump speed adjustment, the electric regulating valve opening increment, and the cooling fan frequency change. The original control actions are then transformed by action space mapping to obtain the optimal parameter control strategy.

8. The method for optimizing liquid cooling and temperature control of an aircraft case according to claim 7, characterized in that, The process involves generating candidate action vectors based on the action probability distribution using a Gaussian distribution sampling mechanism. Simultaneously, the optimized three-dimensional state space and the candidate action vectors are input into the dual-Critic network of the SAC reinforcement learning decision network to calculate the state-action value function, yielding the target Q-value evaluation result, including: Based on the mean vector and variance vector in the action probability distribution, Gaussian distribution random sampling is performed using the reparameterization technique to generate candidate action vectors. The temperature tracking prediction, liquid cooling flow deviation, and pressure fluctuation data in the three-dimensional optimized state space are concatenated and combined with the candidate action vector to form a state-action input vector. The state-action input vector is input into the first Critic network of the dual Critic network in the SAC reinforcement learning decision network, and the first Q-value evaluation result is obtained by calculating the first Critic network. The state-action input vector is simultaneously input into the second Critic network of the dual Critic network in the SAC reinforcement learning decision network, and the second Q-value evaluation result is obtained by calculating the second Critic network. The minimum value selection operation is performed on the first Q-value evaluation result and the second Q-value evaluation result to obtain the target Q-value evaluation result.

9. A liquid cooling heat dissipation and temperature control optimization system for an aircraft case, characterized in that, For performing the aircraft case liquid cooling heat dissipation and temperature control optimization method as described in any one of claims 1-8, the aircraft case liquid cooling heat dissipation and temperature control optimization system comprises: The calculation module is used to collect instantaneous temperature data, temperature change rate data, and heat load data through the temperature sensor array on the surface of the flight case, and calculate the instantaneous temperature deviation value, temperature change rate deviation value, and heat load deviation value. The generation module is used to generate a comprehensive error evaluation vector based on the instantaneous temperature deviation value, the temperature change rate deviation value, and the heat load deviation value; The prediction module is used to input the error comprehensive evaluation vector into the predictive temperature control tracking algorithm and generate temperature tracking prediction values ​​by combining the characteristics of the flight case movement conditions. The collaborative optimization module is used to perform collaborative optimization of liquid cooling parameters based on the temperature tracking prediction value, determine the optimal parameter control strategy, and synchronously adjust the speed of the liquid cooling pump, the opening degree of the regulating valve, and the frequency of the cooling fan according to the optimal parameter control strategy.

10. A liquid-cooled heat dissipation and temperature control optimization device for aircraft cases, characterized in that, The liquid cooling and heat dissipation temperature control optimization device for the flight case includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the aircraft case liquid cooling heat dissipation temperature control optimization device to execute the aircraft case liquid cooling heat dissipation temperature control optimization method as described in any one of claims 1-8.

Images