Single-phase and two-phase immersion liquid cooling method and system based on AI intelligent decision

Abstract

The invention relates to the technical field of data center heat dissipation, and particularly provides a single-phase and two-phase immersion liquid cooling method and system based on AI intelligent decision, and the method comprises the steps: injecting coupled data into a dynamic feature extraction engine, and outputting a thermodynamic state evolution tensor which comprises the characteristics of a temperature change rate, a load-heat flux density coupling coefficient and the like; the thermodynamic state evolution tensor is input into the deep neural network model, the temperature and pressure matched with the current thermodynamic state evolution tensor are calculated, and a closed-loop control instruction set capable of being executed by equipment is generated; a closed-loop control instruction set is injected into an execution mechanism set, execution mechanisms execute power reconstruction and flow channel switching according to instructions, gaseous fluorinated liquid is liquefied and flows back through an efficient condenser in a two-phase mode, and heat dissipation mode self-adaptive switching and heat cycle reconstruction are achieved. The system comprises a server, an AI algorithm controller, a cooling liquid storage device, a condenser, a circulating pump, an electric valve, a pressure release valve and a temperature sensor. The heat dissipation efficiency and the system reliability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of data center heat dissipation technology, and in particular to a single-phase and two-phase immersion liquid cooling method and system based on AI intelligent decision-making. Background Art

[0002] Traditional air cooling technology, limited by the specific heat capacity and heat dissipation efficiency of air, struggles to meet growing heat dissipation demands. Liquid cooling technology has emerged and is rapidly developing. With the rapid development of technologies like artificial intelligence and big data, data center scale continues to expand, server power density is skyrocketing, and heat dissipation demands are facing unprecedented challenges. Traditional air cooling technology is no longer sufficient for high-heat flux scenarios, making immersion liquid cooling a key industry focus. Existing single-phase and two-phase immersion liquid cooling systems often rely on simple threshold control, determining mode switching based solely on the coolant temperature parameter. This inability to accurately handle complex operating conditions. For example, during sudden spikes in computing power in data centers (such as the instantaneous start of AI model training), server loads fluctuate significantly within a short period of time, resulting in uneven and rapidly changing heat flux distribution. Traditional control modes are prone to switching lag or excessive switching, leading to localized overheating that impacts equipment lifespan, or increased system energy consumption due to frequent switching. Furthermore, traditional systems lack comprehensive consideration of the interplay of multiple factors, including coolant characteristics, real-time server load, and ambient temperature and humidity. This makes it difficult to achieve an optimal balance between heat dissipation efficiency and energy consumption, limiting the efficient and stable operation of data centers.

[0003] Prior art 1, Chinese patent application number 202211150784.7, discloses an immersion liquid cooling heat dissipation method, immersion liquid cooling heat dissipation structure, and liquid cooling system. The immersion liquid cooling heat dissipation method includes: immersing a heat exchange fluid of the immersion liquid cooling system in a heat dissipation component of the liquid cooling system; and a fluid driving component driving the heat exchange fluid upstream of the heat dissipation component to increase the flow rate of the heat exchange fluid through the heat dissipation component, wherein the fluid driving component is positioned horizontally relative to the heat dissipation component. Although the heat exchange fluid is immersed in the heat dissipation component and the fluid driving component is positioned horizontally relative to the heat dissipation component to drive the heat exchange fluid upstream of the heat dissipation component to increase the flow rate of the heat exchange fluid through the heat dissipation component, thereby effectively improving heat exchange efficiency and increasing heat transfer capacity, the flow rate is increased solely through mechanical fluid drive, lacking the ability to respond to dynamic changes in heat flux density, making it impossible to adaptively adjust the cooling mode (single-phase / two-phase) according to load fluctuations, and failing to establish a closed-loop relationship between thermodynamic state and execution control.

[0004] Prior art 2, Chinese patent, application number 202410606085.1 discloses a temperature control system and method for an immersion liquid cooling device, the temperature control system includes an immersion liquid cooling device and a temperature control device, the immersion liquid cooling device is filled with immersion liquid, the immersion liquid flows to the temperature control device through the liquid inlet pipe, and flows back to the immersion liquid cooling device through the liquid return pipe, the temperature control device includes: a temperature detection unit, arranged on the liquid inlet pipe and the liquid return pipe of the immersion liquid, for detecting the inlet temperature and the return temperature of the immersion liquid; a variable frequency liquid pump, connected to the liquid inlet pipe or the liquid return pipe of the immersion liquid, for driving the immersion liquid to flow; a cooling liquid circuit assembly, including a heat exchanger for heat exchange between the immersion liquid and the cooling liquid and a cooling liquid pipeline for conveying the cooling liquid to the heat exchanger, and a throttle valve is connected to the cooling liquid pipeline; and a controller unit, electrically connected to the temperature detection unit, the variable frequency liquid pump and the throttle valve, for measuring data based on the temperature detection unit. Although the operating frequency of the variable frequency liquid pump and the opening of the throttle valve are controlled to regulate the temperature of the immersion liquid, the traditional PID temperature control logic cannot handle nonlinear thermodynamic processes. It only adjusts the flow rate and valve opening, and does not solve the problems of phase change working fluid recovery and thermal cycle reconstruction. The temperature detection unit layout is simple and it is difficult to reflect the local hot spots of the server.

[0005] Prior art three, Chinese patent application number 202410642116.9, discloses a data center immersion liquid cooling system and control method. The cooling system includes dual cooling sources, hybrid cooling capacity distribution, air cooling, and immersion liquid cooling modules. The dual cooling source module is located outside the data center computer room and has three cooling modes: fully natural cooling, mechanical cooling, and dual cooling source cooling. The hybrid cooling capacity distribution, air cooling, and immersion liquid cooling modules within the computer room constitute an air-liquid hybrid cooling system. During operation, the secondary immersion cooling liquid circuit and air cooling circuit remove heat from the equipment and exchange heat with the primary chilled water circuit. The heat is then transferred to the outdoor environment through the hybrid cooling capacity distribution module. Although the energy consumption of the cooling source is reduced by fully utilizing the natural cooling source, the cooling efficiency of the immersion liquid cooling system is improved by coupling active and passive combined heat transfer enhancement technology, and the efficient and low-carbon operation of the dual cooling source air-liquid hybrid cooling system in the data center is achieved through intelligent operation and control strategies, the hybrid cooling system has a complex structure, hysteresis in mode switching, lacks in-depth coordinated control of the natural cooling source utilization and immersion liquid cooling, and does not address the optimization of condensation reflux under two-phase flow conditions.

[0006] Currently, existing technologies 1, 2, and 3 suffer from cooling mode rigidity, dynamic response hysteresis, two-phase flow control instability, and poor adaptability to abnormal operating conditions. Therefore, the present invention provides a single-phase and two-phase immersion liquid cooling method and system based on AI intelligent decision-making. Based on AI intelligent decision-making, a mode switching and heat dissipation optimization system is constructed using multi-dimensional data fusion, accurately adapting to complex operating conditions and breaking through traditional control bottlenecks. Summary of the Invention

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: One aspect of the present invention provides a single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making, comprising the following steps: The thermodynamic state evolution tensor, which includes characteristics such as the temperature change rate and the load-heat flux density coupling coefficient, is input into the deep neural network model. The temperature and pressure adapted to the current thermodynamic state evolution tensor are calculated. The output of the deep neural network model is dynamically corrected by the reinforcement learning compensator based on the deviation between the current actual pressure and temperature and the target value. The corrected instructions are then converted into a closed-loop control instruction set executable by the device through a physical signal converter. The closed-loop control instruction set is injected into the actuator group, and the actuator executes power reconstruction and flow channel switching according to the instructions. The single-phase and two-phase flow channel opening and closing combinations are triggered by the heat flux density gradient threshold; when the flow channel switching is completed, the condensation-reflux closed loop is activated synchronously. In the two-phase mode, the gaseous fluorinated liquid is liquefied and refluxed through the high-efficiency condenser, realizing adaptive switching of the heat dissipation mode and reconstruction of the thermal cycle.

[0008] In an optional implementation, the process of generating a closed-loop control instruction set executable by a device includes the following steps: Based on the output thermodynamic state evolution tensor, multimodal feature decoupling is performed; the temperature change rate feature and the load-heat flux density coupling coefficient contained in the thermodynamic state evolution tensor are separated into independent control channels; Decoupled features are input into a dual-channel deep neural network model for joint inference. The temperature channel network uses historical pressure fluctuation patterns as constraints to output a predicted value for the phase change critical point. The load channel network combines filtered real-time load data to generate optimal heat exchange efficiency parameters. The combined inference results are fed into reinforcement learning compensation, using the deviation between the temperature and pressure data measured by the current sensor and the output of the deep neural network model to construct a dynamic compensation matrix. The compensated control parameters form the preliminary command vector. In the physical signal conversion stage, the preliminary instruction vector is mapped into the device control topology, and the abstract control quantity in the preliminary instruction vector is converted into the physical dimension of the specific execution unit according to the spatial parameters provided by the server heat flux density distribution model.

[0009] In an optional implementation, the process of constructing a dynamic compensation matrix includes the following steps: Based on the predicted values ​​of the temperature channel phase change critical point and the load channel heat exchange efficiency parameters output by the dual-channel neural network, an initial deviation field is established. A three-dimensional difference operation is performed between the sensor's measured temperature and pressure data and the deep neural network's predicted values ​​to form an original deviation vector field with physical dimensions. The original deviation vector field is input into the trend backtracking filter for processing. The generated thermodynamic state evolution tensor is used as the historical benchmark, and the time series pattern of the temperature change rate and load coupling coefficient is extracted as the filtering weight. The filtered deviation field becomes the steady-state deviation representation. Steady-state deviation characterization enters spatial mapping. Based on the spatial topological relationship provided by the heat flux density distribution model, the two-dimensional deviation data is reprojected to the actual physical coordinates of the heat-sensitive area of ​​the server chip, forming a deviation distribution matrix with spatial resolution. A compensation matrix is ​​generated through dynamic weight fusion. The deviation distribution matrix after spatial mapping is used as the basis, and the heat flux density gradient information shared by the middle layer of the dual-channel neural network is superimposed as the adjustment coefficient. Each element value of the fused matrix represents the compensation intensity required for a specific spatial position, completing the conversion from the original deviation to the compensation parameter.

[0010] In an optional embodiment, the process of superimposing the heat flux density gradient information shared by the middle layer of the dual-channel neural network as the adjustment coefficient includes the following steps: The deviation distribution matrix is ​​decomposed into characteristic domains, and the obtained spatial projection results are separated according to the temperature channel deviation component and the load channel deviation component to form two orthogonal temperature channel sub-matrices and load channel sub-matrices; For the temperature channel submatrix, the temperature sensitivity coefficient is extracted from the heat flux density gradient information shared by the middle layer of the dual-channel neural network. The temperature sensitivity coefficient is nonlinearly combined with the credibility index of the phase transition critical point prediction value to generate a temperature compensation weight field. For the load channel submatrix, the spatial correlation intensity in the heat flux density gradient information is used as the modulation factor, carrying the load dynamic characteristics in the thermodynamic state evolution tensor. Combined with the historical fluctuation data of the heat exchange efficiency parameters output by the load channel network, the load compensation weight field is generated through sliding window variance analysis. The compensation matrix is ​​constructed through the tensor product operation of the dual-channel weight field, and the operation results are superimposed through the spatial topological rules of physical signal conversion to form a complete dynamic compensation matrix.

[0011] In an optional embodiment, the process of generating the temperature compensation weight field and the load compensation weight field includes the following steps: Extracting the temperature sensitivity coefficient from the heat flux density gradient information shared by the middle layer of the dual-channel neural network. The temperature sensitivity coefficient includes the instantaneous temperature change rate characteristics and historical phase transition behavior patterns in the thermodynamic state evolution tensor. Dynamically couple the temperature sensitivity coefficient with the output phase transition critical point prediction value credibility index, which reflects the confidence level of the neural network in determining the current phase transition boundary. The output result is the temperature sensitivity-credibility fusion field; The trend backtracking filter established by inputting the temperature sensitivity-credibility fusion field calculates the time decay coefficient of the compensation weight using the time series pattern of the temperature change rate in the historical thermodynamic state evolution tensor; The time attenuation coefficient is nonlinearly superimposed with the fusion field to finally generate a temperature compensation weight field, whose spatial distribution is aligned with the mapped deviation distribution area, forming a high-weight gradient band in the phase transition critical region.

[0012] In an optional embodiment, the process of nonlinearly superposing the time attenuation coefficient and the fusion field includes the following steps: The time decay coefficient output from the trend backtracking filter is calculated based on the temporal pattern of the temperature change rate in the historical thermodynamic state evolution tensor. The time decay coefficient is dynamically adjusted as the trend of the temperature change rate changes. The temperature sensitivity-credibility fusion field already contains the weight distribution information of the phase transition active zone. Each weight value in the temperature sensitivity-credibility fusion field is adjusted using nonlinear mapping, so that the weight presents a smooth transition under the influence of the time attenuation coefficient, rather than a linear mutation. The time attenuation coefficient acts on the weight distribution of the fusion field, making the weight adjustment process conform to the law of thermal inertia; the final generated dynamic compensation weight field forms a high-weight gradient zone in the critical region of phase change.

[0013] In an optional implementation, the process of making the weight adjustment process conform to the thermal inertia law of the system includes the following steps: Extract the temporal fluctuation characteristics of the temperature change rate from the historical thermodynamic evolution tensor and generate a time decay coefficient with spatiotemporal continuity; The preset weight distribution in the temperature sensitivity-credibility fusion field is deeply coupled with the time attenuation coefficient; The adjusted weight field eventually forms a dynamic gradient structure with thermodynamic adaptability: in the phase change active zone, the weight difference between adjacent grid points is controlled within the dynamic threshold determined by the attenuation coefficient, and the dynamic threshold varies negatively with the thermal inertia strength of the system; while in the area where thermal disturbances occur frequently, the weight change rate is constrained within the product of the time attenuation coefficient and the local temperature sensitivity.

[0014] In an optional embodiment, the process of setting the heat flux gradient threshold comprises the following steps: The load-heat flux density coupling coefficient is separated from the thermodynamic state evolution tensor output by the dynamic feature extraction engine. This is then integrated with the temperature change rate predicted by the deep neural network model to generate a regional heat flux dynamic index that comprehensively reflects the dynamic evolution trend of the heat load per unit area under the current operating conditions. The pressure-temperature deviation correction output by the reinforcement learning compensator and the regional heat flow dynamic index are input into the nonlinear mapping module to generate the initial gradient threshold benchmark; The initial gradient threshold is adaptively calibrated using the actuator's flow channel switching history data. When the single-phase flow channel is continuously open for longer than the critical point of condenser efficiency attenuation, the initial gradient threshold is proportionally reduced according to the real-time efficiency coefficient of the condensation-reflux closed loop. If the heat flux oscillation amplitude is detected to exceed the stability margin during two-phase flow channel operation, the initial gradient threshold is dynamically increased according to the inverse relationship of the oscillation frequency.

[0015] In an optional implementation, the real-time data streams of temperature, pressure, and load are screened for outliers using the 3σ principle, and the screened data streams are coupled with the server heat flux density distribution model for spatiotemporal alignment; the coupled data are injected into a dynamic feature extraction engine to output a thermodynamic state evolution tensor.

[0016] Another aspect of the present invention provides a single-phase and two-phase immersion liquid cooling system based on AI intelligent decision-making, which is used to implement the single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making, comprising: a server, an AI algorithm controller, a coolant reservoir, a condenser, a circulation pump, an electric valve, a pressure relief valve, and a temperature sensor; Among them, the top of the server is connected to the inlet of the coolant reservoir through a coolant pipe, the outlet of the coolant reservoir is connected to the inlet of the condenser through a coolant pipe, the outlet of the condenser is connected to the inlet of the circulation pump through a coolant pipe, the outlet of the circulation pump is connected to one end of the electric valve, and the other end of the electric valve is connected to the bottom of the server; a pressure relief valve is installed on the right side of the top of the server, the pressure relief valve is connected to the AI ​​algorithm controller through a data cable, the AI ​​algorithm controller is connected to the temperature sensor through a data cable, and the temperature sensor is embedded in the server.

[0017] This invention leverages AI-driven temperature-pressure coordinated control to fully utilize the characteristics of this boiling-point fluorinated liquid in both single-phase (stable at normal pressure) and two-phase (efficient phase transition at reduced pressure) modes. This improves heat dissipation efficiency compared to traditional single-phase modes, and offers rapid response times, effectively addressing scenarios such as sudden server load changes and environmental fluctuations. The model's self-learning mechanism enables continuous system optimization over time, environmental conditions, and business operations, eliminating the need for frequent manual adjustments. Energy consumption is significantly optimized. Through dynamic pressure regulation and precise mode matching, energy consumption is reduced during off-peak hours, lowering operating costs throughout the entire lifecycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1This is a flow chart of the single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making provided in Example 1 of the present invention; Figure 2 Schematic diagram of the single-phase and two-phase immersion liquid cooling methods based on AI intelligent decision-making provided in Example 1 of the present invention; Figure 3 A process diagram of outputting the thermodynamic state evolution tensor provided in Example 2 of the present invention; Figure 4 A process diagram of generating a closed-loop control instruction set executable by a device provided in Example 4 of the present invention; Figure 5 1. A diagram of the process for setting the heat flux gradient threshold value provided in Example 10 of the present invention; Figure 6 Schematic diagram of the structure of the single-phase and two-phase immersion liquid cooling system based on AI intelligent decision-making provided in Example 11 of the present invention; Figure 7 This is a flow chart of the AI ​​algorithm controller provided in Example 11 of the present invention; Figure 8 This is a schematic diagram of the AI ​​algorithm controller instruction conversion provided in Example 11 of the present invention; Figure 9 A block diagram of the electronic device provided by the present invention; Figure 10 Block diagram of the computer-readable storage medium provided by the present invention. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0020] In the following, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified with "first," "second," etc., may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0021] In the present invention, unless otherwise clearly specified and limited, the term "connection" should be understood in a broad sense. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integrated one; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. In addition, unless otherwise clearly specified and limited, the term "coupling" should be understood in a broad sense. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components, or it can be understood as the electrical connection between different components in a circuit structure through a physical line that can transmit electrical signals, such as printed circuit board (PCB) copper foil or wire, so as to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in an airless / non-contact manner, such as electrical connection between two components using capacitive coupling to transmit electrical signals.

[0022] In an embodiment of the present invention, directional terms such as "up", "down", "left" and "right" may be defined including but not limited to the orientation relative to the schematic placement of the components in the drawings. It should be understood that these directional terms may be relative concepts, which are used for relative description and clarification, and may change accordingly according to changes in the orientation of the components in the drawings.

[0023] Example 1: Figure 1 As shown, an embodiment of the present invention provides a single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making, comprising the following steps: Step S100: The real-time data streams of temperature, pressure, and load are screened for outliers using the 3σ principle. The screened data streams are then spatially and temporally aligned with the server heat flux distribution model. The coupled data is then fed into a dynamic feature extraction engine, which outputs a thermodynamic state evolution tensor containing features such as the temperature change rate and the load-heat flux coupling coefficient. Step S200: The thermodynamic state evolution tensor is input into the deep neural network model, and the temperature and pressure adapted to the current thermodynamic state evolution tensor are calculated. The output of the deep neural network model is dynamically corrected by the reinforcement learning compensator based on the deviation between the current actual pressure and temperature and the target value. The corrected instructions are converted into a closed-loop control instruction set executable by the device through a physical signal converter, including pump power setpoints, valve opening instructions, etc. Step S300: The closed-loop control instruction set is injected into the actuator group. The actuator executes the power reconstruction and flow channel switching according to the instructions. The single-phase and two-phase flow channel opening and closing combinations are triggered by the heat flux density gradient threshold. When the flow channel switching is completed, the condensation-reflux closed loop is synchronously activated. In the two-phase mode, the gaseous fluorinated liquid is liquefied and refluxed through the high-efficiency condenser to realize the adaptive switching of the heat dissipation mode and the reconstruction of the thermal cycle.

[0024] In the above embodiment, Figure 2 This is a schematic diagram of a single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making. This embodiment achieves intelligent dynamic control of the immersion liquid cooling system through the systematic coupling of three key steps. The combined effects of its technical features are as follows: data reliability is ensured through 3σ outlier screening, spatiotemporal alignment coupling eliminates temporal / spatial deviations between sensor data and the server heat flux density distribution model (the server heat flux density distribution model is constructed by fusing computational fluid dynamics simulation data with historical infrared thermal imaging data. A three-dimensional convolutional neural network is used to spatially encode the internal chip layout of the server, establishing a nonlinear mapping relationship between the heat source location and the heat dissipation path. A temporal convolution layer is introduced to capture transient thermal load variation characteristics, ultimately outputting a heat flux density gradient field with spatiotemporal continuity. During the training phase, the model enhances data generalization capabilities through a generative adversarial network, ensuring high-precision modeling of the thermal distribution characteristics of different server architectures). The dynamic feature extraction engine compresses multidimensional data into a thermodynamic state tensor containing key parameters such as the temperature change rate and the load-heat flux coupling coefficient, providing high-fidelity input for subsequent decision-making. A deep neural network model (using a dual-channel heterogeneous architecture, with the temperature prediction channel employing a temporal convolutional network to extract temperature rate characteristics from the thermodynamic state evolution tensor, and the load prediction channel using a graph neural network to model the spatial correlation of the heat flux coupling coefficients) shares heat flux gradient information through a cross-modal attention mechanism. The output layer then integrates historical pressure fluctuation pattern constraints and real-time load data features. Finally, a residual connection preserves the physical properties of the original tensor, forming an intelligent decision-making model that combines temporal dynamics and spatial correlation. This model predicts optimal temperature and pressure parameters based on the thermodynamic tensor. A reinforcement learning compensator dynamically corrects model output errors through real-time bias feedback. A physical signal converter converts the correction instructions into executable control instructions (such as pump power and valve opening), forming a control loop with online adaptability. The actuator dynamically adjusts the flow path topology based on the instructions: switching between single-phase and two-phase flow paths is triggered by a heat flux gradient threshold, enabling adaptive cooling mode selection. In the two-phase mode, the condensation-reflux loop is simultaneously activated, maintaining working fluid balance through liquefaction and recovery of gaseous fluorinated liquid. Coordinated control of power reconfiguration and flow path switching enables rapid reconstruction of the thermal cycle system.

[0025] In summary, this embodiment improves the system's response accuracy to transient load changes through a three-level architecture of abnormal data processing, dynamic modeling, and reinforcement learning compensation. A dual-mode switching mechanism based on the heat flux density gradient threshold achieves dynamic optimal matching of heat dissipation capacity and energy consumption. The coordinated control of the condensation reflux system and the flow channel topology ensures the stability of the working fluid cycle under two-phase working conditions. A complete intelligent control closed loop is formed from data perception to execution feedback, significantly improving heat dissipation efficiency and system reliability.

[0026] Example 2: Figure 3 As shown, based on Example 1, the process of outputting the thermodynamic state evolution tensor in step S100 provided in this embodiment of the present invention includes the following steps: Step S101: The real-time data streams of temperature, pressure and load are input into a dynamic screening boundary field generated based on a historical operating condition library to perform thermodynamic outlier annihilation; Step S102: The annihilated data stream is subjected to heat flow lag compensation with the heat conduction phase field of the server chip material. The compensation amount = heat flux density × material relaxation time, thus completing the spatiotemporal energy alignment. Step S103: Injecting energy alignment data into the intrinsic mode decomposition cavity and outputting the thermodynamic state evolution tensor.

[0027] In the above embodiment, this embodiment identifies and eliminates abnormal data points by dynamically screening boundary fields, ensuring that the input data meets the normal operating range of the system, providing a high-quality data foundation for subsequent processing. The heat flow lag compensation module corrects the timing deviation between the measured data and the actual thermal state caused by the thermal conductivity characteristics of the chip material, and realizes strict spatiotemporal synchronization of the sensor data and the physical process. The intrinsic mode decomposition cavity decomposes the synchronized multi-dimensional data into a state tensor containing the essential thermodynamic characteristics of the system. This tensor fully characterizes the core features such as the dynamic change of temperature, the coupling relationship between load and heat dissipation. The three-stage processing forms a serial data processing chain: anomaly filtering, timing correction, and feature extraction. The final output thermodynamic state evolution tensor has the following characteristics: it eliminates the interference of sensor noise and material thermal inertia, establishes a direct correlation between load changes and heat flow responses, and provides an adaptive data structure for standardized neural network input.

[0028] In summary, this embodiment achieves reliable conversion from raw sensor data to decision-making thermodynamic characteristics, providing accurate state input for subsequent intelligent control.

[0029] Example 3: Based on Example 2, the process of injecting energy alignment data into the eigenmode decomposition cavity in step S103 provided in this embodiment of the present invention includes the following steps: Step S1031: Based on the output spatiotemporal energy alignment data, a multi-dimensional thermodynamic field reconstruction is performed; the compensated temperature gradient distribution, pressure fluctuation characteristics, and load dynamic variation are mapped into a three-dimensional field to form a spatiotemporally continuous heat flux density-material response joint distribution field; Step S1032: The heat flux-material response joint distribution field enters feature decoupling. Based on the topological structure of the heat conduction path within the server chip, the field data is decomposed into an axial conduction component and a radial diffusion component. The axial component reflects the heat accumulation characteristics in the vertical direction of the chip, and the radial component represents the planar heat dissipation capacity. The weights are automatically adjusted by the proportion of historical operating conditions recorded in the dynamic screening boundary field. Step S1033: Nonlinearly superimpose the axial component and the radial component in the boundary transition region. During the superposition process, the calculated heat flux compensation amount is compared in real time to ensure feature fusion under the energy conservation constraint. A transition state thermodynamic characteristic matrix is ​​generated, and its dimension corresponds to the spatial resolution of the chip's thermally sensitive area. Step S1034: Generate a thermodynamic state evolution tensor through variable-scale feature distillation. Based on the transition state matrix, slide and intercept the feature window in the time dimension. Perform the following operations in each window: extract the pressure fluctuation pattern corresponding to the extreme point of temperature change, associate it with the load change rate of the current window, combine the data distribution characteristics retained after the outlier annihilation, and output tensor slices with spatiotemporal correlation. Continuous slices are stacked according to the thermal relaxation cycle to form a complete thermodynamic state evolution tensor.

[0030] In the above-mentioned embodiment, the thermodynamic state evolution tensor of this embodiment is essentially a computable state representation carrier formed through multi-level field reconstruction and feature distillation, combining the thermal response hysteresis characteristics of the chip material, the energy rebalancing process under dynamic load, and the clean data features after filtering out abnormal operating conditions. Using the energy distribution corrected for the material relaxation time as the input reference, feature decoupling and reorganization are achieved based on the chip's thermal conduction topology, and a sliding window distillation mechanism is used to maintain temporal continuity. The entire process forms an end-to-end conversion chain from energy-aligned data to the state tensor.

[0031] Example 4: Figure 4 As shown, based on Example 1, the process of generating a closed-loop control instruction set executable by the device in step S200 provided in this embodiment of the present invention includes the following steps: Step S201: Based on the output thermodynamic state evolution tensor, multimodal feature decoupling is performed; the temperature change rate feature and the load-heat flux density coupling coefficient contained in the thermodynamic state evolution tensor are separated into independent control channels, where the temperature change rate channel is associated with the stability of the phase change process, and the coupling coefficient channel corresponds to the heat dissipation efficiency optimization requirement; Step S202: Decoupled features are input into a dual-channel deep neural network model for joint inference. The temperature channel network uses historical pressure fluctuation patterns as constraints to output a predicted value for the phase change critical point. The load channel network combines the filtered real-time load data to generate optimal heat exchange efficiency parameters. The middle layers of the two networks share heat flux gradient information through a feature cross-attention mechanism. Step S203: The combined inference results are used for reinforcement learning compensation. A dynamic compensation matrix is ​​constructed using the deviation between the temperature and pressure data measured by the current sensor and the output value of the deep neural network model. The update strategy of the dynamic compensation matrix is ​​based on the historical trend component extracted from the thermodynamic state evolution tensor. The compensation amount is adaptively adjusted through time backtracking comparison. The compensated control parameters form the preliminary command vector. Step S204: In the physical signal conversion stage, the preliminary instruction vector is mapped into the device control topology. According to the spatial parameters provided by the server heat flux density distribution model, the abstract control quantity in the preliminary instruction vector is converted into the physical dimension of the specific execution unit. The pump power setting value is generated by the load channel output parameter after correction by the compensation matrix, and the valve opening instruction is determined by combining the temperature channel prediction value and the real-time phase change monitoring data. The conversion process retains the credible data boundary formed after screening out abnormal values ​​according to the 3σ principle.

[0032] In the above-mentioned embodiment, the closed-loop control instruction set generated by this embodiment essentially transforms the system dynamic characteristics represented by the thermodynamic state tensor into a spatiotemporally coordinated execution command sequence through a continuous process of four stages: feature decoupling, dual-channel reasoning, deviation compensation, and physical dimension mapping. A dual-channel decoupling control strategy based on the heat flux density distribution characteristics, a reinforced compensation mechanism utilizing the historical trend of the state evolution tensor, and an instruction conversion method that maintains physical constraints are employed. Each processing step strictly relies on the data characteristics and control parameters output by the previous step, forming a complete closed loop from state representation to execution instructions.

[0033] Example 5: Based on Example 4, the process of constructing the dynamic compensation matrix in step S203 provided in this embodiment of the present invention includes the following steps: Step S2031: Based on the temperature channel phase change critical point prediction value and the load channel heat exchange efficiency parameter output by the dual-channel neural network, an initial deviation field is established; a three-dimensional difference operation is performed on the sensor's measured temperature and pressure data and the deep neural network prediction value, where the temperature dimension difference is associated with the decoupled temperature change rate characteristic channel, and the pressure dimension difference is bound to the load-heat flux density coupling coefficient channel, forming an original deviation vector field with physical dimensions; Step S2032: The original deviation vector field is input into the trend backtracking filter for processing. The generated thermodynamic state evolution tensor is used as the historical benchmark, and the time series pattern of the temperature change rate and the load coupling coefficient is extracted as the filtering weight. During the filtering process, the current deviation vector field is matched with the historical trend data through a sliding window correlation, retaining the deviation components that conform to the system inertia characteristics and filtering out sudden interference signals. The filtered deviation field becomes the steady-state deviation representation. Step S2033: Steady-state deviation characterization enters spatial mapping. Based on the spatial topology provided by the heat flux density distribution model, the two-dimensional deviation data is re-projected to the actual physical coordinates of the heat-sensitive area of ​​the server chip. During the projection process, the load channel deviation is preferentially mapped to the high heat flux density area, while the temperature channel deviation is distributed more heavily in the phase change active area, forming a deviation distribution matrix with spatial resolution. Step S2034: Generate a compensation matrix through dynamic weight fusion, based on the deviation distribution matrix after spatial mapping, and superimpose the heat flux density gradient information shared by the middle layer of the dual-channel neural network as an adjustment coefficient; the temperature channel compensation weight is inversely proportional to the credibility of the phase change critical point prediction value, and the load channel compensation weight is adaptively adjusted with the historical fluctuation amplitude of the heat exchange efficiency parameter; each element value of the fused matrix represents the compensation intensity required for a specific spatial position, completing the conversion from the original deviation to the compensation parameter.

[0034] Among the above embodiments, this embodiment is based on a deviation filtering method based on the historical trend of the thermodynamic state tensor, combined with the projection mapping rules of the spatial characteristics of the heat flux density, and a dynamic weight allocation strategy involving the features of the intermediate layers of the neural network. The entire process inherits the feature decoupling results, joint reasoning output, and spatiotemporal alignment characteristics to form a closed-loop compensation system with physical interpretability. As the key conversion layer connecting intelligent decision-making and executive control, the compensation matrix not only retains the predictive advantages of the neural network, but also ensures the physical rationality of the control instructions through multi-dimensional deviation processing.

[0035] Example 6: Based on Example 5, the process of superimposing the heat flux density gradient information shared by the middle layer of the dual-channel neural network as the adjustment coefficient in step S2034 provided by the embodiment of the present invention includes the following steps: Step S20341: performing eigendomain decomposition on the deviation distribution matrix, separating the obtained spatial projection results into temperature channel deviation components and load channel deviation components, thereby forming two orthogonal temperature channel sub-matrices and load channel sub-matrices; Step S20342: For the temperature channel submatrix, extract the temperature sensitivity coefficient from the heat flux density gradient information shared by the middle layer of the dual-channel neural network; perform a nonlinear combination of the temperature sensitivity coefficient and the phase transition critical point prediction value credibility index to generate a temperature compensation weight field; For the load channel submatrix, the spatial correlation intensity in the heat flux density gradient information is used as the modulation factor, carrying the load dynamic characteristics in the thermodynamic state evolution tensor. Combined with the historical fluctuation data of the heat exchange efficiency parameters output by the load channel network, the load compensation weight field is generated through sliding window variance analysis. Step S20343: The compensation matrix is ​​constructed through the tensor product operation of the dual-channel weight field. The temperature compensation weight field and the temperature channel sub-matrix are Hadamard-multiplied, and the product result reflects the stability requirement of the phase change process. The load compensation weight field and the load channel sub-matrix are Kronecker-multiplied, and the operation result represents the heat dissipation efficiency optimization constraint. The two operation results are superimposed through the spatial topology rules of physical signal conversion to form a complete dynamic compensation matrix.

[0036] Among the above embodiments, this embodiment is based on a weight generation mechanism of the intermediate layer characteristics of the neural network and the historical system behavior, a compensation field distribution strategy that follows the laws of thermodynamic evolution, and a matrix fusion method that maintains the unity of physical dimensions.

[0037] Example 7: Based on Example 6, the process of generating the temperature compensation weight field and the load compensation weight field in step S20342 provided in this embodiment of the present invention includes the following steps: Step S203421: extracting a temperature sensitivity coefficient from the heat flux density gradient information shared by the middle layer of the dual-channel neural network, where the temperature sensitivity coefficient includes the instantaneous temperature change rate characteristics and the historical phase change behavior pattern in the thermodynamic state evolution tensor; The temperature sensitivity coefficient is dynamically coupled with the output phase transition critical point prediction value credibility index, which reflects the neural network's confidence in the current phase transition boundary. The coupling process adopts an adaptive weighting strategy to enhance the compensation weight in low-confidence areas and converge the compensation weight in high-confidence areas. The output result is a temperature sensitivity-credibility fusion field, which shows a smooth transition in the phase transition active region and maintains low gain characteristics in the stable region; Step S203422: Input the temperature sensitivity-credibility fusion field into the established trend backtracking filter, and use the temperature change rate time series pattern in the historical thermodynamic state evolution tensor to calculate the time attenuation coefficient of the compensation weight; The time attenuation coefficient is nonlinearly superimposed on the fusion field to ensure that the dynamic adjustment of the weight field complies with the thermal inertia law of the system and avoids compensation oscillation caused by high-frequency disturbances; Finally, a temperature-compensated weight field is generated, whose spatial distribution is aligned with the mapped deviation distribution area, forming a high-weight gradient band in the phase transition critical region.

[0038] In the above-mentioned embodiments, this embodiment constructs a dynamic adaptive compensation mechanism for precisely adjusting the temperature and pressure control of the immersion liquid cooling system. Temperature sensitivity coefficient extraction and dynamic coupling of credibility - Integrating the real-time temperature change characteristics of the thermodynamic state evolution tensor with the neural network's predicted confidence in the phase transition critical point to form a compensation weight distribution with physical constraints, ensuring that the compensation strategy strengthens control in the phase transition critical region and reduces intervention in the stable region to avoid overcompensation. Trend backtracking filtering and time decay coefficient superposition - Using the historical temperature change rate model to inertially constrain the compensation weight, so that dynamic adjustment conforms to the system's thermodynamic transient characteristics, suppressing control oscillations caused by high-frequency disturbances, and improving closed-loop stability. Spatial alignment and gradient band distribution optimization - The final temperature compensation weight field strictly matches the deviation distribution area, forming a high-weight gradient band in the phase transition active region, ensuring that the compensation effect is precisely applied to the key heat dissipation area, and improving the system's thermal management efficiency under transient load changes.

[0039] In summary, this embodiment implements an adaptive compensation strategy by integrating real-time sensor data, neural network prediction confidence, and historical thermodynamic evolution laws, which can optimize the heat dissipation efficiency of the phase change critical zone while maintaining system stability.

[0040] Example 8: Based on Example 7, the process of nonlinearly superimposing the time attenuation coefficient and the fusion field in step S203422 provided in the embodiment of the present invention includes the following steps: Step S2034221: The time attenuation coefficient output from the trend backtracking filter is calculated based on the time series pattern of the temperature change rate in the historical thermodynamic state evolution tensor to reflect the thermal inertia characteristics of the system at different operating stages; the time attenuation coefficient is dynamically adjusted according to the trend of the temperature change rate. If the historical data shows a sharp temperature fluctuation, the attenuation coefficient is increased accordingly; if the temperature changes smoothly, the attenuation coefficient is correspondingly decreased; Step S2034222: The temperature sensitivity-credibility fusion field already contains the weight distribution information of the phase transition active region, where the weights of the high-credibility region have converged and the weights of the low-credibility region have increased. Nonlinear mapping is used to adjust each weight value in the temperature sensitivity-credibility fusion field so that the weight exhibits a smooth transition under the influence of the time attenuation coefficient, rather than a linear abrupt change. Step S2034223: The time attenuation coefficient acts on the weight distribution of the fusion field so that the weight adjustment process conforms to the thermal inertia law of the system. If the attenuation coefficient is large (the system thermal inertia is strong), the weight adjustment amplitude is reduced to avoid over-compensation; if the attenuation coefficient is small (the system thermal inertia is weak), the weight adjustment amplitude is appropriately increased to ensure rapid response; the final generated dynamic compensation weight field forms a high-weight gradient band in the phase change critical region, while maintaining stability in the high-frequency disturbance area, ensuring that the system's thermal management process is both sensitive and reliable.

[0041] In the above embodiment, the time attenuation coefficient of this embodiment is calculated based on the historical temperature change trend, reflecting the thermal inertia characteristics of the system; the fusion field provides the initial weight distribution, combined with the dynamic adjustment of the time attenuation coefficient, to optimize the adaptability of the compensation strategy; the final weight field not only meets the mapping requirements of the deviation distribution area, but also avoids compensation oscillation, thereby improving the stability and control accuracy of the system.

[0042] Example 9: Based on Example 8, the process of making the weight adjustment process conform to the thermal inertia law of the system in step S2034223 provided by the embodiment of the present invention includes the following steps: Step S20342231: Extract the temporal fluctuation characteristics of the temperature change rate from the historical thermodynamic evolution tensor and generate a time decay coefficient with spatiotemporal continuity to reflect the cumulative effect of the system's thermal inertia. When the standard deviation of the temperature change rate exceeds a critical threshold, the coefficient value increases exponentially with the cube root of the fluctuation amplitude. Step S20342232: Deeply couple the preset weight distribution in the temperature sensitivity-credibility fusion field with the time attenuation coefficient. This coupling process is achieved through a nonlinear transformation of the hyperbolic tangent function: for each grid point in the fusion field, the relative deviation between its original weight and the neighborhood credibility mean is obtained, and this relative deviation is substituted into an adaptive adjustment function containing the time attenuation coefficient. When the attenuation coefficient is large, the function output value is compressed to the square root range of the original value; when the coefficient is small, the output value is amplified inversely to 1.5 times the baseline value. Step S20342233: The adjusted weight field eventually forms a dynamic gradient structure with thermodynamic adaptability: in the phase change active area, the weight difference between adjacent grid points is controlled within the dynamic threshold determined by the attenuation coefficient, and the dynamic threshold changes negatively with the thermal inertia strength of the system; in the area where thermal disturbances occur frequently, the weight change rate is constrained within the product of the time attenuation coefficient and the local temperature sensitivity.

[0043] In the above embodiments, this embodiment ensures that the system can track rapid heat flow changes while suppressing high-frequency noise interference. This cascaded parameter transfer mechanism ensures that each processing step strictly relies on the characteristic quantity output by the previous step, forming a closed-loop optimized thermodynamic response system.

[0044] Example 10: Figure 5 As shown, based on Example 1, the process of setting the heat flux gradient threshold in step S300 provided in this embodiment of the present invention includes the following steps: Step S301: Separate the load-heat flux density coupling coefficient from the thermodynamic state evolution tensor output by the dynamic feature extraction engine, perform feature fusion with the temperature change rate predicted by the deep neural network model, and generate a regional heat flux dynamic index that comprehensively reflects the dynamic evolution trend of the heat load per unit area under the current working conditions; Step S302: The pressure-temperature deviation correction value output by the reinforcement learning compensator and the regional heat flow dynamic index are input into a nonlinear mapping module to generate an initial gradient threshold reference. The initial gradient threshold reference increases logarithmically with the increase of the pressure-temperature deviation correction value and is adjusted by the quadratic term of the heat flow dynamic index. Step S303: The initial gradient threshold is adaptively calibrated using the actuator's flow channel switching history data; when the single-phase flow channel is continuously opened for a period exceeding the condenser efficiency attenuation critical point, the initial gradient threshold is proportionally reduced according to the real-time efficiency coefficient of the condensation-reflux closed loop; if the heat flux density oscillation amplitude is detected to exceed the stability margin during the operation of the two-phase flow channel, the initial gradient threshold is dynamically increased according to the inverse relationship of the oscillation frequency.

[0045] In the above embodiment, the setting process of this embodiment ensures that the threshold is always within the balance range between the prediction range of the thermodynamic state evolution tensor and the actual response capability of the actuator, forming a coherent decision chain from feature extraction to control closed loop.

[0046] Example 11: Figure 6 As shown, based on Examples 1 to 10, the single-phase and two-phase immersion liquid cooling systems based on AI intelligent decision-making provided by the embodiments of the present invention include: a server 1, an AI algorithm controller 2, a coolant reservoir 3, a condenser 4, a circulating pump 5, an electric valve 6, a pressure relief valve 7, and a temperature sensor 8.

[0047] Among them, the top of the server 1 is connected to the inlet end of the coolant reservoir 3 through a coolant pipe, the outlet end of the coolant reservoir 3 is connected to the inlet end of the condenser 4 through a coolant pipe, the outlet end of the condenser 4 is connected to the inlet end of the circulation pump 5 through a coolant pipe, the outlet end of the circulation pump 5 is connected to one end of the electric valve 6, and the other end of the electric valve 6 is connected to the bottom end of the server 1; a pressure relief valve 7 is installed on the right side of the top of the server 1, the pressure relief valve 7 is connected to the AI ​​algorithm controller 2 through a data cable, the AI ​​algorithm controller 2 is connected to the temperature sensor 8 through a data cable, and the temperature sensor 8 is embedded in the server 1.

[0048] In the above embodiment, the formation of the immersion liquid cooling instruction is the process of the AI ​​algorithm controller 2 calculating the optimal temperature-pressure coordination strategy based on real-time operating data through a pre-trained model. The core is the progressive logic of "multi-source data fusion-feature mapping-multi-objective optimization-decision output". The system switches the operating mode through the AI ​​algorithm controller 2, maintaining a single-phase immersion liquid cooling mode at low temperatures or low heat flux density, and switching to a two-phase mode when the temperature exceeds the set value. The coolant used in the system is in a mixed state and is a coexistent fluorinated liquid. A suitable boiling point (for example, 70°C) is selected. When the boiling point is exceeded, the two-phase coolant undergoes a phase change, taking away more heat and returning it to the liquid cooling box through circulating cooling. The AI ​​algorithm controller 2 collects multi-source data such as temperature, heat flux, server load, environmental parameters, system pressure, etc. in real time, and inputs the data into the pre-trained deep neural network model after data cleaning and feature extraction. The model outputs the optimal operating mode (single-phase / two-phase) and equipment control parameters (pump power, valve opening, pressure value), driving the circulation pump, intelligent valve group, and pressure regulating device to work together to achieve precise heat dissipation.

[0049] In specific applications (refer to the attached Figure 7 and attached Figure 8 System Initialization: Technicians inject fluorinated liquid into coolant reservoir 3, ensuring that server 1 is completely submerged. A deep neural network model trained on over 100,000 complex operating conditions (covering different seasons, workloads, and heat flux density distribution scenarios) is imported into AI algorithm controller 2. The model is self-learning, fine-tuning its weights every morning based on newly collected data from the previous day. The intelligent valve group (electric valve 6, pressure relief valve 7) defaults to single-phase flow, maintaining the system's initial pressure at atmospheric pressure. Single-phase liquid cooling mode: Upon system startup, the system defaults to single-phase mode. Circulation pump 5 drives the fluorinated liquid circulation. AI algorithm controller 2 continuously receives data from temperature sensor 8 (the fluorinated liquid absorbs sensible heat in its liquid state and is then cooled by the cooling tower), returning the liquid to the water tank where server 1 is located, completing the circulation.

[0050] AI Algorithm Controller 2 first receives multi-dimensional real-time data from the system and performs targeted processing to provide "high-quality raw materials" for deep neural network model decision-making. This includes core monitoring data: temperature data (coolant inlet and outlet temperature difference (ΔT), server 1 chip surface temperature, average internal temperature of the water tank, and ambient temperature); pressure data (current system absolute pressure, circulation pump 5 outlet pressure, and two-phase flow channel pressure difference); and associated parameters (server 1 CPU / GPU load rate (0-100%), heat flux, circulation pump 5 current power, and real-time valve opening).

[0051] Data preprocessing: Outlier elimination: Use the 3σ principle or isolation forest algorithm to eliminate jump values ​​caused by temperature sensor 8 failure (such as the temperature suddenly soaring to 100°C but the load is 0); feature engineering: Extract dynamic features such as temperature change rate, load fluctuation rate, and coupled features such as load rate-heat flux density ratio and temperature difference-pressure. These features can more accurately reflect operating trends. A sudden increase in load causes a rapid increase in heat flux density, requiring early prediction of pressure adjustment needs.

[0052] Instructions are not static outputs but are adjusted in real time through closed-loop feedback: The deep neural network model receives feedback from the temperature sensor 8 (such as actual pressure and actual temperature) every 10ms, calculates the deviation from the target value, and dynamically corrects the next round of instructions through reinforcement learning. If the actual pressure is too high, the opening command of the pressure reducing device is increased. Instruction conversion is the process of converting the "abstract control parameters" output by the AI ​​model (such as "target pressure 0.08MPa") into physical signals (such as electrical signals or mechanical signals) that the device can recognize. The converted signals act on the device through drive circuits (such as relays and power amplifiers), triggering specific actions, and real-time feedback ensures accurate execution of instructions.

[0053] AI drives the switch to two-phase mode: When the load of server 1 suddenly increases, the AI ​​model makes a decision: after feature extraction of the data (such as an increase in the heat flux density gradient and an increase in the load change rate exceeding the threshold), the input model determines that the current working condition is a complex one of "high heat flux density-high load-high temperature environment" and outputs a switching instruction; the actuator responds: the AI ​​algorithm controller 2 increases the power of the circulation pump and sends a signal to the intelligent valve group at the same time to close the single-phase main channel, open the two-phase dedicated flow channel, and start the pressure reducing device; the gaseous fluorinated liquid is liquefied by the high-efficiency condenser 4 built into the cooling tower and flows back to the bottom of the box to complete the two-phase circulation.

[0054] AI-driven reversion to single-phase mode: When the load on server 1 decreases: The AI ​​model determines that the operating condition is "low heat flux density, low load, and normal temperature environment," and outputs a reversion command; the actuator responds: the power of circulating pump 5 is reduced, the intelligent valve group closes the two-phase flow channel, opens the single-phase main channel, and the pressure regulating device returns the system pressure to normal pressure. The fluorinated liquid then stops phase change, and the system returns to a single-phase sensible heat dissipation cycle. Model self-learning and optimization: The system records "operating condition-mode-energy consumption-heat dissipation effect" data every 15 minutes, and uses the new data to fine-tune the AI ​​model between 2 and 4 a.m. every day (using an online learning algorithm). For example, during high temperature periods in summer, the model automatically increases the weight of the ambient temperature on the pressure regulation parameters; for specific business periods (such as peak financial trading periods), the correlation characteristics between load changes and heat flux density are optimized to ensure the accuracy and efficiency of the system's long-term operation.

[0055] Figure 9A block diagram is shown of an exemplary electronic device suitable for implementing embodiments of the present invention.

[0056] The electronic device may include a central processing unit / microprocessor / main control chip, etc. 9; a storage medium 10, coupled to the central processing unit / microprocessor / main control chip, etc. 9, and storing computer executable instructions therein for performing the steps of each method of an embodiment of the present invention when executed by the processor.

[0057] The central processing unit / microprocessor / main control chip 9 may include but is not limited to one or more processors or microprocessors.

[0058] The storage medium 10 may include, but is not limited to, for example, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (such as hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0059] In addition, the electronic device may also include (but not limited to) a data bus 11, an input / output bus / external bus / device bus 12, a display 13, and input / output devices 14 (eg, keyboard, mouse, speaker, etc.).

[0060] The central processing unit / microprocessor / main control chip etc. 9 can communicate with external devices ( 13 , 14 etc.) via an I / O bus 12 via a wired or wireless network (not shown).

[0061] The storage medium 10 may also store at least one computer executable instruction for executing the various functions and / or method steps in the embodiments described in this technology when executed by the central processing unit / microprocessor / main control chip 9.

[0062] In one embodiment, the at least one computer executable instruction may also be compiled into or constitute a software product, wherein one or more computer executable instructions are executed by a processor to perform the various functions and / or method steps in the embodiments described in the present technology.

[0063] Figure 10 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0064] like Figure 10As shown, the non-transitory computer-readable storage medium 16 stores instructions, such as computer-readable instructions 15. When the computer-readable instructions 15 are executed by the processor, the various methods described above can be executed. Non-transitory computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. Non-transitory non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium 16 can be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions 15 stored on the computer-readable storage medium 16, the various methods described above can be performed.

[0065] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0066] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0067] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0068] 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, or the portion 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 for executing all or part of the steps of the various embodiments of the method of the present invention via a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0069] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making, characterized in that: The following steps are involved: The thermodynamic state evolution tensor, which includes the temperature change rate and load-heat flux density coupling coefficient, is input into the deep neural network model. The temperature and pressure adapted to the current thermodynamic state evolution tensor are calculated. The output of the deep neural network model is dynamically corrected by the reinforcement learning compensator based on the deviation between the current actual pressure and temperature and the target value. The corrected instructions are passed through a physical signal converter to generate a closed-loop control instruction set that can be executed by the device; The closed-loop control instruction set is injected into the actuator group, and the actuator executes power reconstruction and flow channel switching according to the instructions. The single-phase and two-phase flow channel opening and closing combinations are triggered by the heat flux density gradient threshold; when the flow channel switching is completed, the condensation-reflux closed loop is activated synchronously. In the two-phase mode, the gaseous fluorinated liquid is liquefied and refluxed through the high-efficiency condenser, realizing adaptive switching of the heat dissipation mode and reconstruction of the thermal cycle.

2. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 1 is characterized in that: The process of generating a closed-loop control instruction set executable by the device includes the following steps: Based on the output thermodynamic state evolution tensor, multimodal feature decoupling is performed; the temperature change rate feature and the load-heat flux density coupling coefficient contained in the thermodynamic state evolution tensor are separated into independent control channels; Decoupled features are input into a dual-channel deep neural network model for joint reasoning. The temperature channel network uses historical pressure fluctuation patterns as constraints to output the predicted value of the phase transition critical point. The load channel network combines the screened real-time load data to generate the optimal heat exchange efficiency parameters; The combined inference results are fed into reinforcement learning compensation, using the deviation between the temperature and pressure data measured by the current sensor and the output of the deep neural network model to construct a dynamic compensation matrix. The compensated control parameters form the preliminary command vector. In the physical signal conversion stage, the preliminary instruction vector is mapped into the device control topology, and the abstract control quantity in the preliminary instruction vector is converted into the physical dimension of the specific execution unit according to the spatial parameters provided by the server heat flux density distribution model.

3. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 2 is characterized in that: The process of constructing the dynamic compensation matrix includes the following steps: Based on the predicted values ​​of the temperature channel phase change critical point and the load channel heat exchange efficiency parameters output by the dual-channel neural network, an initial deviation field is established. A three-dimensional difference operation is performed between the sensor's measured temperature and pressure data and the deep neural network's predicted values ​​to form an original deviation vector field with physical dimensions. The original deviation vector field is input into the trend backtracking filter for processing. The generated thermodynamic state evolution tensor is used as the historical benchmark, and the time series patterns of the temperature change rate and load coupling coefficient are extracted as filtering weights. The filtered deviation field becomes the steady-state deviation representation; Steady-state deviation characterization enters spatial mapping. Based on the spatial topological relationship provided by the heat flux density distribution model, the two-dimensional deviation data is reprojected to the actual physical coordinates of the heat-sensitive area of ​​the server chip, forming a deviation distribution matrix with spatial resolution. A compensation matrix is ​​generated through dynamic weight fusion. The deviation distribution matrix after spatial mapping is used as the basis, and the heat flux density gradient information shared by the middle layer of the dual-channel neural network is superimposed as the adjustment coefficient. Each element value of the fused matrix represents the compensation intensity required for a specific spatial position, completing the conversion from the original deviation to the compensation parameter.

4. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 3 is characterized in that: The process of superimposing the heat flux density gradient information shared by the middle layer of the dual-channel neural network as the adjustment coefficient includes the following steps: The deviation distribution matrix is ​​decomposed into characteristic domains, and the obtained spatial projection results are separated according to the temperature channel deviation component and the load channel deviation component to form two orthogonal temperature channel sub-matrices and load channel sub-matrices; For the temperature channel submatrix, the temperature sensitivity coefficient is extracted from the heat flux density gradient information shared by the middle layer of the dual-channel neural network. The temperature sensitivity coefficient is nonlinearly combined with the credibility index of the phase transition critical point prediction value to generate a temperature compensation weight field. For the load channel submatrix, the spatial correlation intensity in the heat flux density gradient information is used as the modulation factor, carrying the load dynamic characteristics in the thermodynamic state evolution tensor. Combined with the historical fluctuation data of the heat exchange efficiency parameters output by the load channel network, the load compensation weight field is generated through sliding window variance analysis. The compensation matrix is ​​constructed through the tensor product operation of the dual-channel weight field, and the operation results are superimposed through the spatial topological rules of physical signal conversion to form a complete dynamic compensation matrix.

5. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 4 is characterized in that: The process of generating a temperature compensation weight field and a load compensation weight field includes the following steps: Extracting the temperature sensitivity coefficient from the heat flux density gradient information shared by the middle layer of the dual-channel neural network. The temperature sensitivity coefficient includes the instantaneous temperature change rate characteristics and historical phase transition behavior patterns in the thermodynamic state evolution tensor. Dynamically couple the temperature sensitivity coefficient with the output phase transition critical point prediction value credibility index, which reflects the confidence level of the neural network in determining the current phase transition boundary. The output result is the temperature sensitivity-credibility fusion field; The trend backtracking filter established by inputting the temperature sensitivity-credibility fusion field calculates the time decay coefficient of the compensation weight using the time series pattern of the temperature change rate in the historical thermodynamic state evolution tensor; The time attenuation coefficient is nonlinearly superimposed with the fusion field to finally generate a temperature compensation weight field, whose spatial distribution is aligned with the mapped deviation distribution area, forming a high-weight gradient band in the phase transition critical region.

6. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 5, characterized in that: The process of nonlinear superposition of the time attenuation coefficient and the fusion field includes the following steps: The time decay coefficient output from the trend backtracking filter is calculated based on the temporal pattern of the temperature change rate in the historical thermodynamic state evolution tensor. The time decay coefficient is dynamically adjusted as the trend of the temperature change rate changes. The temperature sensitivity-credibility fusion field already contains the weight distribution information of the phase transition active zone. Each weight value in the temperature sensitivity-credibility fusion field is adjusted using nonlinear mapping, so that the weight presents a smooth transition under the influence of the time attenuation coefficient, rather than a linear mutation. The time attenuation coefficient acts on the weight distribution of the fusion field, making the weight adjustment process conform to the law of thermal inertia; the final generated dynamic compensation weight field forms a high-weight gradient zone in the critical region of phase change.

7. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 6, characterized in that: The process of making the weight adjustment process conform to the thermal inertia law of the system includes the following steps: Extract the temporal fluctuation characteristics of the temperature change rate from the historical thermodynamic evolution tensor and generate a time decay coefficient with spatiotemporal continuity; The preset weight distribution in the temperature sensitivity-credibility fusion field is deeply coupled with the time attenuation coefficient; The adjusted weight field eventually forms a dynamic gradient structure with thermodynamic adaptability: in the phase change active zone, the weight difference between adjacent grid points is controlled within the dynamic threshold determined by the attenuation coefficient, and the dynamic threshold varies negatively with the thermal inertia strength of the system; while in the area where thermal disturbances occur frequently, the weight change rate is constrained within the product of the time attenuation coefficient and the local temperature sensitivity.

8. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 1 is characterized in that: The process of setting the heat flux gradient threshold includes the following steps: The load-heat flux density coupling coefficient is separated from the thermodynamic state evolution tensor output by the dynamic feature extraction engine. This is then integrated with the temperature change rate predicted by the deep neural network model to generate a regional heat flux dynamic index that comprehensively reflects the dynamic evolution trend of the heat load per unit area under the current operating conditions. The pressure-temperature deviation correction output by the reinforcement learning compensator and the regional heat flow dynamic index are input into the nonlinear mapping module to generate the initial gradient threshold benchmark; The initial gradient threshold is adaptively calibrated using the actuator's flow channel switching history data. When the single-phase flow channel is continuously open for longer than the critical point of condenser efficiency attenuation, the initial gradient threshold is proportionally reduced according to the real-time efficiency coefficient of the condensation-reflux closed loop. If the heat flux oscillation amplitude is detected to exceed the stability margin during two-phase flow channel operation, the initial gradient threshold is dynamically increased according to the inverse relationship of the oscillation frequency.

9. The single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to claim 1, characterized in that: The real-time data streams of temperature, pressure, and load are screened for outliers using the 3σ principle. The screened data streams are then spatially aligned with the server heat flux density distribution model. The coupled data is injected into the dynamic feature extraction engine, which outputs the thermodynamic state evolution tensor.

10. A single-phase and two-phase immersion liquid cooling system based on AI intelligent decision-making, used to implement the single-phase and two-phase immersion liquid cooling method based on AI intelligent decision-making according to any one of claims 1 to 9, characterized in that: Includes: server, AI algorithm controller, coolant storage, condenser, circulation pump, electric valve, pressure relief valve, temperature sensor; Among them, the top of the server is connected to the inlet of the coolant reservoir through a coolant pipe, the outlet of the coolant reservoir is connected to the inlet of the condenser through a coolant pipe, the outlet of the condenser is connected to the inlet of the circulation pump through a coolant pipe, the outlet of the circulation pump is connected to one end of the electric valve, and the other end of the electric valve is connected to the bottom of the server; a pressure relief valve is installed on the right side of the top of the server, the pressure relief valve is connected to the AI ​​algorithm controller through a data cable, the AI ​​algorithm controller is connected to the temperature sensor through a data cable, and the temperature sensor is embedded in the server.

Images