Efficient liquid cooling data center waste heat recovery and reutilization device and method

Abstract

The invention discloses an efficient liquid cooling data center waste heat recovery and reutilization device and method, and relates to the field of data center energy efficiency and waste heat recovery. The method comprises the steps of calculating a heat source value density field and a heat sink demand urgency degree field based on waste heat data of a liquid cooling data center and demand data of a heat user side, and performing evolution prediction by using a preset field quantity prediction model to obtain a heat source value density prediction field and a heat sink demand urgency degree prediction field; based on the predicted field quantity, the optimal transportation path of waste heat transportation is solved in the heat supply pipe network, and the conduction entropy of the transportation path is calculated; according to the predicted field quantity, the transport path conduction entropy and the system operation cost, constructing a system situation function, and solving an optimal control instruction by maximizing the system situation function; and executing the optimal control instruction, and updating the parameters of the field quantity prediction model based on the deviation between the actual value and the predicted value of the system situation function. Efficient waste heat utilization is achieved through field quantity modeling, path optimization and dynamic control.

Description

Technical Field

[0001] This invention relates to the field of data center energy efficiency and waste heat recovery, specifically to a high-efficiency liquid-cooled data center waste heat recovery and reuse device and method. Background Technology

[0002] With the rapid development of the digital economy, including cloud computing and artificial intelligence, the energy consumption of global data centers continues to surge. Liquid-cooled data centers account for 30% to 40% of the total energy consumption of data centers. Meanwhile, a large amount of low- and medium-grade heat generated by server operation is typically discharged directly into the environment through liquid-cooled data centers, causing not only enormous energy waste but also exacerbating the urban heat island effect. Therefore, waste heat recovery and reuse from liquid-cooled data centers has become a key technological approach to improve their energy efficiency and reduce carbon emissions.

[0003] Currently, waste heat recovery technology for data centers has been applied, for example, in district building heating or agricultural greenhouse heating. However, existing systems still have room for improvement in achieving precise matching of heat supply and demand, optimizing waste heat transport paths, and enhancing predictive and adaptive capabilities. For example, existing control strategies are mostly based on temperature thresholds or fixed flow rates, which are insufficient to fully reflect the spatiotemporal distribution of waste heat and fluctuations in user demand; in complex heating networks, the selection of transport paths has not yet been dynamically optimized, potentially leading to higher energy consumption and heat loss; furthermore, the system's predictive and adaptive capabilities need to be strengthened to more effectively cope with changes in IT load or the external environment. Summary of the Invention

[0004] Based on the shortcomings of the prior art described above, the purpose of this invention is to provide a high-efficiency liquid-cooled data center waste heat recovery and reuse device and method to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for efficient waste heat recovery and reuse in liquid-cooled data centers, comprising: S1: Based on waste heat data from liquid-cooled data centers and demand data from heat users, calculate the heat source value density field and the heat sink demand urgency field. S2: Using the heat source value density field and the heat sink demand urgency field as the initial state, the heat source value density prediction field and the heat sink demand urgency prediction field are obtained by using the preset field quantity prediction model to perform evolution prediction. S3: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, solve the optimal transport path for waste heat transport in the heating network, and calculate the transport path conduction entropy of the optimal transport path. S4: Based on the heat source value density prediction field, the heat sink demand urgency prediction field, the transport path transmission entropy and the system operating cost, construct the system status function, and solve for the optimal control command by maximizing the system status function; S5: Execute the optimal control command and update the parameters of the field quantity prediction model based on the deviation between the actual and predicted values ​​of the system status function.

[0006] The present invention is further configured such that S1 includes: The waste heat data includes temperature distribution data, cooling medium flow distribution data, and server power distribution data. The demand data includes heat load demand data, temperature requirement data, and heating priority weight data. Based on temperature distribution data, cooling medium flow distribution data, server power distribution data, and ambient temperature data, a heat source value density field is generated. Based on heat load demand data, temperature requirement data, heating priority weight data, and system available temperature data, a heat sink demand urgency field is generated.

[0007] The present invention is further configured such that the field quantity prediction model in S2 includes a spatiotemporal evolution prediction unit and a dynamic response prediction unit: The spatiotemporal evolution prediction unit performs spatiotemporal evolution prediction of the heat source value density field based on historical heat source value density field sequence data and IT load prediction data, and generates a heat source value density prediction field. The dynamic response prediction unit dynamically predicts the heat demand urgency field based on historical heat demand urgency field sequence data and external environment forecast data, generating a heat demand urgency prediction field. The external environment forecast data includes ambient temperature, wind speed, and solar radiation intensity data.

[0008] The present invention is further configured such that S3 includes: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, combined with the topological attributes and real-time operating parameters of the heating network, the transport path conduction entropy of each path in the heating network during the waste heat transport process is calculated. The field strength distribution of the heat source value density prediction field is used as the heat source value distribution, the field strength distribution of the heat sink demand urgency prediction field is used as the heat sink demand distribution, and the transmission path entropy is used as the path loss metric to construct a path search optimization function. Solve the path search optimization function to obtain the optimal set of transport paths for waste heat transport and the corresponding allocated mass flow rate.

[0009] The present invention is further configured such that S4 includes: Based on the intensity of the heat source value density prediction field, the intensity of the heat sink demand urgency prediction field, the intensity of the transport path entropy, and the system operating cost, a system status function is constructed. Within the preset prediction time domain, the optimal control command sequence is solved with the goal of maximizing the cumulative value of the system status function at each time point. Output the optimal control command for the current moment to the liquid cooling system actuator to control the operation of the liquid-cooled data center.

[0010] The present invention is further configured such that S5 includes: Execute optimal control commands and collect real-time operating data from the liquid-cooled data center and heat user terminals; Based on real-time operational data, the actual heat source value density field, heat sink demand urgency field, and transport path entropy are recalculated. The actual value of the system status function is calculated based on the actual heat source value density field, heat sink demand urgency field, transport path transmission entropy, and actual system operating cost. Using the deviation between the actual and predicted values ​​of the system state function as the optimization objective, the gradient descent algorithm is employed to update the parameters of the field quantity prediction model.

[0011] The present invention is further configured such that the calculation steps for the system operating cost include: For each path in the optimal transport path set, calculate the energy cost of the path based on the allocated mass flow rate, pumping pressure boost, pump efficiency, and real-time electricity price. The system operating cost is obtained by summing the energy costs of all paths.

[0012] The present invention is further configured such that when the actual value of the system status function is less than a preset threshold, an abnormal alarm signal is generated.

[0013] The present invention is further configured such that the method also includes visually displaying the heat source value density field, the heat sink demand urgency field, the transport path conduction entropy, and the system state function changes through a graphical interface.

[0014] The present invention also provides a high-efficiency liquid-cooled data center waste heat recovery and reuse device, the device comprising: Calculation module: Based on waste heat data from liquid-cooled data centers and demand data from heat users, calculate the heat source value density field and the heat sink demand urgency field; Field quantity prediction module: Taking the heat source value density field and the heat sink demand urgency field as the initial state, the module uses the preset field quantity prediction model to perform evolution prediction, and obtains the heat source value density prediction field and the heat sink demand urgency prediction field. Path planning module: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, it solves the optimal transport path for waste heat in the heating network and calculates the transport path conduction entropy of the optimal transport path. Optimization control module: Based on the heat source value density prediction field, the heat sink demand urgency prediction field, the transport path conduction entropy and the system operating cost, a system status function is constructed, and the optimal control command is solved by maximizing the system status function; Correction module: Executes optimal control commands and updates the parameters of the field quantity prediction model based on the deviation between the actual and predicted values ​​of the system status function.

[0015] This invention provides a high-efficiency liquid-cooled data center waste heat recovery and reuse device and method. The method includes: S1: calculating the heat source value density field and the heat sink demand urgency field based on waste heat data from the liquid-cooled data center and heat user demand data; S2: using the heat source value density field and the heat sink demand urgency field as initial states, performing evolution prediction using a preset field quantity prediction model to obtain the heat source value density prediction field and the heat sink demand urgency prediction field; S3: solving for the optimal transport path for waste heat transport in the heating network based on the heat source value density prediction field and the heat sink demand urgency prediction field, and calculating the transport path conduction entropy of the optimal transport path; S4: constructing a system status function based on the heat source value density prediction field, the heat sink demand urgency prediction field, the transport path conduction entropy, and the system operating cost, and solving for the optimal control command by maximizing the system status function; S5: executing the optimal control command, and updating the field quantity prediction model parameters based on the deviation between the actual value and the predicted value of the system status function. The beneficial effects include: 1. By introducing the heat source value density field and the heat sink demand urgency field, the spatiotemporal distribution of waste heat supply and demand is quantitatively described, so as to achieve refined matching of supply and demand and improve the efficiency of waste heat recovery. 2. Construct a path search optimization function and solve for the optimal transport path. Combine the system status function to optimize the output control command, so as to maximize the waste heat utilization efficiency and reduce the system's operating energy consumption while ensuring heating demand. 3. By collecting operational data in real time and using a deviation correction prediction model, the system can dynamically adapt to load fluctuations and environmental changes, thereby improving prediction accuracy and system operational stability.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating an exemplary embodiment of the present invention is provided, showing a method for efficient liquid-cooled data center waste heat recovery and reuse. Figure 2 This is a schematic diagram illustrating the structure of a high-efficiency liquid-cooled data center waste heat recovery and reuse device, as an exemplary embodiment of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0019] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0020] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention. Example 1

[0021] A method for efficient waste heat recovery and reuse in liquid-cooled data centers, such as Figure 1 As shown, it includes: S1: Based on waste heat data from liquid-cooled data centers and demand data from heat users, calculate the heat source value density field and the heat sink demand urgency field. S2: Using the heat source value density field and the heat sink demand urgency field as the initial state, the heat source value density prediction field and the heat sink demand urgency prediction field are obtained by using the preset field quantity prediction model to perform evolution prediction. S3: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, solve the optimal transport path for waste heat transport in the heating network, and calculate the transport path conduction entropy of the optimal transport path. S4: Based on the heat source value density prediction field, the heat sink demand urgency prediction field, the transport path transmission entropy and the system operating cost, construct the system status function, and solve for the optimal control command by maximizing the system status function; S5: Execute the optimal control command and update the parameters of the field quantity prediction model based on the deviation between the actual and predicted values ​​of the system status function.

[0022] The present invention is further configured such that S1 includes: The waste heat data includes temperature distribution data, cooling medium flow distribution data, and server power distribution data. The demand data includes heat load demand data, temperature requirement data, and heating priority weight data. Based on temperature distribution data, cooling medium flow distribution data, server power distribution data, and ambient temperature data, a heat source value density field is generated. Based on heat load demand data, temperature requirement data, heating priority weight data, and system available temperature data, a heat sink demand urgency field is generated. Specifically, a distributed temperature sensor array deployed in the liquid-cooled data center collects real-time temperature distribution data at key locations such as server rack inlets and outlets and cooling coil surfaces; electromagnetic flowmeters collect real-time volumetric flow rate data of the cooling medium in the pipes and convert it into cooling medium flow distribution data; the server management interface obtains real-time power consumption data of each computing node to obtain server power distribution data; and the heating network monitoring system obtains heat load demand data and required heating temperature data for each heat exchange station. The system retrieves preset heating priority weighting coefficients from the energy management system. These coefficients are determined based on a comprehensive assessment of heat user type and heating security level. Ambient temperature data is obtained in real-time from weather stations. The collected discrete sensor data is spatially gridded. A three-dimensional computational domain is defined based on the physical geometry of the liquid-cooled data center and discretized into grid points. A spatial interpolation algorithm is used to reconstruct the discrete temperature distribution data into a continuous three-dimensional temperature field. For each grid point in the three-dimensional spatial field, the difference between the temperature value at that grid point and the ambient dry-bulb temperature value is calculated, and the difference is squared. This squared operation is used for nonlinear amplification. The value of high-grade heat is calculated by multiplying the squared temperature difference by the density and specific heat capacity of the cooling medium at that grid point, and then relating this value to the heat capacity per unit volume to obtain the basic heat energy value. This basic heat energy value is then divided by the sum of the temperature value at that grid point and an adjustment constant to obtain the heat source value density value at that grid point. The adjustment constant is used to ensure the stability and rationality of the numerical calculation, avoiding calculation anomalies where the denominator is zero or negative at extremely low temperatures; its value range is [1K, 10K]. After traversing all spatial grid points to complete the heat source value density value calculation, a heat source value density field is generated. This heat source value density field visually reflects the waste heat at different locations within the data center. Thermodynamic value is assessed; for each heat user, the difference between the user's required heating temperature and the system's available temperature is calculated, and this difference is input into a negative exponential function for mapping, outputting a matching coefficient between 0 and 1. When the system's available temperature is much lower than the required heating temperature, the matching coefficient approaches 0, indicating a poor matching; when the two are close, the matching coefficient rapidly approaches 1, indicating a good matching. The heat load demand data, matching coefficient, and heating priority weight coefficient of each heat user are weighted and fused to calculate the heat sink demand urgency value for that heat user. After completing the calculation for all heat users, a heat sink demand urgency field is generated for the entire heat user set.

[0023] The present invention is further configured such that the field quantity prediction model in S2 includes a spatiotemporal evolution prediction unit and a dynamic response prediction unit: The spatiotemporal evolution prediction unit performs spatiotemporal evolution prediction of the heat source value density field based on historical heat source value density field sequence data and IT load prediction data, and generates a heat source value density prediction field. The dynamic response prediction unit dynamically predicts the heat demand urgency field based on historical heat demand urgency field sequence data and external environment forecast data, generating a heat demand urgency prediction field. The external environment forecast data includes ambient temperature, wind speed, and solar radiation intensity data. Specifically, the field prediction model consists of two independent prediction units: a spatiotemporal evolution prediction unit and a dynamic response prediction unit. The spatiotemporal evolution prediction unit is configured to process the heat source value density field with spatial continuity, constructing a physical field evolution model based on the convection-diffusion equation combined with the server heat source term. This physical field evolution model describes the mathematical relationship between heat flowing with the cooling medium, spontaneous diffusion within the medium, and server heat generation. The dynamic response prediction unit is configured to process the urgency field of heat sink demand from users, constructing a time series prediction model based on historical heat sink demand urgency field sequence data and external environment forecast data. The time series forecasting model can capture the dynamic response relationship between changes in the external environment and heat demand. It uses continuously collected heat source value density field sequence data from historical periods as the initial state input to the physical field evolution model, while simultaneously receiving server power distribution data within a preset prediction time domain from the IT management system. Based on the cooling medium flow distribution data obtained in the preceding steps, a velocity vector field is generated through interpolation to simulate the driving effect of cooling medium flow on heat transport. The thermal diffusivity coefficient is dynamically determined based on the turbulence state parameters of the cooling medium. The Reynolds number is calculated based on real-time collected cooling medium density, flow velocity, dynamic viscosity, and pipe hydraulic diameter to quantify the turbulence intensity of the current flow. The Reynolds number is then substituted into pre-tested data obtained from experiments or other methods. In the correlation equations calibrated by computational fluid dynamics simulation, the corresponding thermal diffusivity coefficient is obtained to simulate the diffusion process of heat under turbulent conditions. IT load prediction data, i.e., server power distribution data, is converted into the heat generation rate of the corresponding spatial location according to a preset ratio, serving as the heat source term driving the evolution of the physical field. The physical effects of convection, diffusion, and the heat source term are substituted into the governing equations describing the spatiotemporal evolution of the heat source value density field, i.e., the convection-diffusion-reaction equations, to construct a closed mathematical model of the physical field evolution. This mathematical model is then solved using numerical discretization methods, such as discretizing the three-dimensional computational domain into a mesh using the finite volume method, establishing discrete equations on each mesh cell, and forming a large system of linear algebraic equations. Then, the computational fluid dynamics simulation is called... A dynamic solver or a trained physical information neural network numerically solves the equation set. During this process, a historical heat source value density field sequence is used as the initial condition, and iterative calculations are performed within the prediction time domain at a preset time step, ultimately outputting a continuous heat source value density prediction field. Continuously collected heat sink demand urgency field sequence data from the historical time period is used as the initial state input to the time series prediction model. Simultaneously, external environmental forecast data within the preset prediction time domain, including ambient temperature, wind speed, and solar radiation intensity, are received. The environmental forecast data is then comprehensively calculated using a nonlinear mapping function to obtain a meteorological impact factor. This meteorological impact factor is used to quantify the comprehensive impact of meteorological conditions on building heat load or user heating habits.The core of the time series forecasting model is the difference equation, which represents the rate of change of heat sink demand urgency at the next moment as a linear combination of the current heat sink demand urgency and meteorological influence factors, weighted by state autoregressive coefficients and environmental driving gain coefficients, respectively. The state autoregressive coefficients and environmental driving gain coefficients are trained using historical heat load demand data and historical external environmental forecast data. The state autoregressive coefficients quantify the influence of the system's historical state on future changes, while the environmental driving gain coefficients quantify the driving strength of external environmental changes on heat demand changes. The state autoregressive coefficients range from [0.6, 0.9], and the environmental driving gain coefficients range from [0.1, 0.4]. Starting from the last moment of the historical heat sink demand urgency field sequence, the trained time series forecasting model and environmental forecast data are used to iteratively calculate the heat sink demand urgency of each heat user at each time step within the preset forecast time domain, generating a continuous heat sink demand urgency forecast field.

[0024] The present invention is further configured such that S3 includes: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, combined with the topological attributes and real-time operating parameters of the heating network, the transport path conduction entropy of each path in the heating network during the waste heat transport process is calculated. The field strength distribution of the heat source value density prediction field is used as the heat source value distribution, the field strength distribution of the heat sink demand urgency prediction field is used as the heat sink demand distribution, and the transmission path entropy is used as the path loss metric to construct a path search optimization function. Solving the path search optimization function yields the optimal set of waste heat transport paths and their corresponding allocated mass flow rates. Specifically, the heating network is abstracted as a graph structure, where nodes represent heat sources, heat sinks, heat user access points, and pipeline junctions, and edges represent pipeline segments connecting these nodes, thus forming a graph model that can be used for waste heat transport path calculation and optimization. Sensors deployed at key pipeline nodes are used to collect real-time data on the fluid medium temperature, fluid mass flow rate, and valve opening status for each pipeline. For each pipeline in the heating network, the temperature gradient along the pipeline length is calculated based on temperature measurements at both ends. The square of this temperature gradient is then compared with the thermal properties of the pipeline material. Multiplying the conductivity by the pipe cross-sectional area yields the thermal entropy yield per unit length. This yield reflects the energy quality degradation caused by temperature difference during the spontaneous heat transfer from a high-temperature to a low-temperature region. Multiplying the cube of the mass flow rate of the fluid medium in the pipe by the coefficient of friction, and then dividing by the square of the fluid density, the hydraulic diameter of the pipe, and the square of the pipe cross-sectional area, yields the flow friction entropy yield per unit length. This yield reflects the energy dissipation caused by the pump work consumed to overcome the fluid's viscous resistance. The coefficient of friction depends on the fluid flow state and the roughness of the pipe wall, and can be determined by consulting a Moody's diagram or empirical formulas, such as the Kolbrook-White formula. The rates are summed, and the sum is integrated along the entire length of the pipeline to obtain the transport path conduction entropy value of the pipeline. This transport path conduction entropy value is used to quantify the total energy quality loss caused by irreversible thermodynamic processes when heat is transported within the pipeline. After calculating the transport path conduction entropy values ​​of all pipelines, each edge of the pipeline network diagram is assigned its corresponding transport path conduction entropy value. The path search optimization function consists of a transport value term and a transport loss term. The transport value term is obtained based on the heat source value density prediction field and the heat sink demand urgency prediction field. For each possible combination of heat source node and heat sink node, the product of its heat source value density value and heat sink demand urgency value is calculated. This product reflects the transfer of heat from the heat source. The comprehensive value created by transporting goods to the heat sink is obtained by summing the products of all possible pairings to get the total transport value that needs to be maximized. The transport loss term is obtained by traversing all candidate paths, accumulating the transport path conduction entropy values ​​of the pipelines traversed by each path, multiplying them by the path length for weighting, and obtaining the loss of each path. The sum of all path losses is the total transport loss that needs to be minimized. The total transport value is multiplied by the value weight coefficient and then the total transport loss is subtracted to form the path search optimization function. This path search optimization function is used to seek the best balance between maximizing transport value and minimizing transport loss. The value weight coefficient is used to adjust the relative importance of the total transport value in the path search optimization function, and its value range is [0, 1].[5,2]; Based on the above path search optimization function, constraints are set, including that all pipeline junction nodes satisfy the flow conservation constraint, i.e., the net mass flow into non-source and non-sink nodes is zero, and the pipeline capacity constraint that the mass flow allocated to each pipeline does not exceed its design capacity; this constrained optimization problem is modeled as a network flow problem and solved using mathematical programming algorithms, such as the Lagrange relaxation method combined with subgradient optimization, to output the optimal set of transport paths that satisfy the constraints and achieve optimal transport efficiency or cost optimization indicators, along with the allocated mass flow for each path.

[0025] The present invention is further configured such that S4 includes: Based on the intensity of the heat source value density prediction field, the intensity of the heat sink demand urgency prediction field, the intensity of the transport path entropy, and the system operating cost, a system status function is constructed. Within the preset prediction time domain, the optimal control command sequence is solved with the goal of maximizing the cumulative value of the system status function at each time point. The optimal control command for the current moment is output to the liquid cooling system actuator to control the operation of the liquid-cooled data center; the invention is further configured such that the calculation steps for the system operating cost include: For each path in the optimal transport path set, calculate the energy cost of the path based on the allocated mass flow rate, pumping pressure boost, pump efficiency, and real-time electricity price. The system operating cost is obtained by summing the energy consumption costs of all paths. Specifically, the system status function is used to quantitatively evaluate the overall operating status of the waste heat recovery system at a future moment. This system status function is constructed using four key performance indicators: heat source value, heat sink demand, path loss, and operating cost. The heat source value is obtained by summing the squares of the values ​​of the heat source value density prediction field at all spatial grid points and taking the square root of the summation result to obtain the overall intensity scalar of the heat source value density prediction field. This scalar is used to comprehensively measure the total value potential of the data center that can provide waste heat within the prediction time domain. The heat sink demand is obtained by summing the squares of the values ​​of the heat sink demand urgency prediction field at all heat user points and taking the square root of the summation result. The overall intensity scalar of the heat sink demand urgency prediction field is obtained by squaring the data. This scalar measures the urgency of heat users' total demand for waste heat within the prediction time domain. The path loss term is obtained by summing the squares of the transport path conduction entropy of all paths in the optimal transport path set and taking the square root of the summation. This scalar reflects the overall irreversible loss caused by transporting heat on the selected path. The operating cost term is obtained by calculating the energy consumption cost of the pumping equipment on each path in the optimal transport path set by multiplying the allocated mass flow rate, pumping pressure increase, and real-time electricity price of the path by the pump efficiency. The system operating cost is obtained by summing the energy consumption costs of all paths. Preset weighting coefficients are assigned to the heat source value item, heat sink demand item, path loss item, and operating cost item. The system status function is constructed by multiplying the heat source value item and heat sink demand item by their respective weighting coefficients and summing the results, and then subtracting the sum of the path loss item and operating cost item multiplied by their respective weighting coefficients. The weighting coefficients are used to adjust the relative importance of each performance indicator in the comprehensive decision-making process, and their values ​​range from [0,1]. The preset prediction time domain is divided into several discrete time steps, and the cumulative value of the system status function at each time step is used as the optimization objective, which needs to be maximized. The optimization variables are the system control commands at each time step in the prediction time domain, such as the specific speed command of the water pump and the specific opening command of the valve. The optimization process must meet the physical limits of the equipment. Constraints are applied to ensure that control commands remain within the permissible operating range of the equipment, and the system dynamics model ensures that changes in control commands conform to the inherent physical response laws of the liquid-cooled data center, such as the relationship between flow rate, valve position, and pump speed. The aforementioned optimization problem is a constrained nonlinear programming problem, which can be solved using numerical optimization algorithms, such as sequential quadratic programming, to obtain the optimal control command sequence for each time step within the preset prediction time domain. Only the optimal control command at the current moment is output to the liquid-cooled system actuators. The optimal control command obtained at the current moment, such as specific pump speed and valve position signals, is sent to the underlying actuators of the liquid-cooled data center, such as frequency converters and electric valves, to adjust the mass flow distribution and system pressure in real time, thereby controlling the operating status of the waste heat recovery process.

[0026] The present invention is further configured such that S5 includes: Execute optimal control commands and collect real-time operating data from the liquid-cooled data center and heat user terminals; Based on real-time operational data, the actual heat source value density field, heat sink demand urgency field, and transport path entropy are recalculated. The actual value of the system status function is calculated based on the actual heat source value density field, heat sink demand urgency field, transport path transmission entropy, and actual system operating cost. Using the deviation between the actual and predicted values ​​of the system state function as the optimization objective, the gradient descent algorithm is employed to update the parameters of the field quantity prediction model. Specifically, after the control command is executed, real-time operating data of the liquid-cooled data center and heat user terminals are collected synchronously. This real-time operating data includes the cooling medium temperature at key points within the data center, the mass flow rate of the medium in the main and branch pipes, and the actual heat load demand readings of each heat exchange station. Based on the collected real-time operating data, the heat source value density field, heat sink demand urgency field, and transport path conduction entropy are recalculated using the same calculation rules as steps S1 and S3 to obtain the true state of the system at the current moment. Real-time data is then utilized... Based on temperature distribution data, ambient temperature data, and cooling medium properties, the actual heat source value density field for the current moment is recalculated and generated. This actual heat source value density field reflects the true quality and spatial distribution value of the current data center waste heat. Using real-time heat load demand, system available temperature, and heating priority weights for each user, the actual heat sink demand urgency field for the current moment is recalculated and generated. This actual heat sink demand urgency field reflects the current actual urgency of heat users' needs. Based on the real-time temperature gradient and mass flow rate data of the pipeline network, the actual transport path conduction entropy for each transport path under the current operating conditions is recalculated. This transport path conduction entropy is used to quantify heat. Irreversible losses during actual transport; substituting the actual heat source value density field, actual heat sink demand urgency field, actual transport path conduction entropy, and the actual system operating cost calculated based on actual power consumption into the system status function formula defined in step S4, the actual value of the system status function is calculated; this actual value is compared with the system status function value predicted in step S4 of the previous control cycle at the corresponding time, and the deviation between the two is calculated. This deviation is used to quantify the accuracy of the prediction model in the previous cycle; with the goal of minimizing the prediction deviation of the system status function, a model parameter optimization problem is constructed. The essence of this problem is to adjust the key parameters in the field quantity prediction model so that the model's prediction output is as close as possible to the expected value. To approximate the actual operating results of the system, the field quantity prediction model parameters include the thermal diffusivity coefficient, the state autoregressive coefficient, and the environment-driven gain coefficient. A gradient descent algorithm is used to automatically correct the field quantity prediction model parameters. This algorithm determines the adjustment direction and step size of the parameters by calculating the gradient of the prediction deviation value of the system's state function relative to the model parameters to be adjusted, such as the thermal diffusivity coefficient in the physical field evolution model. Specifically, the model parameters are fine-tuned along the opposite direction of the obtained gradient with a preset learning rate, thereby updating the model parameters in a direction that reduces prediction deviation. The adjusted new parameter values ​​are then updated in the field quantity prediction model, replacing the old parameters, completing one online learning cycle of the model.The updated field quantity prediction model parameters are fed back into the field quantity prediction model in step S2. At the start of the next control cycle, the field quantity prediction model will perform a new round of predictions based on the latest corrected model parameters, enabling the field quantity prediction model to continuously adapt to changes in the system's dynamic characteristics, thereby forming an adaptive predictive control closed loop with self-improvement capabilities.

[0027] The invention is further configured such that when the actual value of the system status function is less than a preset threshold, an abnormal alarm signal is generated. Specifically, the preset threshold is obtained by collecting a historical sequence of system status function values ​​during long-term normal operation, removing data segments containing faults, maintenance, or abnormal operating conditions, and retaining only sample data under healthy operating conditions. Statistical analysis is performed on the filtered sample data to calculate the weighted mean or weighted median of the system status function values ​​as the typical performance level under optimal operating conditions. A safety factor is introduced based on the typical performance level to obtain the preset threshold. The safety factor is set according to the sample variance or quantile calculation results to ensure that the threshold can effectively distinguish between healthy and abnormal states even with slight disturbances. The system status function is calculated in real time during each control cycle. When the actual value of the system status function is less than a preset threshold, the system is determined to be in a state of performance degradation or supply-demand imbalance, and an abnormal alarm signal is generated. This signal is output to the upper-level monitoring platform and human-machine interface in the form of digital or logical quantities. This signal can be used to prompt maintenance personnel to perform manual inspection and intervention, and trigger backup control modes, such as switching to a conservative PID control strategy, to ensure the basic operational safety of the system when optimization fails. Through the above design, this embodiment realizes performance degradation identification, abnormal alarm and linkage control based on real-time system status function monitoring, ensuring that the system still has safety protection capabilities when the adaptive optimization effect is insufficient.

[0028] The invention is further configured such that the method also includes visually displaying the heat source value density field, heat sink demand urgency field, transport path entropy, and system state function changes through a graphical interface; specifically, the spatiotemporal distribution and dynamic changes of the heat source value density field, heat sink demand urgency field, transport path entropy, and system state function are rendered in real time on the graphical monitoring interface in a graphical manner. The heat source value density field is superimposed on the data center physical layout map in the form of a three-dimensional color cloud map or a two-dimensional contour map, using color depth or contour line density to indicate the relative levels of waste heat quality in different areas; the heat sink demand urgency field is represented on the heating network topology map by nodes of different sizes or colors to indicate the urgency level of each heat user, facilitating rapid identification of high-priority users; the transport path entropy is represented in the network map by pipelines of different thicknesses or colors to indicate the entropy of each path. The size is used to intuitively reflect the degree of transportation loss; the system status function displays historical and real-time change trends in the form of a time-series curve, and marks the predicted and actual values ​​for performance evaluation and deviation analysis; the graphical interface is dynamically linked with the real-time data stream and calculation results of steps S1 to S5, ensuring that the displayed content is automatically refreshed with the system operating status, enabling operators to quickly locate efficient heat sources, identify transportation bottlenecks, and evaluate the overall system performance based on intuitive graphical information, thereby improving the transparency of the system operating status and providing support for scheduling decisions; the graphical interface also integrates and displays the real-time operating status data of the thermal storage equipment, including the current heat storage capacity, heat charging and discharging rate, and equipment capacity percentage. This status data is obtained in real time from the thermal storage equipment control system through the data interface and presented in the form of dynamic curves or filled animations, providing operators with an intuitive basis for adjusting thermal storage strategies. Example 2

[0029] Please see Figure 2 This exemplary high-efficiency liquid-cooled data center waste heat recovery and reuse device includes: Calculation module: Based on waste heat data from liquid-cooled data centers and demand data from heat users, calculate the heat source value density field and the heat sink demand urgency field; Field quantity prediction module: Taking the heat source value density field and the heat sink demand urgency field as the initial state, the module uses the preset field quantity prediction model to perform evolution prediction, and obtains the heat source value density prediction field and the heat sink demand urgency prediction field. Path planning module: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, it solves the optimal transport path for waste heat in the heating network and calculates the transport path conduction entropy of the optimal transport path. Optimization control module: Based on the heat source value density prediction field, the heat sink demand urgency prediction field, the transport path conduction entropy and the system operating cost, a system status function is constructed, and the optimal control command is solved by maximizing the system status function; Correction module: Executes optimal control commands and updates the parameters of the field quantity prediction model based on the deviation between the actual and predicted values ​​of the system status function.

[0030] It should be noted that the high-efficiency liquid-cooled data center waste heat recovery and reuse device provided in the above embodiments and the high-efficiency liquid-cooled data center waste heat recovery and reuse method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the high-efficiency liquid-cooled data center waste heat recovery and reuse device provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0031] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for efficient waste heat recovery and reuse in liquid-cooled data centers, characterized in that, include: S1: Based on waste heat data from liquid-cooled data centers and demand data from heat users, calculate the heat source value density field and the heat sink demand urgency field. S2: Using the heat source value density field and the heat sink demand urgency field as the initial state, the heat source value density prediction field and the heat sink demand urgency prediction field are obtained by using the preset field quantity prediction model to perform evolution prediction. S3: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, solve the optimal transport path for waste heat transport in the heating network, and calculate the transport path conduction entropy of the optimal transport path. S4: Based on the heat source value density prediction field, the heat sink demand urgency prediction field, the transport path transmission entropy and the system operating cost, construct the system status function, and solve for the optimal control command by maximizing the system status function; S5: Execute the optimal control command and update the parameters of the field quantity prediction model based on the deviation between the actual and predicted values ​​of the system status function.

2. The method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 1, characterized in that, S1 includes: The waste heat data includes temperature distribution data, cooling medium flow distribution data, and server power distribution data. The demand data includes heat load demand data, temperature requirement data, and heating priority weight data. Based on temperature distribution data, cooling medium flow distribution data, server power distribution data, and ambient temperature data, a heat source value density field is generated. Based on heat load demand data, temperature requirement data, heating priority weight data, and system available temperature data, a heat sink demand urgency field is generated.

3. The method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 1, characterized in that, The field quantity prediction model in S2 includes a spatiotemporal evolution prediction unit and a dynamic response prediction unit: The spatiotemporal evolution prediction unit performs spatiotemporal evolution prediction of the heat source value density field based on historical heat source value density field sequence data and IT load prediction data, and generates a heat source value density prediction field. The dynamic response prediction unit dynamically predicts the heat demand urgency field based on historical heat demand urgency field sequence data and external environment forecast data, generating a heat demand urgency prediction field. The external environment forecast data includes ambient temperature, wind speed, and solar radiation intensity data.

4. The method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 1, characterized in that, S3 includes: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, combined with the topological attributes and real-time operating parameters of the heating network, the transport path conduction entropy of each path in the heating network during the waste heat transport process is calculated. The field strength distribution of the heat source value density prediction field is used as the heat source value distribution, the field strength distribution of the heat sink demand urgency prediction field is used as the heat sink demand distribution, and the transmission path entropy is used as the path loss metric to construct a path search optimization function. Solve the path search optimization function to obtain the optimal set of transport paths for waste heat transport and the corresponding allocated mass flow rate.

5. The method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 1, characterized in that, S4 includes: Based on the intensity of the heat source value density prediction field, the intensity of the heat sink demand urgency prediction field, the intensity of the transport path entropy, and the system operating cost, a system status function is constructed. Within the preset prediction time domain, the optimal control command sequence is solved with the goal of maximizing the cumulative value of the system status function at each time point. Output the optimal control command for the current moment to the liquid cooling system actuator to control the operation of the liquid-cooled data center.

6. The method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 1, characterized in that, S5 includes: Execute optimal control commands and collect real-time operating data from the liquid-cooled data center and heat user terminals; Based on real-time operational data, the actual heat source value density field, heat sink demand urgency field, and transport path entropy are recalculated. The actual value of the system status function is calculated based on the actual heat source value density field, heat sink demand urgency field, transport path transmission entropy, and actual system operating cost. Using the deviation between the actual and predicted values ​​of the system state function as the optimization objective, the gradient descent algorithm is employed to update the parameters of the field quantity prediction model.

7. The method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 5, characterized in that, The calculation steps for the system operating cost include: For each path in the optimal transport path set, calculate the energy cost of the path based on the allocated mass flow rate, pumping pressure boost, pump efficiency, and real-time electricity price. The system operating cost is obtained by summing the energy costs of all paths.

8. The method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 6, characterized in that, An abnormal alarm signal is generated when the actual value of the system status function is less than a preset threshold.

9. A method for efficient liquid-cooled data center waste heat recovery and reuse according to claim 1, characterized in that, The method also includes visually displaying the heat source value density field, heat sink demand urgency field, transport path entropy, and system state function changes through a graphical interface.

10. A high-efficiency liquid-cooled data center waste heat recovery and reuse device, used to implement the high-efficiency liquid-cooled data center waste heat recovery and reuse method according to any one of claims 1-9, characterized in that, include: Calculation module: Based on waste heat data from liquid-cooled data centers and demand data from heat users, calculate the heat source value density field and the heat sink demand urgency field; Field quantity prediction module: Taking the heat source value density field and the heat sink demand urgency field as the initial state, the module uses the preset field quantity prediction model to perform evolution prediction, and obtains the heat source value density prediction field and the heat sink demand urgency prediction field. Path planning module: Based on the heat source value density prediction field and the heat sink demand urgency prediction field, it solves the optimal transport path for waste heat in the heating network and calculates the transport path conduction entropy of the optimal transport path. Optimization control module: Based on the heat source value density prediction field, the heat sink demand urgency prediction field, the transport path conduction entropy and the system operating cost, a system status function is constructed, and the optimal control command is solved by maximizing the system status function; Correction module: Executes optimal control commands and updates the parameters of the field quantity prediction model based on the deviation between the actual and predicted values ​​of the system status function.

Images