Method and device for residual heat cascade adaptive distribution in data center phase change immersion cooling

Abstract

The application discloses a data center phase change immersion cooling waste heat stepwise self-adaptive distribution method and device, and relates to the technical field of liquid cooling waste heat distribution.The method comprises the following steps: S1, collecting stepwise waste heat distribution data in real time and performing data preprocessing; S2, determining the control parameters of a steam pump, performing steam pumping according to the control parameters of the steam pump, determining the control parameters of a power pump in the process of steam entering a plate heat exchanger, and adjusting the backflow of condensate; S3, judging the waste heat distribution proportion of three branches of a first level, a second level and a third level, and driving the three branches to distribute waste heat to a heat consumption end according to needs; and S4, continuously monitoring the running state and optimizing each algorithm.The method solves the problems that in the existing two-phase immersion cooling system, a large amount of waste heat is directly discharged, the waste heat utilization rate is low, there is a lack of stepwise distribution and dynamic self-adaptive control mechanism, energy is wasted, it is difficult to cope with load fluctuation, and the high energy efficiency and low-carbon operation requirements of a data center cannot be met.

Description

Technical Field

[0001] This invention relates to the field of liquid cooling waste heat distribution technology, specifically to a method and apparatus for adaptive cascade distribution of waste heat from phase change immersion cooling in data centers. Background Technology

[0002] With the rapid development of the digital economy, the global demand for computing power in data centers has increased significantly, and the power density of single racks has risen dramatically. The centralized deployment of high-density computing equipment has made heat dissipation a core bottleneck restricting the improvement of data center energy efficiency. To address the challenge of high power density heat dissipation, two-phase immersion cooling technology, with its advantages of high phase change heat transfer efficiency and good temperature uniformity, is gradually replacing traditional air cooling as the mainstream solution. Its core principle is to immerse the entire server or core heat-generating components in a low-boiling-point fluorinated liquid. The fluorinated liquid absorbs the heat from the chip, undergoes a phase change to generate steam, and carries away a large amount of latent heat through the steam, achieving efficient heat dissipation.

[0003] For example, the invention patent with announcement number CN119356104B announces a smart liquid cooling control method and system for data centers, belonging to the field of liquid cooling technology. It includes the following steps: S101: Obtaining optimization constraints, and determining the flow parameter range, rotation speed parameter range, and cooling parameter range based on the optimization constraints; S102: Obtaining an initial antibody population based on an artificial immune algorithm; S103: Obtaining the affinity function and affinity, and filtering multiple random antibody feature vectors based on the affinity, retaining multiple preferred antibody feature vectors; S104: Performing cloning and mutation operations on multiple preferred antibody feature vectors to obtain multiple copies of preferred antibody feature vectors and multiple mutated antibody feature vectors; S105: Establishing a new antibody population, performing iterative optimization, and adopting a mutation resistance strategy; S106: When the maximum number of iterations is reached, exporting the optimized feature vector with the highest affinity, importing it into the liquid cooling control module, and controlling the cooling unit to adjust to the corresponding parameters to save energy.

[0004] For example, invention patent CN117806151A discloses an immersion liquid cooling heat dissipation control optimization method, system, and application, including the following steps: S1: Obtain the current and real-time temperatures of the components to be cooled in the liquid-cooled cabinet from the PLC, and calculate the real-time temperature change; S2: Input the real-time temperature change, the heat power of each part, and the coolant oil temperature into a neural network, with the goal of ensuring the components to be cooled meet the control point temperature and minimizing system energy consumption, and calculate the real-time temperature and PI controller parameters for the next moment through the neural network. This immersion liquid cooling heat dissipation control optimization method, system, and application has the advantages of high efficiency, flexibility, and energy saving. This method combines neural network self-learning and adjustment of liquid-cooled cabinet control parameters, aiming to improve the cooling efficiency of data centers, reduce energy consumption, and ensure stable server operation.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0006] In existing two-phase immersion cooling systems, a significant amount of waste heat generated by servers is directly discharged into the environment, failing to achieve effective recovery and utilization through tiered and temperature-zoned processes, resulting in low waste heat utilization efficiency. Existing systems generally lack tiered waste heat distribution and dynamic adaptive control mechanisms for multi-stage heat-consuming terminals, making it impossible to flexibly adjust the waste heat distribution ratio according to actual load changes, the needs of different heat-consuming terminals, or seasonal operating conditions. This leads to singular energy utilization and difficulty in improving energy efficiency. Furthermore, existing solutions are insufficiently adaptable to complex operating scenarios in data centers, such as high heat density and large load fluctuations. Waste heat recovery and energy efficiency control are lagging behind, failing to meet the comprehensive requirements of next-generation data centers for high energy efficiency, low carbon emissions, and intelligent operation and maintenance, thus hindering the promotion and development of green and low-carbon data centers.

[0007] Therefore, in order to address the above problems, there is an urgent need for a method and device for adaptive distribution of waste heat in phase change immersion cooling of data centers. Summary of the Invention

[0008] Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides a method and apparatus for adaptive cascade distribution of waste heat from phase change immersion cooling in data centers. This solves the problems in existing two-phase immersion cooling systems where a large amount of waste heat is directly discharged, resulting in low waste heat utilization and a lack of cascade distribution and dynamic adaptive control mechanisms. These issues lead to energy waste, difficulty in coping with load fluctuations, and an inability to meet the high energy efficiency and low carbon operation requirements of data centers.

[0010] Technical solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: a data center phase change immersion cooling waste heat cascade adaptive distribution method, comprising the following steps: S1, real-time acquisition of cascade waste heat distribution data and data preprocessing of the cascade waste heat distribution data; S2, real-time determination of steam pump control parameters based on cascade waste heat distribution data, adjustment of steam pump for steam extraction based on steam pump control parameters, and determination of power pump control parameters based on cascade waste heat distribution data during steam entry into plate heat exchanger, and adjustment of condensate reflux to achieve liquid level stability and heat transfer based on power pump control parameters; S3, determination of waste heat distribution ratio of the first, second, and third-stage branches based on cascade waste heat distribution data, and driving of the three branches to distribute waste heat to the heat-using end on demand based on waste heat distribution ratio; S4, continuous monitoring of cascade waste heat distribution data, steam pump control parameters, power pump control parameters, and waste heat distribution ratio, dynamic optimization of each algorithm, and realization of closed-loop control throughout the entire process.

[0012] Furthermore, the specific process of real-time acquisition and preprocessing of cascade waste heat distribution data is as follows: Multiple types of sensors distributed in the boiling chamber, plate heat exchanger, main hot water pipe, and three branch waste heat application ends are used to collect cascade waste heat distribution data in real time. This data includes: server power consumption, chip outlet temperature, chip inlet temperature, boiling chamber pressure, boiling chamber liquid level, steam flow rate, steam pipe pressure, steam pipe temperature, condensate return flow rate, condensate inlet temperature, condensate outlet temperature, main hot water pipe temperature, first-level branch return water temperature, second-level branch return water temperature, third-level branch return water temperature, first-level branch load, second-level branch load, third-level branch load, equipment operating status, and a working fluid property data table embedded in the edge controller. The system obtains the specific heat capacity, latent heat of vaporization, and density of the coolant based on its type, and the steam density based on the steam pipeline pressure and temperature. It then uses sliding mean filtering and median filtering to denoise and smooth the original cascade waste heat distribution data, and employs the standard fraction method to remove sudden extreme values ​​and invalid data. All data acquisition points are calibrated with unified timestamps and data alignment. The system performs unit conversion and normalization on the cascade waste heat distribution data. For lost or abnormal main acquisition signals, it employs multi-source redundancy switching, Kalman filtering, and physical model compensation. Simultaneously, it performs data integrity verification through cyclic redundancy check, logical consistency check, and temporal integrity check. Finally, it establishes a cascade waste heat distribution database and stores the cascade waste heat distribution data within it.

[0013] Furthermore, based on the cascade waste heat distribution data, the specific process for determining the steam pump's control parameters in real time is as follows: While the server heat source is immersed in the boiling chamber, the high-density chip heats up, causing the low-boiling-point working fluid to undergo a boiling phase change, generating steam that flows into the main hot water pipe. During this process, cascade waste heat distribution data is acquired in real time. The difference between the chip outlet temperature and the chip inlet temperature is calculated and multiplied by the coolant's specific heat capacity to obtain the sensible heat value. The sensible heat value is then added to the coolant's latent heat of vaporization to obtain the total heat requirement per unit mass. The server power consumption is multiplied by the total heat requirement per unit mass to obtain the heat load energy input value. The steam density is multiplied by the coolant's latent heat of vaporization to obtain the working fluid density. The denominator term is calculated; the basic target flow rate is obtained by dividing the heat load energy input value by the working fluid physical denominator term; based on the sliding time window, the boiling chamber pressure when the server power consumption is less than the power consumption threshold is selected, and the average value is calculated to obtain the boiling chamber target pressure; the boiling chamber pressure difference is obtained by subtracting the boiling chamber target pressure from the boiling chamber pressure; the steam flow path pressure difference is obtained by subtracting the boiling chamber pressure from the steam pipeline pressure; the pressure correction coefficient is obtained by adding a constant to the product of the boiling chamber pressure difference and the target pressure response weight factor, and subtracting the product of the steam flow path pressure difference and the pipeline resistance correction weight factor; the basic target flow rate is multiplied by the pressure correction coefficient to obtain the steam extraction control value.

[0014] Furthermore, the specific process of adjusting the steam pump for steam extraction based on the steam pump's control parameters is as follows: The steam pump, based on the steam extraction control value, fits the characteristic curve of the steam pump's flow rate and rotational speed using a ridge regression algorithm, converting the steam extraction control value into the steam pump's speed. The controller then issues a speed adjustment command to maintain a balance between steam extraction and server heat generation. When the boiling chamber pressure is greater than or equal to the normal pressure threshold, the steam pump speed is increased to accelerate steam extraction. When the boiling chamber pressure is less than the normal pressure threshold, the steam pump speed is decreased to reduce steam extraction speed. The standard deviation of the steam extraction control value is calculated within the sliding time window. When the standard deviation of the steam extraction control value exceeds the calculated steam extraction fluctuation threshold, it is recorded and an alarm is triggered, activating a fault-tolerant extraction path.

[0015] Furthermore, the specific process of determining the control parameters of the power pump based on the stage waste heat distribution data during the steam entry into the plate heat exchanger is as follows: Steam is pumped to the plate heat exchanger. During the process of steam entering the plate heat exchanger, counter-current heat exchange and condensation with the cold source, releasing latent heat and converting into a heat transfer medium, the stage waste heat distribution data is acquired in real time. Based on a sliding time window, boiling chamber liquid levels with a standard deviation less than the liquid level fluctuation threshold are selected, and the average value is calculated to obtain the target value of the boiling chamber liquid level. The liquid level deviation is obtained by subtracting the boiling chamber liquid level from the target value, and multiplied by the liquid level deviation weighting factor to obtain the liquid level deviation response. The difference between the condensate inlet temperature and the condensate outlet temperature is calculated and compared with the cold source temperature... The condensation heat transfer power of the plate heat exchanger is obtained by multiplying the condensate reflux flow rate, coolant density, and coolant specific heat capacity. The difference between the condensate outlet temperature and the chip inlet temperature when the condensate flows back to the boiling chamber is calculated and multiplied by the coolant density to obtain the heat transfer term per unit volume. The heat transfer demand is obtained by dividing the condensation heat transfer power of the plate heat exchanger by the heat transfer term per unit volume, and then multiplied by the heat load flow rate weighting factor to obtain the heat transfer demand response. The rate of change of the liquid level in the boiling chamber is obtained by dividing the difference between two consecutive boiling chamber liquid levels by the sampling time interval, and then multiplied by the liquid level rate weighting factor to obtain the liquid level rate response. The condensate reflux control value is obtained by adding the liquid level deviation response, the heat transfer demand response, and the liquid level rate response.

[0016] Furthermore, the specific process of adjusting the reflux of condensate to achieve liquid level stability and heat transfer based on the control parameters of the power pump is as follows: Based on the reflux control value, the characteristic relationship between the pump flow rate and speed, and between the flow rate and valve opening is obtained through polynomial fitting and inverse solution. The reflux control value is then converted into the target speed of the power pump and the target valve opening, which are adjusted and fed back by the central controller to perform closed-loop control of the boiling chamber liquid level. If the standard deviation of the boiling chamber liquid level is greater than the liquid level fluctuation threshold, or the standard deviation of the reflux control value is greater than the reflux fluctuation threshold, a safety plan is triggered, the pump speed is reduced, and an alarm is issued. At the same time, the steam extraction control value and the reflux control value are stored in the cascade waste heat distribution database in real time.

[0017] Furthermore, based on the waste heat distribution data, the specific process for determining the waste heat distribution ratio of the first, second, and third-level branches is as follows: When the working fluid flows into the three-way valve in the main hot water pipe, the waste heat distribution data of the cascade is acquired in real time, and the waste heat distribution ratio of the first, second, and third-level branches is calculated: Based on the sliding time window, the maximum historical load of the i-th branch is selected as the maximum operating load of the i-th branch; the load of the i-th branch is divided by the maximum operating load of the i-th branch to obtain the load ratio of the i-th branch; the return water temperature of the i-th branch is subtracted from the temperature of the main hot water pipe to obtain the available temperature difference of the waste heat in the i-th branch; the distribution weight factor of the i-th branch, the load ratio of the i-th branch, and the available temperature difference of the waste heat in the i-th branch are multiplied to obtain the distribution weight term of the i-th branch; the distribution weight terms of the first, second, and third-level branches are added together to obtain the total distribution weight sum, and the distribution weight term of the i-th branch is divided by the total distribution weight sum to obtain the waste heat distribution ratio of the i-th branch.

[0018] Furthermore, the specific process of distributing waste heat to the heat-consuming end on demand by driving the three branches according to the waste heat distribution ratio is as follows: Based on the waste heat distribution ratio of the i-th branch, a waste heat distribution command is generated and sent to the three-way valve; for the first-level branch, the working fluid flows through the three-way valve to the ORC power generation module, driving the organic Rankine cycle to generate electricity, thus utilizing the first-level waste heat; for the second-level branch, the working fluid is diverted through the three-way valve to the absorption refrigeration module, utilizing the second-level waste heat to produce chilled water, replacing mechanical refrigeration and reducing PUE; for the tertiary branch, the working fluid is diverted through the shut-off valve to the heating module and dehumidification module, and seasonal adjustments are made based on the outdoor temperature sensor signal. Mode switching: During winter operation, the heat transfer fluid flows to the heating module to provide heating for the data center building and adjacent areas; during summer operation, the heat transfer fluid flows to the dehumidification module to regulate the humidity of the data center environment; when any of the branch equipment's operating parameters, including temperature, flow rate, pressure, and power, exceed the safety threshold, and the actual load continues to decrease, accompanied by continuous interruptions in data acquisition, the allocation weight factor is reduced to the minimum, the branch waste heat allocation is suspended, and the remaining waste heat resources are reallocated; at the same time, the waste heat allocation ratio of each branch is written into the cascade waste heat allocation database to optimize the allocation weight factor of each branch.

[0019] Furthermore, the specific process of continuously monitoring the waste heat distribution data of the cascade, the control parameters of the steam pump, the control parameters of the power pump, and the waste heat distribution ratio, and dynamically optimizing each algorithm to achieve closed-loop control of the entire process is as follows: Global aggregation and visualization monitoring of the waste heat distribution data of the cascade, steam extraction control values, condensate return control values, and the waste heat distribution ratio of each branch are established to create a full-link operational status; the central controller continuously collects the execution status of all sensors, actuators, and waste heat distribution in each branch, and synchronously maps the waste heat distribution data of the cascade to the digital twin platform to visualize equipment health, load distribution, energy efficiency trends, and alarm information; based on the full-link operational status, adjustment commands are output, and redundant equipment is linked and load reduction protection is implemented under abnormal and extreme operating conditions; through continuous comparison of historical steam extraction control values, condensate return control values, and waste heat distribution ratios of each branch, sliding window and self-learning optimization methods are used to correct each weight factor and each algorithm parameter online, achieving closed-loop and energy-efficient optimal operation of the entire process.

[0020] The second aspect of this invention provides a data center phase change immersion cooling waste heat cascade adaptive distribution device, comprising: a data acquisition and preprocessing module for real-time acquisition of cascade waste heat distribution data and preprocessing the data; a phase change cooling and condensation reflux closed-loop module for real-time determination of steam pump control parameters based on cascade waste heat distribution data, adjusting the steam pump for steam extraction based on the control parameters, and adjusting the condensate reflux to achieve liquid level stability and heat transfer during steam entry into the plate heat exchanger based on the cascade waste heat distribution data; an adaptive waste heat cascade distribution module for determining the waste heat distribution ratio of the first, second, and third-level branches based on the cascade waste heat distribution data, and driving the three branches to distribute waste heat to the heat-consuming end on demand based on the waste heat distribution ratio; and an intelligent control and data feedback optimization module for continuously monitoring the cascade waste heat distribution data, steam pump control parameters, power pump control parameters, and waste heat distribution ratio, dynamically optimizing each algorithm, and achieving full-process closed-loop control.

[0021] Beneficial effects

[0022] The present invention has the following beneficial effects:

[0023] (1) This invention realizes real-time acquisition and highly robust preprocessing of waste heat in the data center through multiple types of sensors and edge controllers, providing a high-quality data foundation for subsequent adaptive control.

[0024] (2) This invention constructs a closed-loop control method for phase change cooling and condensation reflux that combines sliding window, dynamic target setting and multi-parameter adaptive formula, which effectively ensures stable liquid level in boiling chamber, efficient cooling process and automatic adaptation to load fluctuations.

[0025] (3) This invention integrates three indicators: energy efficiency weight, load ratio and waste heat grade, to achieve optimized waste heat cascade allocation for primary, secondary and tertiary heat-consuming ends, thereby improving the comprehensive energy utilization rate and intelligent scheduling level.

[0026] (4) This invention constructs a full-link digital twin feedback and self-learning optimization mechanism to dynamically correct the control parameters and allocation weights online, forming a closed-loop adaptive optimization capability, thereby maximizing energy efficiency, protecting against abnormal redundancy and supporting intelligent operation and maintenance, and improving the waste heat utilization efficiency and overall operational safety of the data center.

[0027] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0028] Figure 1 Flowchart of a phase change immersion cooling waste heat cascade adaptive allocation method for data centers;

[0029] Figure 2 Module diagram of a phase change immersion cooling waste heat cascade adaptive distribution device for data centers;

[0030] Figure 3 Flowchart for adaptive cascade allocation of waste heat from phase change immersion cooling in data centers;

[0031] Figure 4 This is the closed-loop control logic diagram for the steam pump.

[0032] Figure 5 This is the logic diagram for the closed-loop control of the liquid level in the return circuit of the power pump.

[0033] Figure 6 Pie chart showing the allocation ratio of waste heat branches at each level;

[0034] Figure 7 This is the closed-loop control logic diagram for the three-way valve of the first-level branch.

[0035] In the diagram, 1. Boiling chamber; 2. Steam pump; 3. Plate heat exchanger; 4. Power pump; 5. Three-way valve; 6. ORC power generation module; 7. Absorption refrigeration; 8. Heating module; 9. Dehumidification module; 10. Shut-off valve. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Please see Figures 1-7 This invention provides a technical solution: a method and apparatus for adaptive cascade distribution of waste heat from phase change immersion cooling in data centers, such as... Figure 1 As shown, the process includes the following steps: S1, real-time acquisition of waste heat distribution data in the cascade, and data preprocessing of the waste heat distribution data; S2, real-time determination of the control parameters of steam pump 2 based on the waste heat distribution data, adjustment of steam pump 2 to pump steam according to the control parameters of steam pump 2, and during the process of steam entering plate heat exchanger 3, determination of the control parameters of power pump 4 based on the waste heat distribution data, and adjustment of condensate reflux to achieve liquid level stability and heat transfer according to the control parameters of power pump 4; S3, determination of the waste heat distribution ratio of the three branches of primary, secondary and tertiary stages based on the waste heat distribution data, and driving the three branches to distribute waste heat to the heat-using end on demand according to the waste heat distribution ratio; S4, continuous monitoring of waste heat distribution data in the cascade, control parameters of steam pump 2, control parameters of power pump 4 and waste heat distribution ratio, dynamic optimization of each algorithm, and realization of closed-loop control of the entire process.

[0038] Specifically, the real-time acquisition and preprocessing of waste heat distribution data in the cascade is as follows: Multiple types of sensors, including temperature sensors, pressure sensors, flow meters, level sensors, and power meters, are distributed across the boiling chamber 1, plate heat exchanger 3, main hot water pipe, and three branch waste heat application ends. These sensors, combined with an industrial automation data acquisition network, enable high-frequency, distributed, real-time acquisition of waste heat distribution data. This data includes: server power consumption, chip outlet temperature, chip inlet temperature, boiling chamber pressure, boiling chamber level, steam flow rate, steam pipe pressure, steam pipe temperature, condensate return flow rate, condensate inlet temperature, and condensate outlet temperature. The system includes the following parameters: main hot water pipe temperature, primary branch return water temperature, secondary branch return water temperature, tertiary branch return water temperature, primary branch load, secondary branch load, tertiary branch load, equipment operating status, and working fluid property data tables embedded in the edge controller. This includes the coolant specific heat capacity, latent heat of vaporization, and coolant density obtained based on the coolant type, and the steam density obtained based on the steam pipe pressure and temperature. Server power consumption is obtained through the server power metering module, chip temperature data is collected in real-time by onboard temperature sensors, and equipment operating status includes online, offline, alarm, and fault status information. Status identification and diagnosis are achieved through the edge intelligent controller. Moving average filtering and median filtering are used to denoise and smooth the original cascade waste heat allocation data to eliminate short-term noise and improve signal smoothness. The standard score method is combined to remove sudden extreme values ​​and invalid data. All acquisition points are calibrated with unified timestamps and data alignment. Unit conversion and normalization are performed on the cascade waste heat allocation data. For the loss or anomaly of the main acquisition signal, multi-source redundancy switching, Kalman filtering, and physical model compensation are employed. Multi-source redundancy switching refers to activating a backup data source when the main acquisition signal is abnormal. Simultaneously, data integrity is verified through cyclic redundancy check, logical consistency check, and temporal integrity detection. A cascade waste heat allocation database is established, and the cascade waste heat allocation data is stored in this database, supporting high-concurrency data writing and fast historical querying, ensuring a data foundation for subsequent control and algorithm optimization.

[0039] like Figure 3The diagram shows the adaptive distribution flow of waste heat from phase change immersion cooling in a data center. It illustrates the overall physical process of the adaptive distribution system, including the following key steps and functions: The high heat density generated during server operation is absorbed by the low-boiling-point working fluid in boiling chamber 1, triggering phase change boiling and forming steam. Steam pump 2 adaptively adjusts its extraction rate based on the waste heat distribution data, sending the steam into plate heat exchanger 3. Here, it exchanges heat counter-currently with the cold source working fluid, condenses, and releases heat, becoming a hot working fluid. Power pump 4 uses closed-loop level control to precisely return the condensate to boiling chamber 1, ensuring stable level and fluid circulation. The hot working fluid from the plate heat exchanger outlet flows into the main hot water pipe, and the distribution ratio is adjusted in real time through a three-way valve 5. It enters the ORC power generation module 6 through a primary branch, undergoes absorption cooling 7 through a secondary branch, and provides heating and dehumidification through a tertiary branch, distributing the heat to multiple waste heat consumption terminals as needed. The third-level branch circuit achieves seasonal switching via shut-off valve 10, connecting to heating module 8 and dehumidification module 9 to meet the needs of winter heating and summer dehumidification. The entire process is achieved through closed-loop adaptive regulation by the central controller and multi-source sensors, optimizing the parameters and allocation ratios of each link in real time. This enables server heat dissipation, multi-level utilization of waste heat, energy efficiency optimization, and intelligent switching, thereby improving the waste heat utilization rate and low-carbon operation capability of the data center.

[0040] This implementation plan achieves real-time acquisition of key physical quantities across multiple dimensions throughout the entire phase change immersion cooling process of the data center, improving the comprehensiveness and timeliness of operational monitoring. Employing various advanced data preprocessing technologies enhances the accuracy, completeness, and robustness of the data, effectively eliminating the impact of noise, outliers, and signal loss. This provides a high-quality, traceable data foundation for subsequent waste heat distribution optimization and adaptive control, ensuring the realization of energy efficiency improvements, operational safety, and intelligent management.

[0041] Specifically, the process of determining the control parameters of steam pump 2 in real time based on the waste heat distribution data of the cascade is as follows: While the server heat source is immersed in the boiling chamber 1, the high-density chip heats up, causing the low-boiling-point working fluid to undergo a boiling phase change, generating steam that flows into the main hot water pipe. During this process, the waste heat distribution data of the cascade is acquired in real time. The difference between the chip outlet temperature and the chip inlet temperature is calculated. This chip temperature difference reflects the driving force of the heat generated by the server. This difference is multiplied by the specific heat capacity of the coolant to obtain the sensible heat value. The sensible heat value is then added to the latent heat of vaporization of the coolant working fluid to obtain the total heat required per unit mass. Finally, the server power consumption is multiplied by the total heat required per unit mass to obtain the heat load. Energy input value; multiplying the steam density by the latent heat of vaporization of the working fluid to obtain the working fluid physical denominator; dividing the heat load energy input value by the working fluid physical denominator to obtain the basic target flow rate; based on a sliding time window, selecting the boiling chamber pressure when the server power consumption is less than the power consumption threshold, and calculating the average value to obtain the boiling chamber target pressure; subtracting the boiling chamber target pressure from the boiling chamber pressure to obtain the boiling chamber pressure difference; subtracting the boiling chamber pressure from the steam pipeline pressure to obtain the steam flow path pressure difference; the boiling chamber pressure difference reflects the deviation between the current actual pressure and the expected working pressure, and the steam flow path pressure difference reflects the change in the flow resistance of steam through the pipeline and components. Adding a constant to the product of the boiling chamber pressure difference and the target pressure response weighting factor, and subtracting the product of the steam flow path pressure difference and the pipeline resistance correction weighting factor, to obtain the pressure correction coefficient; each weighting factor is adaptively adjusted using historical data; multiplying the basic target flow rate by the pressure correction coefficient to obtain the steam extraction control value. The steam extraction control value is comprehensively calculated by the controller, and the multi-channel integrated PID controller generates specific speed control commands for steam pump 2, driving steam pump 2 to adjust in real time, ensuring that the steam extraction flow rate dynamically matches the server heat load, and achieving optimal synergy between boiling chamber pressure, liquid level and waste heat utilization.

[0042] The specific formula for the steam extraction control value is as follows:

[0043] ;

[0044] In the formula, This represents the steam extraction control value, used for real-time quantification and dynamic adjustment of the target flow rate of steam pump 2. This allows steam pump 2 to adaptively match the steam generation and extraction rates based on changes in server heat generation, working fluid properties, temperature, pressure, and pipeline conditions, ensuring stable boiling chamber pressure, efficient heat dissipation, and optimal energy efficiency. This indicates the server's power consumption, reflecting the current heat intensity of the server; This indicates the specific heat capacity of the coolant, representing the amount of heat required to raise the temperature of a unit mass of the working fluid by one degree Celsius. Indicates the chip outlet temperature; Indicates the chip inlet temperature; This indicates the temperature difference between the chip's outlet and inlet, reflecting the temperature rise of the coolant before and after passing through the server chip, and demonstrating the heat dissipation status. It represents the latent heat of vaporization of the working fluid in the coolant, reflecting the latent heat absorbed per kilogram when the working fluid changes from a liquid to a gaseous state; This represents the heat load energy input value, which indicates the total heat energy that the server generates per unit time that can be used for steaming, and directly determines how much steam can be theoretically generated. This indicates the density of steam, reflecting the mass of steam per unit volume. This represents the physical denominator of the working fluid, converting the total thermal energy into the theoretical steam volume flow rate; Indicates the pressure in the boiling chamber; Indicates the target pressure in the boiling chamber; Indicates the pressure difference in the steam flow path; It represents the pressure correction coefficient, which comprehensively corrects the basic target flow rate and incorporates dynamic factors such as pressure fluctuations, target pressure, and flow path resistance into flow regulation to achieve adaptive adjustment and linkage matching with operating conditions. The target pressure response weighting factor is initially set to 0.05. Then, within each sliding time window, the actual steam pump flow rate and steam extraction control value are continuously collected and recorded. The mean square error of the actual steam pump flow rate and steam extraction control value within this window is calculated. The optimal target pressure response weighting factor is obtained by minimizing the error between the actual steam pump flow rate and the steam extraction control value. The value range is between 0.01 and 0.2. The pipeline resistance correction weight factor is initially set to 0.01. The pressure difference in the steam flow path and the actual steam pump flow rate fluctuations are continuously monitored, and these two indicators are included in a sliding time window. At the end of each window period, the relationship between the change in the steam flow path pressure difference and the fluctuation in the actual steam pump flow rate is analyzed. If the actual steam pump flow rate fluctuation is aggravated by a large change in the steam flow path pressure difference, the pipeline resistance correction weight factor is increased to enhance the suppression capability. If the change in the steam flow path pressure difference has too little impact on the fluctuation in the actual steam pump flow rate, the pipeline resistance correction weight factor is decreased. Through periodic fine-tuning, the pipeline resistance correction weight factor is converged to the optimal range that can effectively suppress the abnormal fluctuations in the actual steam pump flow rate caused by changes in the steam flow path pressure difference, thus obtaining the optimal pipeline resistance correction weight factor, with a value range between 0.001 and 0.1.

[0045] like Figure 4The diagram shows the closed-loop control logic of the steam pump. It illustrates the closed-loop control logic of the intelligent regulation of steam pump 2 in a data center phase change immersion cooling waste heat cascade adaptive distribution system. The specific process is as follows: First, the steam flow sensor and boiling chamber pressure sensor collect the actual flow rate and pressure of the steam output from boiling chamber 1 in real time, serving as key input parameters. Simultaneously, the server heat load and operating status are collected synchronously to provide basic heat source data for the regulation strategy. Next, this multi-source data is sent to the steam pump regulation parameter calculation module. Based on the current operating conditions, physical properties, load, and pressure difference input, combined with the sliding window method and adaptive algorithm, the optimal steam extraction regulation value is calculated. The steam extraction regulation value not only dynamically matches the server heat load and the actual working fluid evaporation rate but also takes into account the fluctuation of boiling chamber pressure and pipeline resistance factors, ensuring stable pressure and optimal energy efficiency. The PID controller comprehensively considers the proportional, integral, and derivative terms based on the steam extraction regulation value, outputting the optimal speed control command for steam pump 2 to achieve real-time, precise, and gradual flow regulation. Steam pump 2 adjusts its speed according to the controller output, dynamically adapting to the actual steam generation rate and pressure changes, and promptly extracts excess steam from boiling chamber 1 to prevent excessive pressure, working fluid entrainment, and cooling lag. Boiling chamber 1, as the core heat source area, ensures the continuity and efficiency of phase change heat dissipation through stable liquid level and pressure. This closed-loop intelligent control mechanism can achieve adaptive balance in steam extraction under conditions of significant fluctuations in data center load and dynamic changes in thermal conditions, maximizing waste heat utilization efficiency and operational safety. It also lays a solid foundation for subsequent energy recovery and waste heat cascade distribution in plate heat exchanger 3.

[0046] In this implementation plan, by collecting data on the cascade waste heat distribution and analyzing the physical quantities throughout the process, high-precision, real-time adaptive determination of the steam pump control parameters is achieved. Furthermore, it accurately reflects changes in server load, boiling chamber pressure fluctuations, and working fluid state, effectively improving the adjustment response speed and steady-state accuracy of the steam extraction flow rate. By combining the basic target flow rate with multi-factor pressure correction coefficients, the output of steam pump 2 can be flexibly adjusted according to actual operating conditions, ensuring the stability of the phase change process in boiling chamber 1, efficient waste heat recovery, and optimized energy utilization. This enhances the dynamic adaptability and operational safety of data center cooling, providing a solid foundation for subsequent end-to-end energy efficiency control and intelligent waste heat distribution.

[0047] Specifically, the process of adjusting the steam pump 2 for steam extraction according to the control parameters of steam pump 2 is as follows: Steam pump 2 fits the characteristic curve of steam pump flow rate and speed using a ridge regression algorithm based on the steam extraction control value, converting the steam extraction control value into the pump speed of steam pump 2. Ridge regression is a linear regression algorithm with a regularization term, which can robustly fit the relationship between flow rate and speed when multicollinearity exists in the data, achieving an accurate mapping between the target flow rate and the actual pump speed. The controller then issues a speed adjustment command to maintain the balance between steam extraction and server heat generation. When the boiling chamber pressure is greater than or equal to the normal pressure threshold, the speed of steam pump 2 is increased to accelerate the steam extraction speed. When the boiling chamber pressure is less than the normal pressure threshold, the speed of steam pump 2 is decreased to reduce the steam extraction speed. The adjustment logic is implemented through pressure sensor feedback and PID controller closed loop to ensure that steam pump 2 dynamically adjusts its operating state in response to boiling chamber pressure fluctuations. Within a sliding time window, the standard deviation of the steam extraction control value is calculated. The standard deviation is a statistical indicator measuring the degree of data fluctuation and can reflect the stability of the steam extraction flow rate in real time. When the standard deviation of the steam extraction control value exceeds the calculated steam extraction fluctuation threshold, it is recorded and an alarm is triggered, activating the fault-tolerant extraction path. The fault-tolerant extraction path uses redundant pumps. When the main extraction path experiences severe fluctuations or fails, it switches to the backup extraction equipment to ensure cooling and pressure safety. Alarm information is pushed through multiple channels, including on-site alarms and the monitoring platform, achieving timely response to anomalies and operational redundancy protection.

[0048] This implementation scheme combines multi-dimensional physical construction of steam pump control parameters, ridge regression algorithm fitting, and intelligent closed-loop control to achieve high-precision, real-time adaptive adjustment of the steam pump speed, dynamically matching the server's heat generation and steam extraction requirements of boiling chamber 1. It features sensitive pressure feedback, fluctuation warning, and fault-tolerant redundancy mechanisms, which not only improve the operational safety and stability of steam extraction but also effectively prevent risks caused by abnormal operating conditions such as excessively high or low pressure or extraction path failure, ensuring the continuous and reliable operation of efficient waste heat recovery and cooling in the data center.

[0049] Specifically, the process of determining the control parameters of the power pump 4 based on the stage waste heat distribution data during the steam entry into the plate heat exchanger 3 is as follows: Steam is pumped to the plate heat exchanger 3. During the process of steam entering the plate heat exchanger 3, counter-current heat exchange and condensation with the cold source, releasing latent heat and converting into a heat transfer medium, the stage waste heat distribution data is acquired in real time. Based on a sliding time window, boiling chamber liquid levels with a standard deviation less than the liquid level fluctuation threshold are selected, and the average value is calculated to obtain the target value of the boiling chamber liquid level. The liquid level deviation is obtained by subtracting the boiling chamber liquid level from the target value, and multiplied by the liquid level deviation weighting factor to obtain the liquid level deviation response. The liquid level deviation weighting factor is used to balance the influence of liquid level regulation on reflux control. The difference between the condensate inlet temperature and the condensate outlet temperature is calculated and multiplied by the condensate reflux flow rate, coolant density, and coolant specific heat capacity to obtain the condensation heat exchange power of the plate heat exchanger, reflecting the total heat released by condensation per unit time, which is the basis for the reflux regulation of the power pump 4. The condensate outlet temperature is calculated... The temperature difference between the temperature at the inlet of the chip and the temperature at which the condensate flows back to boiling chamber 1 is multiplied by the density of the coolant to obtain a heat term per unit volume of temperature difference. This term is used to measure the actual heat exchange capacity of the reflux liquid and the heat level brought back to boiling chamber 1. The heat exchange demand is obtained by dividing the condensation heat exchange power of the plate heat exchanger by the heat term per unit volume of temperature difference. This heat demand response is then multiplied by the heat load flow rate weighting factor to obtain the heat demand response. The heat demand response reflects the heat recovery target that the power pump 4 should meet under the current condensation and actual reflux conditions. The heat load flow rate weighting factor is used to adjust the reflux priority under different heat load scenarios. The difference between two consecutive boiling chamber liquid levels is calculated and divided by the sampling time interval to obtain the liquid level change rate. This rate is then multiplied by the liquid level rate weighting factor to obtain the liquid level rate response. The liquid level rate response reflects the dynamic change trend of the liquid level and helps to identify the reflux pump regulation lag or liquid level runaway problem in a timely manner. The weighting factor can be adaptively adjusted. The liquid level deviation response, the heat exchange demand response, and the liquid level rate response are added together to obtain the condensate reflux control value. The condensate reflux control value is used as the target flow rate of the power pump, which is directly used to drive the variable frequency power pump 4 and the intelligent valve to regulate the condensate reflux rate. This achieves coordinated optimization of plate heat exchange and boiling chamber liquid level, ensuring heat exchange efficiency and liquid level stability, and supporting efficient waste heat recovery and cooling safety in the data center.

[0050] The specific formula for the condensate reflux control value is as follows:

[0051] ;

[0052] In the formula, This represents the condensate reflux control value, used to quantify the target flow rate of the condensate power pump 4 in real time. It integrates the three factors of liquid level closed-loop control, actual heat load dynamic response, and the influence of liquid level change rate, so that the pump flow rate can accurately maintain the optimal liquid level of boiling chamber 1, and can adaptively cope with heat exchange load and operating disturbances, ensuring the safety, heat exchange efficiency and dynamic stability of boiling chamber 1. Indicates the target value of the liquid level in the boiling chamber; Indicates the liquid level in the boiling chamber; This represents the liquid level deviation response, reflecting the deviation between the current actual liquid level in boiling chamber 1 and the target liquid level. The larger the deviation, the greater the pump flow adjustment is required to quickly restore the liquid level balance. This indicates the condensing heat exchange power of the plate heat exchanger. Indicates the density of the coolant; Indicates the condensate outlet temperature; This indicates the chip inlet temperature when the condensate flows back to boiling chamber 1. It represents the heat exchange demand response, which is the volumetric flow rate of condensate required to maintain energy balance under the current heat exchange load and temperature conditions. The denominator is the heat that can be carried away by a unit volume of condensate, and the numerator is the actual heat that needs to be carried away. The ratio of the two is the theoretically required flow rate, which is used to compensate for liquid level fluctuations caused by dynamic changes in heat load. Indicates the rate of change of liquid level; It represents the liquid level rate response, reflecting the timely adjustment of pump flow when the liquid level is changing rapidly, preventing liquid level overshoot and lag, and achieving more sensitive and stable liquid level control; The liquid level deviation weighting factor is initially set to 0.5. Within each sliding time window, liquid level deviation and actual condensate reflux control value data are continuously collected and recorded. The mean square error of the liquid level deviation change and the actual condensate reflux control value adjustment response within the window is calculated. The optimization objective is to minimize the control error of the liquid level deviation on the condensate reflux control value. The liquid level deviation weighting factor is fine-tuned. When the liquid level deviation response is found to be lagging and the liquid level is difficult to return to the target in time, the liquid level deviation weighting factor is increased to improve sensitivity. If there is a sharp fluctuation in liquid level, the liquid level deviation weighting factor is decreased to suppress over-adjustment. Through adaptive adjustment via the sliding window, it eventually converges to the optimal range of liquid level deviation response to obtain the optimal liquid level deviation weighting factor, with a value range between 0.1 and 2. The heat load flow rate weighting factor is initially set to 1.0. Data on heat exchange demand and actual condensate reflux control values ​​are continuously collected. Within each sliding window period, the coupling relationship between heat exchange demand and actual condensate reflux control values ​​is analyzed, and the mean square error is calculated. The optimization objective is to minimize the impact of changes in heat exchange demand on pump flow control error. The heat load flow rate weighting factor is fine-tuned. If insufficient liquid level regulation response is found under sudden changes in heat exchange load, the heat load flow rate weighting factor is increased to enhance the influence of the heat load term. If flow fluctuations intensify, the heat load flow rate weighting factor is decreased to obtain the optimal heat load flow rate weighting factor, with a value range between 0.5 and 3.0. The liquid level rate weighting factor is initially set to 0.1. It continuously collects data on the rate of change of liquid level and the actual condensate reflux control value, and statistically analyzes the liquid level fluctuation range within the window. With the goal of suppressing flow overshoot caused by sudden liquid level changes and achieving optimal dynamic response, the liquid level rate weighting factor is fine-tuned. When the liquid level rate change has insufficient impact on flow regulation, the liquid level rate weighting factor is increased; if the fluctuation is aggravated, the liquid level rate weighting factor is decreased. Through dynamic adjustment within a sliding window, it eventually converges to the optimal liquid level rate weighting factor range that suppresses abnormal liquid level rates and improves stability. The value range is between 0.01 and 0.5.

[0053] like Figure 5 The diagram shows the closed-loop control logic for the liquid level in the power pump reflux loop. First, the liquid level sensor collects the actual liquid level of the working fluid in boiling chamber 1 in real time and uploads it to the control system. The comparator compares the collected actual liquid level with the target liquid level in the boiling chamber in real time, calculating the liquid level deviation. The liquid level deviation is sent to a multi-channel integrated PID controller, which outputs the optimal condensate reflux control command based on the magnitude and trend of the deviation and the response to heat exchange requirements. Subsequently, the power pump 4 adjusts its speed and flow rate according to the output signal of the PID controller, pumping the condensate reflux liquid into boiling chamber 1. In this way, the working fluid level in boiling chamber 1 can be dynamically adjusted to ensure that the liquid level remains stable within the optimal operating range. The entire process achieves closed-loop control of the boiling chamber liquid level, effectively ensuring safe evaporation, phase change heat transfer efficiency, and cooling stability, supporting data centers to achieve optimal energy efficiency and safe operation under complex conditions of high heat density and load fluctuations.

[0054] In this implementation scheme, by combining multi-point liquid level signal statistics, real-time heat exchange power of plate heat exchanger 3, and multi-dimensional parameters of liquid level dynamic change rate, a step-by-step weighted and adaptive algorithm is adopted to achieve high-precision, dynamic, and adaptive control of the power pump return flow rate. This not only improves the stability of the boiling chamber liquid level and the heat recovery efficiency of plate heat exchanger 3, but also flexibly responds to load and liquid level changes according to actual operating conditions, achieving synergistic optimization of heat exchange control and liquid level management. This effectively enhances the data center cooling system's adaptability to heat load fluctuations and the energy efficiency of waste heat recovery, improving safety, operational stability, and intelligence.

[0055] Specifically, the process of adjusting the reflux of condensate according to the control parameters of power pump 4 to achieve liquid level stability and heat transfer is as follows: Based on the reflux control value, the characteristic relationships between pump flow rate and speed, and flow rate and valve opening are obtained through polynomial fitting and inverse kinematics. Polynomial fitting refers to using historical operating data and flow characteristic curves of power pump 4, and expressing the nonlinear relationship between flow rate / speed and flow rate / valve opening using a cubic polynomial through least squares statistical methods. Combined with the inverse kinematics algorithm, the optimal pump speed and valve opening are calculated under a given target flow rate. The liquid reflux control value is converted into the target speed of power pump 4 and the target valve opening, which is adjusted and fed back by the central controller for closed-loop regulation of the boiling chamber liquid level. If the standard deviation of the boiling chamber liquid level is greater than the liquid level fluctuation threshold, or the standard deviation of the condensate reflux control value is greater than the reflux fluctuation threshold, the safety plan is triggered, the pump speed is reduced, and an alarm is issued. The standard deviation is a statistical indicator reflecting the severity of liquid level fluctuations. Exceeding the reflux fluctuation threshold indicates a strong disturbance, triggering the safety protection mechanism. This reduces the flow output by reducing the pump speed and simultaneously activates the local alarm, notifying maintenance personnel to investigate promptly. At the same time, the steam extraction control value and the condensate reflux control value are stored in the cascade waste heat distribution database in real time.

[0056] In this implementation scheme, high-precision dynamic control of condensate reflux and automatic closed-loop regulation of the boiling chamber level are achieved by mapping the condensate reflux control value to the power pump speed and valve opening. Employing polynomial fitting and inverse kinematics algorithms, the scheme adapts to the nonlinear characteristics of the pump and valves, ensuring optimal flow regulation under various operating conditions. Real-time monitoring of liquid level and reflux fluctuations, triggering safety contingency plans and alarms, enhances anomaly detection and self-protection capabilities. Full-process archiving of control data provides solid support for intelligent operation and maintenance, energy efficiency analysis, and self-learning optimization.

[0057] Specifically, based on the waste heat distribution data, the process for determining the waste heat distribution ratio of the first, second, and third-level branches is as follows: When the working fluid flows into the three-way valve 5 in the main hot water pipe, the waste heat distribution data of the cascade is acquired in real time, and the waste heat distribution ratio of the first, second, and third-level branches is calculated: Based on the sliding time window, the maximum historical load of the i-th branch is selected as the maximum operating load of the i-th branch. The sliding time window refers to dynamically sliding within a historical time period of a set length, and real-time statistics are collected on the extreme value changes of the load of each branch, which can filter out occasional spikes and reflect the true maximum load capacity of the equipment; The load ratio of the i-th branch is obtained by dividing the load of the i-th branch by the maximum operating load of the i-th branch. The load ratio reflects the ratio of the real-time demand of the current heat-consuming end to the ultimate capacity of the equipment, ensuring that the distribution strategy can dynamically adjust. To adapt to load fluctuations, the available waste heat temperature difference of the i-th branch is obtained by subtracting the return water temperature of the i-th branch from the main hot water pipe temperature. The main hot water pipe temperature is the real-time temperature during the distribution of the heat transfer fluid, and the return water temperature of each branch reflects the actual working fluid temperature after utilization at the heat-using end. The larger the temperature difference, the stronger the absorption capacity of the branch for high-grade waste heat. The allocation weight factor of the i-th branch, the load ratio of the i-th branch, and the available waste heat temperature difference of the i-th branch are multiplied to obtain the allocation weight term of the i-th branch. The branch allocation weight factor is adaptively set according to the strategy, energy efficiency optimization, and external economic signals such as electricity price and carbon price, and is used to adjust the priority and energy efficiency target of each branch. The allocation weight terms of the first-level branch, the second-level branch, and the third-level branch are added together to obtain the total allocation weight sum. The waste heat allocation ratio of the i-th branch is obtained by dividing the allocation weight term of the i-th branch by the total allocation weight sum. The branch waste heat distribution ratio is directly used as the control target of the three-way valve 5 distribution actuator to realize the on-demand, dynamic and optimized distribution of waste heat, improve the efficiency and flexibility of waste heat utilization, and support the energy efficiency improvement and multi-objective adaptive operation of the data center.

[0058] The specific formula for the waste heat distribution ratio of the i-th branch is as follows:

[0059] ;

[0060] In the formula, This represents the waste heat distribution ratio of the i-th branch, used to dynamically calculate the target distribution ratio of the data center's waste heat in each branch of the three-way distribution valve, so that the waste heat energy can be optimally distributed according to actual heat demand, energy efficiency priority and available temperature difference, thereby improving overall energy efficiency and energy utilization. Indicates the load of the i-th branch; This represents the maximum operating load of the i-th branch; This represents the load ratio of the i-th branch, ensuring that the allocation ratio changes synchronously with the current demand, with those with higher demand receiving more surplus heat. Indicates the temperature of the main hot water pipe; Indicates the return water temperature of the i-th branch; This represents the available temperature difference of the waste heat in the i-th branch, reflecting the amount of available heat in each branch. Higher grades are allocated with higher priority. The weighting factor for the i-th branch is determined by adjusting the weights based on the historical and real-time energy efficiency indicators, economic benefits, real-time load and operating status data of the i-th branch. The optimal weighting factor for the i-th branch is obtained by fitting the data with the goal of minimizing energy loss and maximizing economic benefits. The value range is between 0.5 and 5. This represents the total allocation weight, ensuring that the sum of the three allocation ratios is 1, thus achieving the globally optimal allocation.

[0061] In this embodiment, Table 1 is a data table of waste heat distribution ratios for the branch circuits. It details the distribution weight factors, loads, maximum operating loads, main hot water pipe temperatures, return water temperatures, and waste heat distribution ratios for the three waste heat distribution branches: Level 1, Level 2, and Level 3. Specifically, the Level 1 branch has a distribution weight factor of 1.2, a load of 75, a maximum operating load of 100, a main hot water pipe temperature of 65, a return water temperature of 54, and a waste heat distribution ratio of 30.96%. The Level 2 branch has a distribution weight factor of 1.0, a load of 45, a maximum operating load of 60, a main hot water pipe temperature of 65, a return water temperature of 49, and a waste heat distribution ratio of 37.52%. The Level 3 branch has a distribution weight factor of 0.8, a load of 30, a maximum operating load of 50, a main hot water pipe temperature of 65, a return water temperature of 44, and a waste heat distribution ratio of 31.52%.

[0062] Table 1. Data on the distribution ratio of waste heat from branch circuits

[0063]

[0064] like Figure 6 The image shows a pie chart illustrating the waste heat distribution ratios for each level of waste heat distribution branch. It displays the waste heat distribution ratios obtained by the primary, secondary, and tertiary waste heat distribution branches under the current operating conditions. Specifically, the primary branch receives 30.96%, the secondary branch receives 37.52%, and the tertiary branch receives 31.52%. (Based on Table 1 and...) Figure 6 It can be seen that in this allocation, the secondary branch received the largest proportion of waste heat, indicating that the current demand for refrigeration and medium-grade heating is higher and greater. The allocation proportions for the primary and tertiary branches are similar, both close to one-third, reflecting that the waste heat allocation is adjusted according to multi-objective optimization to achieve dynamic and balanced utilization of heat.

[0065] This implementation scheme employs a multi-parameter coupling method involving sliding window statistics, load ratio, available temperature difference, and allocation weighting factors to achieve dynamic adaptive optimization of the allocation ratio of waste heat branches at each level. It can reflect changes in heat demand, equipment limits, and energy efficiency targets in real time, flexibly adjusting waste heat allocation strategies and improving the accuracy and on-demand nature of allocation. Through adaptive adjustment of weighting factors and real-time data-driven approaches, it can effectively balance the priorities of multiple heat-consuming ends, maximizing waste heat recovery and comprehensive energy utilization efficiency, and enhancing the intelligent control and operational safety of data center waste heat allocation.

[0066] Specifically, the process of distributing waste heat to the heat-consuming end on demand through three branches based on the waste heat distribution ratio is as follows: A waste heat distribution command is generated and sent to the three-way valve 5 according to the waste heat distribution ratio of the i-th branch. For the first-level branch, the working fluid flows through the three-way valve 5 to the ORC power generation module 6, driving the organic Rankine cycle to generate electricity for primary waste heat utilization. The ORC module can efficiently utilize high-grade waste heat, driving a turbine to generate electricity through the expansion of the working fluid. The generated energy can be fed back to the data center microgrid and grid-connected output, enhancing the energy utilization value of waste heat. For the second-level branch, the working fluid is diverted through the three-way valve 5 to the absorption refrigeration module 7, using secondary waste heat to produce chilled water, replacing mechanical refrigeration and reducing PUE. The absorption refrigeration module 7, such as a lithium bromide unit, can utilize medium-grade waste heat to produce chilled water for data center air conditioning systems and equipment room cooling, achieving cascaded energy utilization. PUE is the data center energy efficiency ratio, and the waste heat distribution scheme can significantly reduce energy consumption. For the three-level branch circuits, the heat transfer fluid is diverted to the heating module 8 and dehumidification module 9 via shut-off valve 10. Seasonal mode switching is performed based on outdoor temperature sensor signals: during winter operation, the heat transfer fluid flows to the heating module 8 to provide heating for the data center building and adjacent areas; during summer operation, it flows to the dehumidification module 9 to regulate the humidity of the data center environment. Branch circuit switching is controlled by the central controller via shut-off valve 10, intelligently adjusting according to ambient temperature and load demand to ensure the continuity and comfort of heating and dehumidification at the terminal. When any of the branch circuit equipment's operating parameters, including temperature, flow rate, pressure, and power, exceeds the safety threshold, and the actual load continues to decrease, accompanied by continuous interruptions in data acquisition, the allocation weight factor is reduced to the minimum, the branch circuit's waste heat allocation is suspended, and the remaining waste heat resources are reallocated. This prevents abnormal branches from continuing to consume heat transfer fluid and causing energy efficiency losses. By adjusting the weight parameters within the allocation optimizer, the allocation ratio of other healthy branches is dynamically increased, ensuring stable operation of the entire process. At the same time, the waste heat allocation ratio of each branch is written into the cascade waste heat allocation database. Combined with the real-time feedback mechanism and self-learning optimization algorithm, the allocation weight factor of each branch is optimized to improve the accuracy and responsiveness of subsequent allocation, forming a complete closed loop of allocation, execution and feedback.

[0067] like Figure 7The diagram shows the closed-loop control logic of the three-way valve for a primary branch. First, the load acquisition module monitors the overall heat load of the branch in real time, while simultaneously acquiring the return temperature and key operating parameters of the primary branch's energy efficiency, such as power generation efficiency, providing dynamic and comprehensive data support for subsequent allocation decisions. The waste heat allocation optimizer is input, and based on real-time load demand, branch return temperature, energy efficiency status, and other external constraints, it dynamically calculates the optimal waste heat allocation ratio for the branch using an adaptive allocation algorithm. The waste heat allocation optimizer is the decision-making center of the entire allocation process, ensuring the synergy of multiple objectives: energy efficiency, economy, and load demand. The optimizer's allocation results are sent to the allocation target comparator, which compares them in real time with the actual allocation ratio of the branch, generating a deviation signal. At this time, the ORC inlet temperature sensor continuously acquires the actual working fluid temperature, providing feedback support for comparison and subsequent adjustment. The deviation signal is processed by a multi-channel integrated PID controller, which generates precise adjustment commands to control the opening of the three-way valve 5, diverting the working fluid from the main hot water pipe to the heat-consuming end according to the allocation ratio. Among them, the ORC evaporator, as a typical representative of the heat end of the primary branch, receives high-grade heat transfer fluid for organic Rankine cycle power generation, achieving efficient recovery and utilization of waste heat. The overall process constitutes a closed-loop intelligent allocation system with data-driven, adaptive optimization, and real-time feedback at its core. Under the conditions of heat load fluctuations, operating condition changes, and multi-objective constraints, it can always achieve dynamic optimization of waste heat allocation and operational safety, providing a solid guarantee for the green and efficient operation of data centers.

[0068] This implementation plan achieves precise allocation and coordinated utilization of waste heat resources at all levels to different heat-consuming ends through a combination of on-demand dynamic allocation, intelligent execution, and closed-loop feedback. The allocation commands are linked with three-way valve 5 and shut-off valve 10, efficiently driving diverse energy consumption scenarios such as power generation, cooling, heating, and dehumidification, comprehensively improving the utilization efficiency of waste heat resources. It features an abnormal branch identification and allocation ratio adjustment mechanism, enabling timely suspension of abnormal branches to ensure efficient waste heat absorption and safe operation. Simultaneously, continuous archiving of allocation data and weight self-learning optimization provide a solid foundation for energy efficiency improvement and intelligent decision-making, enhancing the data center's energy efficiency, intelligence level, and operational resilience.

[0069] Specifically, the process of continuously monitoring the waste heat distribution data of the cascade, the control parameters of steam pump 2, the control parameters of power pump 4, and the waste heat distribution ratio, and dynamically optimizing each algorithm to achieve closed-loop control of the entire process is as follows: Global aggregation and visualization monitoring of the waste heat distribution data of the cascade, steam extraction control values, condensate return control values, and the waste heat distribution ratio of each branch are established to create a full-link operational status; the central controller continuously collects the execution status of all sensors, actuators, and waste heat distribution in each branch, and synchronously maps the waste heat distribution data of the cascade to the digital twin platform for monitoring equipment health, load distribution, and other aspects. The visualization of energy efficiency trends and alarm information is achieved through a digital twin platform that uses actual physical parameters to build a virtual simulation model. This platform enables 3D visualization and real-time interaction of operating parameters, energy efficiency trends, and fault alarm status, helping maintenance personnel to fully grasp the operational dynamics. Based on the overall operational status, adjustment commands are output, triggering redundant equipment and implementing load reduction protection under abnormal and extreme conditions. Abnormal conditions include sensor failures, exceeding limits for critical flow, pressure, and temperature. The central controller can schedule backup equipment, switch cooling paths, and ensure core heat dissipation and safety through load reduction and bypass discharge. By continuously comparing historical steam extraction control values, condensate return control values, and waste heat distribution ratios of each branch, a sliding window and self-learning optimization method are used to correct weight factors and algorithm parameters online, achieving closed-loop and energy-efficient operation throughout the entire process. The sliding window method monitors the statistical trends and abnormal fluctuations of each parameter in real time, while the self-learning optimization method combines machine learning, historical operating condition playback, and parameter feedback to adaptively and dynamically adjust weight factors and distribution strategies, continuously optimizing the waste heat distribution efficiency of each branch. It has achieved a closed-loop, self-evolving control architecture that integrates data acquisition, intelligent computing, precise execution, digital feedback, and continuous optimization, effectively improving the intelligence level and energy efficiency limits of data center cooling.

[0070] In this implementation plan, by continuously aggregating and monitoring multi-dimensional data on cascade waste heat distribution, steam pump 2 and power pump 4 regulation, and distribution ratios, intelligent visualization and dynamic operational status awareness of the entire process are achieved. A digital twin platform is used to map and analyze equipment health, load distribution, and energy efficiency trends in real time, improving transparency and operational intelligence. The central controller can automatically link redundant equipment and implement load reduction protection under abnormal and extreme conditions, ensuring safe and stable operation. Combining sliding window and self-learning optimization methods, the allocation weights and control parameters can be continuously corrected online and adaptively evolved to maximize energy efficiency and waste heat recovery capabilities, achieving a closed-loop process and optimal energy-saving operation for data center cooling.

[0071] Reference Figure 2As shown, the second aspect of the present invention provides a data center phase change immersion cooling waste heat cascade adaptive distribution device, applied to the aforementioned data center phase change immersion cooling waste heat cascade adaptive distribution method, comprising: a data acquisition and preprocessing module, used to acquire cascade waste heat distribution data in real time and preprocess the cascade waste heat distribution data; a phase change cooling and condensation reflux closed-loop module, used to determine the control parameters of steam pump 2 in real time based on the cascade waste heat distribution data, and adjust steam pump 2 to perform steam extraction according to the control parameters of steam pump 2. During the process of steam entering plate heat exchanger 3, the cascade waste heat distribution is adjusted accordingly. The system determines the control parameters of the power pump 4 based on the data, and adjusts the reflux of the condensate according to the control parameters of the power pump 4 to achieve liquid level stability and heat transfer; the adaptive waste heat cascade distribution module is used to determine the waste heat distribution ratio of the three branches of the first, second and third stages based on the waste heat distribution data, and drives the three branches to distribute waste heat to the heat-using end on demand according to the waste heat distribution ratio; the intelligent control and data feedback optimization module is used to continuously monitor the cascade waste heat distribution data, the control parameters of the steam pump 2, the control parameters of the power pump 4 and the waste heat distribution ratio, and to dynamically optimize each algorithm to achieve closed-loop control of the entire process.

[0072] This implementation scheme, through modular design, enables refined, intelligent, and dynamically adaptive control of waste heat utilization across the entire process. Multi-point real-time acquisition and robust preprocessing of various operating condition data ensure the quality of foundational data for subsequent control and optimization. Combined with phase change cooling and condensation reflux closed-loop regulation, it effectively maintains stable boiling chamber liquid level and pressure, achieving optimal synergy between cooling capacity and heat recovery. An adaptive waste heat cascade distribution module dynamically optimizes the waste heat distribution ratio of each branch in the first, second, and third stages based on real-time demand, load, and energy efficiency status, improving waste heat utilization efficiency and overall response flexibility. Simultaneously, it possesses intelligent closed-loop control and data feedback self-learning capabilities, continuously optimizing and adaptively correcting the distribution and control algorithms to maximize energy efficiency, ensure operational safety, and automate the entire process, thereby enhancing the data center's green and low-carbon operation capabilities and comprehensive energy utilization level.

[0073] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0074] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for adaptive cascade distribution of waste heat from phase change immersion cooling in data centers, characterized in that: Includes the following steps: S1, collects cascade waste heat distribution data in real time and performs data preprocessing on the cascade waste heat distribution data; S2, based on the waste heat distribution data of the cascade, the control parameters of the steam pump (2) are determined in real time, and the steam pump (2) is adjusted according to the control parameters of the steam pump (2) to pump steam. During the process of steam entering the plate heat exchanger (3), the control parameters of the power pump (4) are determined according to the waste heat distribution data of the cascade, and the reflux of the condensate is adjusted according to the control parameters of the power pump (4) to achieve liquid level stability and heat transfer. The specific process of determining the control parameters of the steam pump (2) in real time based on the waste heat distribution data of the cascade is as follows: While the server heat source is immersed in the boiling chamber (1), the high-density chip heats up and causes the low-boiling-point working fluid to undergo a boiling phase change, generating steam that flows into the main hot water pipe. In the process, the waste heat distribution data of the cascade is acquired in real time; the difference between the chip outlet temperature and the chip inlet temperature is calculated and multiplied by the specific heat capacity of the coolant to obtain the sensible heat value. The sensible heat value is added to the latent heat of vaporization of the working fluid to obtain the total heat required per unit mass; the server power consumption is multiplied by the total heat required per unit mass to obtain the heat load energy input value; the steam density is multiplied by the latent heat of vaporization of the working fluid to obtain the working fluid physical denominator; the heat load energy input value is divided by the working fluid physical denominator to obtain the basic target flow rate. Based on a sliding time window, the boiling chamber pressure when the server power consumption is less than the power consumption threshold is selected, and the average value is calculated to obtain the target pressure of the boiling chamber. The boiling chamber pressure difference is obtained by subtracting the boiling chamber target pressure from the boiling chamber pressure. The steam flow path pressure difference is obtained by subtracting the boiling chamber pressure from the steam pipeline pressure. The pressure correction coefficient is obtained by adding a constant to the product of the boiling chamber pressure difference and the target pressure response weight factor, and subtracting the product of the steam flow path pressure difference and the pipeline resistance correction weight factor. The steam extraction control value is obtained by multiplying the basic target flow rate by the pressure correction coefficient. S3, based on the waste heat distribution data of the cascade, determines the waste heat distribution ratio of the three branches of the first, second and third stages, and drives the three branches to distribute waste heat to the heat-using end on demand according to the waste heat distribution ratio; S4 continuously monitors the waste heat distribution data of the cascade, the control parameters of the steam pump (2), the control parameters of the power pump (4) and the waste heat distribution ratio, and performs dynamic optimization of each algorithm to achieve closed-loop control of the entire process.

2. The adaptive distribution method for waste heat in phase change immersion cooling of data centers according to claim 1, characterized in that, The specific process of real-time acquisition of cascade waste heat distribution data and data preprocessing of the cascade waste heat distribution data is as follows: Through multiple types of sensors distributed in the boiling chamber (1), plate heat exchanger (3), main hot water pipe and three branch waste heat heat-using ends, the cascade waste heat distribution data is collected in real time. The cascade waste heat distribution data includes: server power consumption, chip outlet temperature, chip inlet temperature, boiling chamber pressure, boiling chamber liquid level, steam flow rate, steam pipe pressure, steam pipe temperature, condensate return flow rate, condensate inlet temperature, condensate outlet temperature, main hot water pipe temperature, first-level branch return water temperature, second-level branch return water temperature, third-level branch return water temperature, first-level branch load, second-level branch load, third-level branch load, equipment operating status, and working fluid property data table embedded in the edge controller, which obtains the specific heat capacity of the coolant, the latent heat of vaporization of the coolant working fluid and the coolant density according to the coolant type, and the steam density obtained according to the steam pipe pressure and steam pipe temperature. The original cascade waste heat distribution data is denoised and smoothed using moving average filtering and median filtering methods, and sudden extreme values ​​and invalid data are removed using the standard fraction method. All acquisition points are calibrated with unified timestamps and data alignment. Unit conversion and normalization are performed on the cascade waste heat distribution data. For loss or anomalies in the main acquisition signal, multi-source redundancy switching, Kalman filtering, and physical model compensation are used. At the same time, data integrity is verified through cyclic redundancy check, logical consistency check, and timing integrity detection. A cascade waste heat distribution database is established, and the cascade waste heat distribution data is stored in the database.

3. The adaptive distribution method for waste heat in phase change immersion cooling of data centers according to claim 1, characterized in that, The specific process of adjusting the steam pump (2) according to the control parameters of the steam pump (2) to perform steam extraction is as follows: According to the steam extraction control value, the characteristic curve of the steam pump flow rate and speed is fitted by the ridge regression algorithm. The steam extraction control value is converted into the pump speed of the steam pump (2), and the controller issues the speed adjustment command to maintain the balance between steam extraction and server heat generation. When the boiling chamber pressure is greater than or equal to the normal pressure threshold, increase the steam pump (2) speed to accelerate the steam extraction speed; when the boiling chamber pressure is less than the normal pressure threshold, decrease the steam pump (2) speed to reduce the steam extraction speed; and calculate the standard deviation of the steam extraction control value within the sliding time window. When the standard deviation of the steam extraction control value is greater than the calculated steam extraction fluctuation threshold, record and trigger an alarm, and enable the fault-tolerant extraction path.

4. The adaptive distribution method for waste heat in phase change immersion cooling of data centers according to claim 1, characterized in that, The specific process of determining the control parameters of the power pump (4) based on the waste heat distribution data of the stepped flow into the plate heat exchanger (3) is as follows: Steam is pumped to plate heat exchanger (3). During the process of steam entering plate heat exchanger (3) and condensing with cold source in countercurrent flow, releasing latent heat and being converted into heat working fluid, the waste heat distribution data of the stage is acquired in real time. Based on the sliding time window, the boiling chamber liquid level with standard deviation less than the liquid level fluctuation threshold is selected, and the average value is calculated to obtain the target value of boiling chamber liquid level. The liquid level deviation is obtained by subtracting the boiling chamber liquid level from the target value of boiling chamber liquid level, and the liquid level deviation response is obtained by multiplying it with the liquid level deviation weighting factor. Calculate the difference between the condensate inlet temperature and the condensate outlet temperature, and multiply it by the condensate return flow rate, coolant density, and coolant specific heat capacity to obtain the condensation heat transfer power of the plate heat exchanger; calculate the difference between the condensate outlet temperature and the chip inlet temperature when the condensate flows back to the boiling chamber (1), and multiply it by the coolant density to obtain the unit volume temperature difference heat term; divide the plate heat exchanger condensation heat transfer power by the unit volume temperature difference heat term to obtain the heat transfer demand, and multiply it by the heat load flow rate weighting factor to obtain the heat transfer demand response; The difference between two consecutive boiling chamber liquid levels is calculated and divided by the sampling time interval to obtain the liquid level change rate. This rate is then multiplied by the liquid level rate weighting factor to obtain the liquid level rate response. The condensate reflux control value is obtained by adding the liquid level deviation response, the heat exchange demand response, and the liquid level rate response.

5. The adaptive distribution method for waste heat in phase change immersion cooling of data centers according to claim 1, characterized in that, The specific process of adjusting the reflux of condensate according to the control parameters of the power pump (4) to achieve liquid level stability and heat transfer is as follows: Based on the condensate reflux control value, the characteristic relationship between the pump flow rate and speed and the flow rate and opening degree is obtained by polynomial fitting and inverse solution method. The condensate reflux control value is converted into the target speed of the power pump (4) and the target opening degree of the valve. The central controller adjusts and provides feedback to carry out closed-loop regulation of the boiling chamber liquid level. If the standard deviation of the boiling chamber liquid level is greater than the liquid level fluctuation threshold, or the standard deviation of the condensate reflux control value is greater than the reflux fluctuation threshold, the safety plan will be triggered, the pump speed will be reduced, and an alarm will be issued. Simultaneously, the steam extraction control value and the condensate reflux control value are stored in the cascade waste heat distribution database in real time.

6. The adaptive distribution method for waste heat in phase change immersion cooling of data centers according to claim 1, characterized in that, The specific process of determining the waste heat distribution ratio of the first, second, and third-level branches based on the waste heat distribution data is as follows: When the working fluid in the main hot water pipe flows into the three-way valve (5), the waste heat distribution data of the cascade is obtained in real time, and the waste heat distribution ratio of the first-level branch, the second-level branch and the third-level branch is calculated: based on the sliding time window, the maximum value of the historical load of the i-th branch is selected as the maximum operating load of the i-th branch; The load ratio of the i-th branch is obtained by dividing the load of the i-th branch by the maximum operating load of the i-th branch. The usable temperature difference of the waste heat in the i-th branch is obtained by subtracting the return water temperature of the i-th branch from the temperature of the main hot water pipe. The weighting factor of the i-th branch, the load ratio of the i-th branch, and the available temperature difference of the waste heat of the i-th branch are multiplied to obtain the weighting term of the i-th branch. The total allocation weight is obtained by adding the allocation weights of the first-level branch, the second-level branch, and the third-level branch. The waste heat allocation ratio of the i-th branch is obtained by dividing the allocation weight of the i-th branch by the total allocation weight.

7. The adaptive distribution method for waste heat in phase change immersion cooling of data centers according to claim 1, characterized in that, The specific process of driving the three branch circuits to distribute waste heat to the heat-consuming end on demand according to the waste heat distribution ratio is as follows: Based on the waste heat distribution ratio of the i-th branch, a waste heat distribution instruction is generated and sent to the three-way valve (5). For the primary branch, the heat medium flows through the three-way valve (5) to the ORC power generation module (6) to drive the organic Rankine cycle to generate electricity and utilize the primary waste heat. For the secondary branch, the heat medium is diverted to the absorption refrigeration (7) module via a three-way valve (5) to prepare chilled water using the secondary waste heat, replacing mechanical refrigeration and reducing PUE; For the three-level branch, the heat medium is diverted to the heating module (8) and the dehumidification module (9) via the shut-off valve (10), and the seasonal mode is switched according to the outdoor temperature sensor signal: when the winter is turned on, the heat medium flows to the heating module (8) to provide heating for the data center building and the adjacent area; when the summer is turned on, the heat medium flows to the dehumidification module (9) to regulate the humidity of the data center environment. When any of the operating parameters of the branch equipment, including temperature, flow rate, pressure, and power, exceed the safety threshold, and the actual load continues to decrease, accompanied by continuous interruption of data acquisition, the allocation weight factor will be reduced to the minimum, the branch waste heat allocation will be suspended, and the remaining waste heat resources will be redistributed; at the same time, the waste heat allocation ratio of each branch will be written into the cascade waste heat allocation database, and the allocation weight factor of each branch will be optimized.

8. The adaptive distribution method for waste heat in phase change immersion cooling of data centers according to claim 1, characterized in that, The specific process of continuously monitoring the waste heat distribution data of the cascade, the control parameters of the steam pump (2), the control parameters of the power pump (4), and the waste heat distribution ratio, and dynamically optimizing each algorithm to achieve closed-loop control of the entire process is as follows: The system performs global aggregation and visual monitoring of waste heat distribution data in the cascade, steam extraction control values, condensate reflux control values, and waste heat distribution ratios in each branch, establishing a full-link operation status. The central controller continuously collects the execution status of all sensors, actuators, and waste heat distribution in each branch, and synchronously maps the waste heat distribution data in the cascade to the digital twin platform to visualize equipment health, load distribution, energy efficiency trends, and alarm information. Based on the overall operational status, adjustment commands are output to link redundant equipment and perform load reduction protection under abnormal and extreme conditions. By continuously comparing historical steam extraction control values, condensate reflux control values, and waste heat distribution ratios of each branch, the weighting factors and algorithm parameters are corrected online using a sliding window and self-learning optimization method to achieve closed-loop and energy-efficient operation of the entire process.

9. A data center phase change immersion cooling waste heat cascade adaptive distribution device, employing the data center phase change immersion cooling waste heat cascade adaptive distribution method as described in any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module is used to acquire cascade waste heat distribution data in real time and perform data preprocessing on the cascade waste heat distribution data. The phase change cooling and condensation reflux closed-loop module is used to determine the control parameters of the steam pump (2) in real time based on the waste heat distribution data of the cascade, and adjust the steam pump (2) to pump steam according to the control parameters of the steam pump (2). During the process of steam entering the plate heat exchanger (3), the control parameters of the power pump (4) are determined based on the waste heat distribution data of the cascade, and the reflux of the condensate is adjusted according to the control parameters of the power pump (4) to achieve liquid level stability and heat transfer. The adaptive waste heat cascade distribution module is used to determine the waste heat distribution ratio of the three branches (primary, secondary, and tertiary) based on the cascade waste heat distribution data, and drive the three branches to distribute waste heat to the heat-consuming end on demand according to the waste heat distribution ratio. The intelligent control and data feedback optimization module is used to continuously monitor the waste heat distribution data of the cascade, the control parameters of the steam pump (2), the control parameters of the power pump (4) and the waste heat distribution ratio, and to dynamically optimize each algorithm to achieve closed-loop control of the entire process.

Images