A method and system for dynamic heat transfer control of condensers based on multi-dimensional sensing and AI

Abstract

This invention discloses a dynamic heat exchange control method and system for condensers based on multi-dimensional sensing and AI, belonging to the field of intelligent heat exchange control technology. The method and system include: S1, collecting condenser heat exchange sensing data and preprocessing the data; S2, performing two rounds of data filtering on sampling points to issue emergency control commands and process abnormal data, constructing an effective condenser operation dataset; S3, evaluating the condenser's heat exchange adaptability, identifying load state categories, and outputting control commands; S4, executing the control commands and collecting execution feedback data, evaluating the condenser's heat exchange performance, determining whether the current control command needs adjustment based on the evaluation results, and transmitting the execution feedback data and evaluation results back for threshold updates and control strategy optimization. This solves the problems of limited phase change and excessive energy consumption under load changes caused by fixed condenser heat exchange control.

Description

Technical Field

[0001] This invention relates to the field of intelligent heat exchange control technology, specifically to a dynamic heat exchange control method and system for condensers based on multi-dimensional sensing and AI. Background Technology

[0002] With the widespread application of high-performance computing, high-density packaging, and smart terminal devices, the heat dissipation capacity of electronic systems has become a key factor restricting their stable operation and energy efficiency optimization. Among them, the condenser, as a core heat dissipation unit, has a significant impact on chip temperature control and overall system power consumption. Traditional condenser control strategies are mainly based on fixed thresholds and rule-driven approaches, lacking the ability to adapt to changes in the operating environment in real time. This can easily lead to problems such as insufficient heat dissipation, excessive energy consumption, or resource waste, making it difficult to meet the refined thermal management requirements under current complex operating conditions.

[0003] For example, the invention disclosed in CN115293373A provides a method, apparatus, and electronic device for determining a condenser maintenance strategy. The method includes: acquiring turbine exhaust steam data, condensate data, fan data, and the heat exchange area of ​​the condenser; the exhaust steam data includes the temperature after heat exchange of the final stage exhaust steam and the initial temperature of the final stage exhaust steam; determining the steam heat release of the turbine based on the turbine exhaust steam data and condensate data; determining the heat exchange temperature difference of the condenser based on the steam heat release, fan data, the temperature after heat exchange of the final stage exhaust steam, and the initial temperature of the final stage exhaust steam; determining the overall heat transfer coefficient based on the steam heat release, the heat exchange temperature difference, and the heat transfer area; and determining the condenser maintenance strategy based on the overall heat transfer coefficient. This invention can determine a reasonable condenser maintenance strategy for the operating conditions of air-cooled thermal power units, ensuring the normal operation of air-cooled thermal power units.

[0004] For example, invention publication CN104166794B discloses a nuclear power plant condenser characteristic test data acquisition and analysis system. This system includes: a data acquisition system for acquiring data required for condenser characteristic tests; an IMP (Inlet and Outlet Temperature) acquisition system for acquiring seawater inlet and outlet temperatures, condensate temperature, and back pressure data, and processing the acquired data; and a calculation and analysis system for real-time reading of data from the IMP acquisition system and receiving user-input data on nuclear island thermal power, generator electrical power, and circulating water salinity, and calculating and displaying test results based on the condenser's design parameters and the received data. Implementing this invention provides simple operation and accurate analysis, ensuring the analysis and evaluation of condenser thermodynamic performance, enabling maintenance and improvement to ensure the condenser's optimal operating condition, thereby enhancing the safe and economical operation of the unit.

[0005] However, although the above technical solutions can improve the safety and reliability of condenser operation in specific scenarios, they still have the following shortcomings: lack of real-time perception, fusion and dynamic processing capabilities of operating data; lack of dynamic control mechanism based on AI strategy optimization, making it difficult to ensure heat dissipation performance under high load and reduce energy consumption under low load; and lack of control logic that establishes closed-loop linkage between condenser heat exchange adaptability assessment, load identification and control commands, resulting in insufficient system intelligence and adaptability.

[0006] Therefore, in order to address the above problems, there is an urgent need for a dynamic heat exchange control method and system for condensers based on multi-dimensional perception and AI. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a dynamic heat exchange control method and system for condensers based on multi-dimensional perception and AI, which solves the problems of limited phase change and excessive energy consumption caused by fixed condenser heat exchange control under load changes.

[0009] Technical solution

[0010] To achieve the above objectives, this invention provides the following technical solution: a dynamic heat exchange control method and system for condensers based on multi-dimensional perception and AI, comprising: S1, collecting condenser heat exchange sensing data, and performing time synchronization, noise separation, physical consistency verification, and normalization mapping processing on the condenser heat exchange sensing data; S2, based on the preprocessed condenser heat exchange sensing data, performing two rounds of data filtering on the sampling points to realize the issuance of emergency control commands and the processing of abnormal point data, and constructing an effective condenser operation dataset; S3, evaluating the condenser heat exchange adaptability based on the effective condenser operation dataset, extracting the chip temperature deviation ratio and the box air pressure deviation ratio, and identifying the load state category; inputting the heat exchange adaptability evaluation results, load state category, and effective condenser operation data into a reinforcement learning decision model, and outputting control commands; S4, executing the control commands and collecting execution feedback data, evaluating the condenser heat exchange execution performance, determining whether the current control commands need to be adjusted based on the evaluation results, and sending back the execution feedback data and evaluation results for threshold updates and control strategy optimization.

[0011] Further, the specific steps for collecting condenser heat exchange sensing data and performing time synchronization, noise separation, physical consistency verification, and normalization mapping on the condenser heat exchange sensing data are as follows: Collect condenser heat exchange sensing data, including chip temperature, housing pressure, cooling coil inlet water temperature, cooling coil outlet water temperature, cooling water density, cooling water flow rate, cooling water specific heat capacity, pump power, and housing saturated steam pressure; perform cross-sensor time synchronization and sampling rate unification processing on the condenser heat exchange sensing data using multi-channel timestamp comparison and sampling delay compensation methods; and combine this with Kalman filtering... A multi-scale signal reconstruction method based on wavelet decomposition is used to eliminate measurement noise and separate periodic disturbances in condenser heat transfer sensing data. A heat-flow integration consistency verification mechanism is established based on the energy conservation equation and the pressure-flow coupling model to identify and remove data from condenser heat transfer sensing data that does not conform to physical laws. Data format standardization and interface compatibility conversion are performed on condenser heat transfer sensing data through multi-dimensional feature encoding and adaptive encapsulation of transmission protocols. Unit conversion and feature scale normalization are performed on condenser heat transfer sensing data through physical quantity normalization mapping and dynamic interval stretching algorithms.

[0012] Furthermore, based on the pre-processed condenser heat exchange sensing data, two rounds of data filtering are performed on the sampling points to realize the issuance of emergency control commands and the processing of abnormal data. The specific steps for constructing an effective condenser operation dataset are as follows: Extract the pre-processed condenser heat exchange sensing data and perform the first round of data filtering on each sampling point: When the chip temperature exceeds the upper limit threshold, the box pressure exceeds the upper limit threshold, or the cooling water flow rate is lower than the lower limit threshold, the data is judged to be abnormal and emergency control is triggered. The inlet water temperature of the target cooling coil and the target pump power are adjusted to the emergency set values ​​and issued to the actuator. The remaining sampling points enter the second round of filtering. For the sampling points that enter the second round of filtering, the temperature difference between the inlet and outlet water of the cooling coil is calculated based on the difference between the inlet water temperature and the outlet water temperature of the cooling coil. Within the dynamic window, the mean and standard deviation of chip temperature, enclosure air pressure, cooling water flow rate, and inlet / outlet water temperature difference of the cooling coil are calculated to assess the degree of condenser operating abnormality and obtain the condenser operating abnormality assessment value. The condenser operating abnormality assessment value and the operating abnormality threshold are compared in real time. When the condenser operating abnormality assessment value exceeds the abnormality scoring threshold, the corresponding sampling point is determined to be an abnormal point. If it is an isolated abnormal point, the moving average of the condenser heat exchange sensing data of adjacent sampling points is used instead. If three or more consecutive sampling points are determined to be abnormal points and the corresponding parameter acquisition source is the same sensor, the data is switched to the backup sensor. Otherwise, the corresponding sampling point is determined to be a normal point and the original data is retained. The condenser heat exchange sensing data retained after two rounds of data filtering are used to construct an effective condenser operation dataset according to the sampling point order.

[0013] Furthermore, the specific steps for assessing the degree of condenser operating abnormality and obtaining the condenser operating abnormality assessment value are as follows: Divide the square of the difference between the chip temperature and the mean chip temperature by the square of the standard deviation of the chip temperature, and add one to obtain the temperature deviation item; divide the square of the difference between the housing pressure and the mean housing pressure by the square of the standard deviation of the housing pressure, and add one to obtain the pressure deviation item; divide the square of the difference between the cooling water flow rate and the mean cooling water flow rate by the square of the standard deviation of the cooling water flow rate, and add one to obtain the flow rate deviation item; divide the square of the difference between the inlet and outlet water temperature of the cooling coil and its mean by the standard deviation of the cooling coil temperature, and add one to obtain the flow rate deviation item. The temperature deviation term is obtained by adding one to the square of the standard deviation of the temperature difference. The temperature deviation term, pressure deviation term, flow rate deviation term, and temperature deviation term are added together and the square root is taken to obtain the normalized deviation sum of multiple parameters. The temperature-pressure cross-correction term is obtained by multiplying the difference between the chip temperature and the target chip safe temperature by the difference between the chamber pressure and the chamber saturated vapor pressure, taking the absolute value, and dividing it by the product of the target chip safe temperature and the chamber saturated vapor pressure. The temperature-pressure cross-correction term is added one to obtain the correction factor. The abnormal assessment value of the condenser operating condition is obtained by multiplying the normalized deviation sum of multiple parameters by the correction factor.

[0014] Furthermore, based on the effective condenser operation dataset, the specific steps for evaluating the condenser heat transfer adaptability and extracting the chip temperature deviation ratio and the box pressure deviation ratio to identify the load state category are as follows: Extract the effective condenser operation dataset and evaluate the condenser heat transfer adaptability: Multiply the cooling water density, cooling water flow rate, and cooling water specific heat capacity by the inlet and outlet water temperature difference of the cooling coil to obtain the heat transfer power component; Take the larger value between the difference between the box saturated steam pressure and the box pressure and the minimum term to obtain the pressure margin component; Divide the heat transfer power component by the pressure margin component to obtain the pressure-corrected heat transfer power. The difference between the chip temperature and the target chip's safe temperature is divided by the target chip's safe temperature to obtain the relative temperature deviation coefficient. The pressure-corrected heat transfer power is multiplied by the relative temperature deviation coefficient to obtain the condenser heat transfer adaptation evaluation value. The average chip temperature and the average chamber pressure are extracted and compared with the chip temperature threshold and the chamber pressure threshold, respectively, to obtain the chip temperature deviation ratio and the chamber pressure deviation ratio. The condenser heat transfer adaptation evaluation value, the chip temperature deviation ratio, and the chamber pressure deviation ratio are combined to form a feature vector, which is input into a BP neural network classifier to output the current load state category, including high load, medium load, and low load.

[0015] Further, the specific steps for inputting the heat exchange adaptation evaluation results, load state categories, and effective condenser operating data into the reinforcement learning decision model and outputting control commands are as follows: Based on historical effective condenser operating data, a reinforcement learning decision model is constructed using the Q-Learning algorithm. The adjustment amount of the cold target cooling coil inlet water temperature and the adjustment amount of the target pump power are set as the action set, and the load state category, condenser heat exchange adaptation evaluation value, and effective condenser operating data are set as the state set. The reward function is to maximize heat dissipation efficiency under high load and minimize energy consumption under low load, thus completing the training of the reinforcement learning decision model. The current load state category, condenser heat exchange adaptation evaluation value, and effective condenser operating data are used as decision input parameters to input into the reinforcement learning decision model, and the control commands are output as follows: under high load, reduce the cold target cooling coil inlet water temperature and increase the target pump power; under low load, increase the cold target cooling coil inlet water temperature and reduce the target pump power; under medium load, maintain the current cold target cooling coil inlet water temperature and target pump power.

[0016] Furthermore, the specific steps for executing control commands and collecting execution feedback data are as follows: receiving control commands, sending the adjustment amount of the target cooling coil inlet water temperature to the cooling water temperature control execution unit, sending the target pump power adjustment amount to the pump speed control unit, and driving the control execution to perform real-time adjustment; during the control execution process, collecting the actual value of the cooling coil inlet water temperature after execution, the actual value of the cooling coil outlet water temperature after execution, the actual value of the pump power after execution, the actual execution response time, and the actual total energy consumption within the control cycle, and constructing an execution feedback dataset.

[0017] Further, the specific steps for evaluating the condenser heat exchange performance are as follows: Based on the execution feedback dataset, multiply the cooling water density, cooling water flow rate, and cooling water specific heat capacity, and then multiply by the difference between the actual value of the cooling coil inlet water temperature and the actual value of the cooling coil outlet water temperature after execution to obtain the heat exchange power component; divide the actual execution response time by the reference response time and add one to obtain the response time correction term; calculate the absolute value of the difference between the actual value of the cooling coil inlet water temperature and the target cooling coil inlet water temperature after execution, calculate the absolute value of the difference between the actual value of the pump power and the target pump power after execution, add the two absolute values ​​and add one to obtain the execution deviation correction term; multiply the actual total energy consumption, response time correction term, and execution deviation correction term within the control cycle to obtain the total corrected energy consumption; divide the heat exchange power component by the total corrected energy consumption to obtain the heat exchange performance evaluation value.

[0018] Furthermore, based on the evaluation results, the specific steps for determining whether to adjust the current control command and transmitting the execution feedback data and evaluation results back for threshold updates and control strategy optimization are as follows: Real-time comparison of the heat exchange performance evaluation value and the performance threshold. When the heat exchange performance evaluation value is greater than or equal to the performance threshold, the current control strategy is maintained for the next control cycle. When the heat exchange performance evaluation value is less than the performance threshold, based on the correspondence between the actual value of the cooling coil inlet water temperature after execution and the target cooling coil inlet water temperature, and the actual value of the pump power after execution and the target pump power recorded in the execution feedback dataset, the cooling coil inlet water temperature deviation and pump power deviation are determined. Corrected cold target cooling coil inlet water temperature and target pump power commands are generated according to the direction and magnitude of the deviations and sent to the actuator. Regardless of whether strategy correction is performed, the execution feedback dataset and heat exchange performance evaluation value are transmitted back to the data filtering stage and the intelligent decision-making stage, respectively. In the data filtering stage, the temperature threshold, air pressure threshold, and flow threshold are updated. In the intelligent decision-making stage, the cold target cooling coil inlet water temperature adjustment strategy and the target pump power adjustment strategy are optimized to achieve closed-loop operation of dynamic heat exchange control.

[0019] The second aspect of this invention provides a dynamic heat exchange control system for condensers based on multi-dimensional sensing and AI, comprising: a data sensing preprocessing module, an anomaly screening emergency control module, an intelligent state judgment decision generation module, and an execution feedback closed-loop optimization module. The data sensing preprocessing module collects condenser heat exchange sensing data and performs time synchronization, noise separation, physical consistency verification, and normalization mapping on the condenser heat exchange sensing data. The anomaly screening emergency control module performs two rounds of data screening on sampling points based on the preprocessed condenser heat exchange sensing data, enabling the issuance of emergency control commands and processing of anomaly point data, thereby constructing an effective cooling system. The system includes: a condenser operation dataset; an intelligent state judgment and decision generation module, used to evaluate the condenser heat exchange adaptability based on the effective condenser operation dataset, extract the chip temperature deviation ratio and the box pressure deviation ratio, and identify the load state category; inputting the heat exchange adaptability evaluation results, load state category, and effective condenser operation data into the reinforcement learning decision model, and outputting control commands; and an execution feedback closed-loop optimization module, used to execute control commands and collect execution feedback data, evaluate the condenser heat exchange performance, determine whether the current control command needs to be adjusted based on the evaluation results, and send the execution feedback data and evaluation results back for threshold updates and control strategy optimization.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) The condenser dynamic heat exchange control method and system based on multi-dimensional perception and AI, by setting safety thresholds for chip temperature, box pressure and cooling water flow rate, combined with a two-round data screening mechanism, realizes real-time identification and emergency control response of abnormal condenser operating conditions. It can automatically issue control commands when the sampled data exceeds the limit, effectively preventing heat exchange failure, abnormal energy consumption or system overheating, and improving the safety and stability of the system.

[0023] (2) The condenser dynamic heat exchange control method and system based on multidimensional perception and AI constructs a condenser heat exchange adaptation evaluation model and combines the deviation ratio between chip temperature and box pressure. It uses a BP neural network to achieve high-precision intelligent identification of the current load state, opens up the feedback path between condenser operation and calculation load, provides targeted input for subsequent heat exchange strategies, and realizes the transformation from static control to state perception-driven.

[0024] (3) The condenser dynamic heat exchange control method and system based on multidimensional perception and AI constructs a decision model by using Q-Learning reinforcement learning algorithm. The optimization goal is to maximize heat dissipation efficiency under high load and minimize energy consumption under low load. The optimal cooling coil water temperature and pump power control command is trained and output, realizing the self-learning and dynamic evolution of the strategy, and significantly improving the energy efficiency ratio and intelligent control level of the condenser.

[0025] (4) The condenser dynamic heat exchange control method and system based on multidimensional perception and AI constructs a heat exchange performance evaluation model, quantifies the control effect in real time, and sends the execution feedback data back to the data screening and decision-making process to complete the dynamic update of thresholds such as temperature, pressure and flow rate and the optimization and iteration of the strategy model, forming a closed-loop adaptive heat exchange control mechanism to ensure the long-term stable and efficient operation of the system. Attached Figure Description

[0026] Figure 1 The flowchart shows a dynamic heat exchange control method for condensers based on multi-dimensional perception and AI.

[0027] Figure 2 A schematic diagram of the structure of a condenser dynamic heat exchange control system based on multi-dimensional perception and AI;

[0028] Figure 3 Trend chart of condenser heat exchanger adaptation evaluation values;

[0029] Figure 4 This is a schematic diagram of a dynamic heat exchange control system for condensers based on multi-dimensional perception and AI.

[0030] In the diagram, 1. Saturated vapor pressure sensor; 2. Cooling coil; 3. Condenser; 4. Server; 5. Temperature sensor; 6. Pump; 7. Coolant reservoir. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0032] Please see Figures 1-4 This invention provides a technical solution: a dynamic heat exchange control method and system for condensers based on multi-dimensional perception and AI, comprising: S1, collecting condenser heat exchange sensing data, and performing time synchronization, noise separation, physical consistency verification, and normalization mapping processing on the condenser heat exchange sensing data; S2, based on the preprocessed condenser heat exchange sensing data, performing two rounds of data filtering on the sampling points to realize the issuance of emergency control commands and the processing of abnormal point data, and constructing an effective condenser operation dataset; S3, evaluating the heat exchange adaptation degree of condenser 3 based on the effective condenser operation dataset, and extracting the chip temperature deviation ratio and the box air pressure deviation ratio to identify the load state category; inputting the heat exchange adaptation evaluation results, load state category, and effective condenser operation data into a reinforcement learning decision model, and outputting control commands; S4, executing the control commands and collecting execution feedback data, evaluating the heat exchange performance of condenser 3, determining whether the current control commands need to be adjusted based on the evaluation results, and sending back the execution feedback data and evaluation results for threshold updates and control strategy optimization.

[0033] Specifically, the steps for collecting condenser heat exchange sensing data and performing time synchronization, noise separation, physical consistency verification, and normalization mapping on the condenser heat exchange sensing data are as follows: Collecting condenser heat exchange sensing data includes chip temperature, housing pressure, cooling coil inlet water temperature, cooling coil outlet water temperature, cooling water density, cooling water flow rate, cooling water specific heat capacity, pump power, and housing saturated steam pressure. Each sensing data item originates from an independent sensing channel and has a corresponding timestamp and sampling frequency label. Using multi-channel timestamp comparison and sampling delay compensation methods, combined with the sampling start time and sampling period of each data channel, cross-sensor time synchronization and unified sampling rate processing are performed on each condenser heat exchange sensing data item to ensure that all types of sensing data are aligned under a unified time reference. Using a multi-scale signal reconstruction method combining Kalman filtering and wavelet decomposition, multi-dimensional filtering and disturbance separation processing are performed on each condenser heat exchange sensing data item to effectively suppress high-frequency measurement noise and extract stable heat exchange trends. Based on energy conservation... A heat flow consistency verification mechanism is established by coupling the constant equation with the pressure difference flow rate model. The theoretical heat transfer is estimated by multiplying the cooling water density data, cooling water flow rate data, and cooling water specific heat capacity data, and then combining this with the temperature difference between the cooling coil inlet water temperature data and the cooling coil outlet water temperature data. This is compared with the load heat flow derived from the chip temperature data and the housing pressure data to identify abnormal sampling points inconsistent with the thermal balance. Operations such as chip temperature data rejection, housing pressure data rejection, and cooling water flow rate data rejection are performed to ensure that the input data follows physical laws. Through multi-dimensional feature encoding and adaptive encapsulation methods of transmission protocols, each condenser heat transfer sensing data is encoded in a unified format, and a data encapsulation structure adaptable to different communication interfaces is constructed to achieve lossless data exchange and parallel processing across systems. Through physical quantity normalization mapping and dynamic interval stretching algorithms, unit conversion and feature scale normalization are performed on each condenser heat transfer sensing data, outputting a standardized sensing vector group as effective input for subsequent heat transfer control algorithms and state judgment models.

[0034] In this implementation scheme, by sequentially performing cross-sensor time synchronization, multi-scale noise separation, physical consistency verification, format standardization encapsulation, and feature normalization mapping on chip temperature data, chamber air pressure data, cooling coil inlet water temperature data, cooling coil outlet water temperature data, cooling water density data, cooling water flow rate data, cooling water specific heat capacity data, pump power data, and chamber saturated steam pressure data, data quality control can be achieved in terms of time alignment, numerical accuracy, physical rationality, interface matching, and dimensional uniformity. This provides a high-confidence data foundation for subsequent anomaly screening, status identification, regulation generation, and feedback evaluation, thereby improving the accuracy, stability, and adjustability of the heat exchange regulation process.

[0035] Specifically, based on the preprocessed condenser heat exchange sensing data, two rounds of data filtering are performed on each sampling point to realize the issuance of emergency control commands and the processing of abnormal data. The specific steps for constructing an effective condenser operation dataset are as follows: Extract the condenser heat exchange sensing data after cross-sensor time-series synchronization processing, multi-scale noise separation processing, physical consistency verification processing, data format standardization processing, and physical quantity normalization mapping processing. For each sampling point, extract the chip temperature data, housing air pressure data, and cooling water flow rate data in sequence and perform the first round of data filtering. When the chip temperature exceeds the upper limit threshold of the chip temperature, the housing air pressure exceeds the upper limit threshold of the housing air pressure, and the cooling water flow rate is lower than the lower limit threshold of the cooling water flow rate, the data corresponding to the sampling point is determined to be abnormal, and the emergency control process is immediately triggered. The inlet water temperature of the target cooling coil is set to the emergency safety value of the corresponding cooling water temperature, and the target pump power is set to the emergency output value within the pump power control range. These two emergency settings are then issued to the corresponding actuators after being packaged into commands. The remaining sampling points that have not triggered abnormalities enter the second filtering process. The actuator is a key component for implementing the control strategy, receiving control commands and performing corresponding physical adjustments. The actuator includes a cooling water temperature control unit and a pump speed control unit. The cooling water temperature control unit precisely adjusts the cooling medium temperature based on the target cooling coil inlet water temperature adjustment; the pump speed control unit controls the pump speed based on the target pump power adjustment, thereby regulating the cooling water circulation rate. The actuator's response speed and adjustment accuracy directly affect the actual execution effect of the heat exchange control strategy, making it a core component for achieving dynamic heat exchange optimization.For sampling points entering the second screening process, the inlet and outlet water temperature difference of the cooling coil is calculated based on the real-time difference between the inlet and outlet water temperatures of the cooling coil. Based on a preset sliding window, the sliding mean and sliding standard deviation of the chip temperature, housing pressure, cooling water flow rate, and inlet and outlet water temperature difference of the cooling coil are calculated respectively to construct a statistical feature space of the condenser 3's thermodynamic parameters. Combining multi-parameter normalization deviation and temperature and pressure interaction correction factors, the degree of abnormality in the condenser 3's operating condition at this sampling point is assessed, ultimately obtaining the condenser operating condition abnormality assessment value corresponding to this sampling point. The condenser operating condition abnormality assessment value is compared with the operating condition abnormality scoring threshold in real time. When the condenser operating condition abnormality assessment value exceeds the currently set operating condition abnormality scoring threshold, the sampling point is determined to be an abnormal point. If the abnormal point exhibits isolated distribution characteristics in the time series, the chip temperature of adjacent sampling points is used. The moving average of four types of condenser heat exchange sensing data—temperature, chamber pressure, cooling water flow rate, and the temperature difference between the inlet and outlet water of the cooling coil—is used to replace the anomaly point. If three or more consecutive sampling points are identified as anomalies, and their corresponding abnormal parameter data all originate from the same sensor channel, it is determined that the acquisition source is unstable, and the corresponding backup sensor is immediately switched to collect values, while retaining the sampling sequence number. Otherwise, the sampling point is determined to be a valid sampling point, and its original condenser heat exchange sensing data is retained without correction. Finally, the nine types of condenser heat exchange sensing data—chip temperature, chamber pressure, cooling water flow rate, cooling coil inlet water temperature, cooling coil outlet water temperature, cooling water density, cooling water specific heat capacity, pump power, and chamber saturated vapor pressure—retained after two rounds of screening are arranged sequentially according to sampling time to construct a valid condenser operation dataset, which is used for subsequent heat exchange adaptation evaluation and control command generation.

[0036] In this implementation scheme, an emergency control trigger mechanism is constructed by performing two rounds of point-by-point screening on the condenser heat exchange sensing data, combining the upper limit threshold of chip temperature, the upper limit threshold of casing air pressure, and the lower limit threshold of cooling water flow. Furthermore, a moving average and moving standard deviation calculation method is introduced to generate anomaly assessment values ​​for condenser operating conditions, effectively identifying sudden abnormal operating conditions and data source stability issues. This method can accurately replace isolated abnormal sampling point data and dynamically switch the acquisition channels of unstable sensors, ultimately selecting a valid condenser operation dataset that meets the statistical laws of the thermodynamic characteristics of condenser 3. This lays an accurate and reliable data foundation for subsequent heat exchange adaptation assessment and control command generation, significantly improving the data processing's anti-interference capability and the execution reliability of condenser 3 control.

[0037] Specifically, the steps for assessing the degree of abnormality in condenser 3 operating conditions and obtaining the condenser operating condition abnormality assessment value are as follows: Square the difference between the chip temperature corresponding to the current sampling point and the mean of all chip temperature sample values ​​within the sliding window, and divide this squared difference by the sum of the square of the standard deviation of the chip temperature sample values ​​and a constant term to obtain the temperature deviation term; square the difference between the chamber pressure corresponding to the sampling point and the mean of the chamber pressure sample values ​​within the sliding window, and divide this squared difference by the sum of the square of the standard deviation of the chamber pressure sample values ​​and a constant term to obtain the pressure deviation term; The squared difference between the cooling water flow rate at the sampling point and the mean cooling water flow rate within the sliding window is then divided by the sum of the squared standard deviation of the cooling water flow rate sample and a constant term to obtain the flow rate deviation term. The difference between the cooling coil inlet water temperature and the cooling coil outlet water temperature is used to obtain the current cooling coil inlet and outlet water temperature difference. The squared difference is then divided by the sum of the squared standard deviation of the cooling coil inlet and outlet water temperature difference and a constant term to obtain the temperature difference deviation term. The temperature deviation term and the air pressure deviation term are then calculated. The flow rate deviation and temperature difference deviation terms are accumulated, and the square root of the accumulated value is taken to obtain the multi-parameter normalized deviation sum, which is used to measure the degree of coordinated deviation between each sensing data and its short-period statistical characteristics. The difference between the current chip temperature and the target chip safe temperature is calculated, and the difference between the current chamber pressure and the chamber saturated vapor pressure is calculated. The two differences are multiplied and the absolute value is taken. The result is then divided by the product of the target chip safe temperature and the chamber saturated vapor pressure to obtain the temperature-pressure interaction correction term, which is used to reflect the degree of coupling deviation between the heat load and the chamber gas-liquid phase equilibrium state. Among them, the target chip safe temperature is the reference temperature threshold in heat exchange control, which represents the highest temperature upper limit at which the chip can still work stably under full load operation. It is used to determine whether the current chip temperature is within the safe range and to guide the dynamic adjustment of the heat exchange control strategy. The temperature-pressure interaction correction term is added to a constant term to obtain the correction factor. Finally, the multi-parameter normalized deviation sum is multiplied by the correction factor to obtain the condenser condition anomaly assessment value corresponding to the sampling point, which is used to characterize the degree of matching deviation between the current thermal operation state of condenser 3 and the chip heat exchange demand.

[0038] The specific formula for calculating the condenser operating condition anomaly assessment value is as follows:

[0039] ;

[0040] In the formula, S represents the abnormal assessment value of the condenser operating condition. Indicates chip temperature. This indicates the average chip temperature. Indicates the standard deviation of chip temperature. Indicates the air pressure inside the chamber. This indicates the average air pressure inside the chamber. This indicates the standard deviation of the air pressure in the enclosure. Indicates the cooling water flow rate. This represents the average cooling water flow rate. Indicates the standard deviation of cooling water flow rate. This indicates the temperature difference between the inlet and outlet water of the cooling coil. This represents the average temperature difference between the inlet and outlet water of the cooling coil. This indicates the standard deviation of the temperature difference between the inlet and outlet water of the cooling coil. Indicates the target chip's safe temperature. This indicates the saturated steam pressure of the chamber.

[0041] In this implementation scheme, an anomaly assessment value for condenser operating conditions is constructed by combining multi-parameter normalized deviation with a temperature and pressure interaction correction term. This effectively integrates the statistical deviation characteristics of four types of condenser heat exchange sensing data: chip temperature, housing pressure, cooling water flow rate, and the temperature difference between the inlet and outlet water of the cooling coil. Furthermore, a coupling correction factor between the target chip safety temperature and the housing saturated vapor pressure is introduced, enhancing the robustness and sensitivity of anomaly identification. This method can achieve precise quantitative determination of anomalies under different condenser operating conditions, improving the accuracy and dynamic adaptability of data screening, and providing a reliable state basis for subsequent control command generation.

[0042] Specifically, the following steps were taken to evaluate the heat exchange adaptability of condenser 3 based on the effective condenser operation dataset and to extract the chip temperature deviation ratio and the box pressure deviation ratio to identify the load state category: The effective condenser operation dataset retained after two rounds of data filtering was extracted, and a parameter combination was constructed to evaluate the heat exchange capacity matching of condenser 3. Specifically, the cooling water density, cooling water flow rate, and cooling water specific heat capacity were multiplied, and then multiplied by the temperature difference between the inlet and outlet water of the cooling coil to obtain the heat exchange power component reflecting the current heat exchange capacity of condenser 3. Next, the difference between the box saturated vapor pressure and the box pressure was compared with the minimum term, and the larger value was taken as the pressure margin. A pressure margin component is used to maintain computational stability. The minterm is a non-zero, minimal positive real number used to provide a minimum positive limit when the difference between the saturated vapor pressure and the gas pressure in the chamber approaches zero, preventing computational anomalies caused by division by zero or numerical instability. The heat transfer power component is then divided by the pressure margin component to obtain the pressure-corrected heat transfer power, incorporating pressure-related factors. The difference between the chip temperature and the target chip's safe temperature is then calculated and divided by the target chip's safe temperature to obtain the standardized relative temperature deviation coefficient. The pressure-corrected heat transfer power is then multiplied by the relative temperature deviation coefficient to obtain the final condenser heat transfer adaptation evaluation value, characterizing the match between heat transfer capacity and load requirements. Furthermore, the mean chip temperature and mean chamber pressure in the condenser heat transfer sensing data are extracted and compared with the chip temperature threshold and chamber pressure threshold, respectively, to calculate the chip temperature deviation ratio and the chamber pressure deviation ratio, reflecting the relative distance between the current heat load state and the safety boundary. The condenser heat exchange adaptation evaluation value, the chip temperature deviation ratio, and the box air pressure deviation ratio are used to form a feature vector, which is then input into a BP neural network classifier. The classifier outputs the load state category of the current condenser 3, which includes high load state, medium load state, and low load state. This category is used to provide key decision-making basis for the generation of subsequent dynamic control commands.

[0043] The specific calculation formula for the condenser heat exchanger adaptation evaluation value is as follows:

[0044] ;

[0045] In the formula, H represents the condenser heat exchange adaptation evaluation value. C represents the density of the cooling water, C represents the specific heat capacity of the cooling water, and Q represents the flow rate of the cooling water. This indicates the temperature difference between the inlet and outlet water of the cooling coil. This indicates the saturated vapor pressure of the chamber. Indicates the air pressure inside the chamber. Indicates chip temperature. Indicates the target chip's safe temperature. Indicates a minus term.

[0046] In this embodiment, Table 1 is a data table of condenser heat exchange adaptation evaluation values, listing the cooling water density, cooling water specific heat capacity, cooling water flow rate, cooling coil inlet and outlet water temperature difference, chamber saturated vapor pressure, chamber air pressure, chip temperature, target chip safe temperature, and the condenser heat exchange adaptation evaluation value calculated based on the above variables at 5 sampling points. Specific data explanations are as follows: At sampling point 1, the cooling water density is 995.99, the cooling water specific heat capacity is 4081.41, the cooling water flow rate is 0.03959, the cooling coil inlet and outlet water temperature difference is 2.72, the chamber saturated vapor pressure is 105.48, the chamber air pressure is 100.8, the chip temperature is 66.15, the target chip safe temperature is 66.1, and the calculated condenser heat exchange adaptation evaluation value is 78.51; At sampling point 2, the cooling water density is 1058.58, the cooling water specific heat capacity is 40... 89.35, cooling water flow rate is 0.03837, cooling coil inlet and outlet water temperature difference is 2.36, chamber saturated steam pressure is 115.59, chamber gas pressure is 111.54, chip temperature is 68.69, target chip safe temperature is 68.63, the calculated condenser heat exchange adaptation evaluation value is 95.51; at sampling point 3, cooling water density is 935.82, cooling water specific heat capacity is 4137.77, cooling water flow rate is 0.03551, cooling coil inlet and outlet water temperature difference is 4.43, chamber... The saturated vapor pressure of the condenser is 113.69 g / L, the chamber pressure is 109.81 g / L, the chip temperature is 65.71 °C, and the target chip's safe temperature is 65.68 °C. The calculated condenser heat transfer compatibility evaluation value is 73.08. At sampling point 4, the cooling water density is 950.60 g / L, the cooling water specific heat capacity is 4046.01 L / L, the cooling water flow rate is 0.04457 g / L, the inlet and outlet water temperature difference of the cooling coil is 2.26 °C, the chamber saturated vapor pressure is 118.51 g / L, the chamber pressure is 113.8 °C, and the chip temperature is 65.12 °C. The target chip's safe temperature is 65.06°C, and the calculated condenser heat exchange compatibility evaluation value is 69.39. At sampling point 5, the cooling water density is 926.49 g / L, the cooling water specific heat capacity is 4160.34 g / L, the cooling water flow rate is 0.04099 g / L, the inlet and outlet water temperature difference of the cooling coil is 2.20 °C, the saturated steam pressure of the housing is 113.31 °C, the housing gas pressure is 110.71 °C, the chip temperature is 68.81 °C, and the target chip's safe temperature is 68.77 °C. The calculated condenser heat exchange compatibility evaluation value is 73.91.

[0047] Table 1. Condenser Heat Exchanger Compatibility Evaluation Values

[0048]

[0049] like Figure 3The figure shows the changing trends of the condenser heat exchange adaptation evaluation values ​​at five different sampling points. The horizontal axis represents the sampling point number, and the vertical axis represents the corresponding heat exchange adaptation evaluation value, used to measure the degree of matching between the current operating condition of condenser 3 and the target heat exchange state. The figure shows that: the evaluation value is highest at sampling point 2, indicating that condenser 3 has optimal heat exchange capacity at this moment; the evaluation value is lowest at sampling point 4, indicating relatively weak heat exchange efficiency, which may require strategic intervention; the overall evaluation value remains within the range of [69, 96], indicating that the operating condition fluctuations of condenser 3 are relatively mild; the data labels clearly identify the value of each point, facilitating the analysis of fluctuation patterns and trend judgment. Figure 3 It can be used to intuitively evaluate the execution effect of the heat exchange control strategy of condenser 3 and the adaptability of the system operation status, and is an important visualization support for reinforcement learning strategy optimization and fault early warning.

[0050] In this implementation scheme, a calculation model coupling multiple parameters such as cooling water density, cooling water flow rate, cooling water specific heat capacity, cooling coil inlet and outlet water temperature difference, saturated steam pressure of the housing, housing air pressure, chip temperature and target chip safe temperature is constructed to quantitatively evaluate the condenser heat exchange adaptation assessment value. Combined with the chip temperature deviation ratio and housing air pressure deviation ratio, a state feature vector that can comprehensively reflect the degree of matching between heat exchange capacity and current heat load is formed. This vector is input into a BP neural network classifier to achieve accurate identification of load state category, improve the accuracy of state judgment before dynamic control of condenser 3, and provide a highly reliable decision-making basis for subsequent reinforcement learning strategy generation.

[0051] Specifically, the steps for inputting the heat exchange adaptation evaluation results, load state categories, and effective condenser operating data into the reinforcement learning decision model and outputting control commands are as follows: Based on historical effective condenser operating data, key parameter sequences including cooling water density, cooling water flow rate, cooling water specific heat capacity, cooling coil inlet and outlet water temperature difference, chip temperature, casing pressure, and pump power are collected to construct a long-term operating state sequence database. The Q-Learning algorithm is used to construct the reinforcement learning decision model. In the model initialization phase, the adjustment amount of the target cooling coil inlet water temperature and the target pump power adjustment amount are used as discrete control variables in the action set. The load state category, condenser heat exchange adaptation evaluation value, and effective condenser operating data including cooling coil inlet water temperature, cooling coil outlet water temperature, cooling water density, cooling water flow rate, cooling water specific heat capacity, pump power, chip temperature, and casing pressure are set as high-dimensional input features in the state set. The composite reward function is to maximize heat dissipation efficiency under high load and minimize unit energy consumption under low load. Iterative training is performed to complete the reinforcement learning decision model training phase. During the strategy reasoning phase, the load state category, condenser heat exchange adaptation evaluation value, and all effective condenser operating data acquired at the current moment are input into the reinforcement learning decision model as decision input parameters, outputting corresponding control commands: Under high load conditions, the inlet water temperature of the target cooling coil is actively reduced while the target pump power is simultaneously increased to enhance the heat exchange intensity of condenser 3; under low load conditions, the inlet water temperature of the target cooling coil is increased while the target pump power is reduced to lower the energy consumption of the cooling system; under medium load conditions, the current inlet water temperature of the target cooling coil and the target pump power are kept constant to maintain stable heat exchange efficiency. The target cooling coil inlet water temperature is the ideal cooling water inlet temperature derived from the current load state and heat exchange requirements. It is used to guide the cooling system to dynamically adjust the cooling intensity, ensuring a balance between chip heat dissipation requirements and energy consumption, and serves as a key control variable for the control commands. The target pump power is a pump operating power setpoint calculated by combining the condenser 3 operating state, heat exchange adaptation degree, and energy consumption optimization target. It is used to adjust the cooling water circulation rate, control the overall energy efficiency level of the system, and compare it with the actual power in the feedback evaluation to guide the subsequent optimization of the control strategy.

[0052] In this implementation scheme, by jointly inputting the condenser 3 heat exchange adaptation evaluation results, load state categories, and effective condenser operating data into a reinforcement learning decision model, dynamic output of the target cooling coil inlet water temperature adjustment and the target pump power adjustment is achieved, constructing a closed-loop control mechanism of perception-decision-execution. This mechanism, based on the Q-Learning algorithm, fully mines the behavioral patterns of key variables such as cooling water density, cooling water flow rate, cooling water specific heat capacity, cooling coil inlet and outlet water temperature difference, chip temperature, casing pressure, and pump power contained in historical effective condenser operating data. With the reward orientation of maximizing heat dissipation efficiency under high load and minimizing energy consumption under low load, a decision model with optimal strategy convergence capability is trained. Compared to traditional fixed control logic that relies on experience, this achieves adaptive control output for condenser 3 operating conditions under different load states, significantly improving the energy efficiency of condenser 3 operation and the flexibility of the control strategy.

[0053] Specifically, the steps for executing control commands and collecting execution feedback data are as follows: Upon receiving the control command, the adjustment amount of the target cooling coil inlet water temperature is sent to the cooling water temperature control execution unit, driving the unit to adjust the temperature based on the target value, thus achieving dynamic control of the cooling coil inlet water temperature; the target pump power adjustment amount is sent to the pump speed control unit, enabling the unit to adjust the pump's operating state according to the target power, achieving precise adjustment of the pump power output. During the execution of the control command, the actual values ​​of the cooling coil inlet water temperature, cooling coil outlet water temperature, and pump power are collected in real time after execution. Simultaneously, the start and end times of the command response are recorded, and the actual execution response time is calculated. Energy consumption data recorded by the power metering unit is simultaneously acquired within the control cycle as the actual total energy consumption. An execution feedback dataset is constructed based on the actual values ​​of the cooling coil inlet water temperature, cooling coil outlet water temperature, pump power, actual execution response time, and actual total energy consumption, providing a complete data foundation for subsequent performance evaluation of the condenser 3 heat exchanger.

[0054] In this implementation scheme, the adjustment amount of the target cooling coil inlet water temperature is precisely sent to the cooling water temperature control execution unit, and the target pump power adjustment amount is simultaneously sent to the pump speed control unit, ensuring efficient response of the control commands at the cooling coil inlet water temperature and pump power levels. During the control execution process, the actual values ​​of the cooling coil inlet water temperature, cooling coil outlet water temperature, pump power, actual execution response time, and actual total energy consumption within the control cycle are collected in real time to construct an execution feedback dataset. This feedback dataset constitutes the necessary input source for subsequent performance evaluation of the condenser 3 heat exchange, providing a basis for dynamically analyzing the cooling coil inlet water temperature adjustment effect and pump power control accuracy, and laying the foundation for judging the execution quality of the cooling control strategy, thereby improving the thermal management adaptability of the condenser 3 under varying load conditions.

[0055] Specifically, the steps for evaluating the heat exchange performance of condenser 3 are as follows: Based on the execution feedback dataset, extract the values ​​of cooling water density, cooling water flow rate, and cooling water specific heat capacity. Multiply the cooling water density by the cooling water flow rate, then by the cooling water specific heat capacity to calculate the total cooling water heat capacity flow rate. Multiply this by the temperature difference between the actual inlet and outlet water temperatures of the cooling coil after execution to obtain the heat exchange power component. Calculate the ratio of the actual execution response time to the reference response time, and add one to this ratio to obtain the response time correction term, which reflects the dynamic delay in the control response process. The reference response time is a reference value representing the ideal time required for the actuator to complete the control command under standard operating conditions, used to measure whether the current control response is lagging or excessive. The speed is measured and used as an important parameter for evaluating execution performance, guiding the dynamic correction of the execution speed of the control strategy. The absolute value of the difference between the actual value of the cooling coil inlet water temperature after execution and the target cooling coil inlet water temperature, as well as the absolute value of the difference between the actual value of the pump power after execution and the target pump power, are calculated separately. The two absolute differences are added together and one is added to obtain the execution deviation correction term, which is used to measure the degree of joint deviation between the cooling water temperature control accuracy and the pump power control accuracy. The total corrected energy consumption is calculated by multiplying the actual total energy consumption, the response time correction term, and the execution deviation correction term within the control cycle, which is used to comprehensively reflect the overall energy efficiency performance during the control process. Finally, the heat exchange power component is divided by the total corrected energy consumption to obtain the heat exchange execution performance evaluation value of condenser 3 in the current control cycle, which serves as an important performance indicator for feedback on the effect of the control strategy.

[0056] The specific formula for calculating the heat exchange performance evaluation value is as follows:

[0057] ;

[0058] In the formula, X represents the heat exchange performance evaluation value. Let Q represent the density of the cooling water, Q represent the flow rate of the cooling water, and C represent the specific heat capacity of the cooling water. This indicates the actual value of the cooling coil inlet water temperature after execution. This represents the actual value of the cooling coil outlet water temperature after execution, and W represents the actual total energy consumption during the control cycle. Indicates the actual execution response time. Indicates the baseline response time. This indicates the target cooling coil inlet water temperature, and G represents the actual pump power value after execution. Indicates the target pump power.

[0059] This implementation plan constructs a heat exchange performance evaluation mechanism with core variables including cooling water density, cooling water flow rate, cooling water specific heat capacity, actual cooling coil inlet water temperature after execution, actual cooling coil outlet water temperature after execution, actual execution response time, baseline response time, target cooling coil inlet water temperature, actual cooling coil inlet water temperature after execution, target pump power, actual pump power after execution, and actual total energy consumption within the control cycle. This effectively achieves dynamic performance quantitative analysis of the control execution process. The method comprehensively evaluates the control response efficiency and execution accuracy by calculating the heat exchange power component, response time correction term, and execution deviation correction term. Using the total corrected energy consumption as the energy efficiency benchmark, it ultimately obtains the heat exchange performance evaluation value. This provides a quantifiable, feedback-able, and closed-loop evaluation basis for subsequent judgment and optimization of control strategies, significantly enhancing the accuracy and adaptability of control decisions.

[0060] Specifically, the steps for determining whether to adjust the current control command based on the evaluation results, and for transmitting the execution feedback data and evaluation results back for threshold updates and control strategy optimization, are as follows: Real-time comparison of the heat exchange performance evaluation value and the performance threshold; at the end of each control cycle, the current heat exchange performance evaluation value is analyzed based on the performance threshold standard; when the heat exchange performance evaluation value is greater than or equal to the performance threshold, the current control strategy is deemed to meet the operational requirements in terms of control accuracy, response speed, and energy efficiency, and the current control strategy is maintained to continue execution in the next control cycle; when the heat exchange performance evaluation value is less than the performance threshold, based on the absolute difference between the actual value of the cooling coil inlet water temperature after execution and the target cooling coil inlet water temperature, and the absolute difference between the actual value of the pump power after execution and the target pump power, recorded centrally in the execution feedback data, the cooling coil inlet water temperature deviation and pump power deviation are constructed. Rate deviation; based on the sign characteristics and numerical amplitude of the cooling coil inlet water temperature deviation and pump power deviation, the current control offset direction and correction degree are calculated, and the corrected cold target cooling coil inlet water temperature adjustment amount and the corrected target pump power adjustment amount are generated. The above control parameters are combined into a correction control command and sent to the actuator to drive the cooling water temperature control execution unit and the pump speed control unit to perform correction adjustment. Regardless of whether strategy correction is performed, the execution feedback dataset collected in the current cycle and the calculated heat exchange performance evaluation value are synchronously transmitted back to the data filtering stage and the intelligent decision stage. In the data filtering stage, the chip temperature threshold, the box air pressure threshold and the cooling water flow threshold are corrected in real time. In the intelligent decision stage, the state behavior mapping relationship corresponding to the cold target cooling coil inlet water temperature adjustment strategy and the target pump power adjustment strategy is updated to realize a multi-dimensional dynamic heat exchange control closed-loop operation mechanism for the operating state of condenser 3.

[0061] In this implementation scheme, a dynamic judgment and correction mechanism for the effect of control commands is realized based on the real-time comparison of heat exchange performance evaluation values ​​and performance thresholds, ensuring the accuracy of control strategies in adjusting cooling coil inlet water temperature and setting target pump power. By constructing corrected control commands based on the deviations in cooling coil inlet water temperature and pump power in the execution feedback data, and sending the execution feedback data and heat exchange performance evaluation values ​​back to the data filtering and intelligent decision-making stages, the control loop is updated with chip temperature thresholds, housing air pressure thresholds, cooling water flow thresholds, and optimized cooling coil inlet water temperature adjustment strategies and target pump power adjustment strategies, respectively. This achieves continuous adaptive correction and enhanced accuracy of the control loop, improving the stability, robustness, and energy efficiency of heat exchange control.

[0062] like Figure 2 As shown, the second aspect of this invention provides a condenser dynamic heat exchange control system based on multi-dimensional perception and AI, including: a data perception preprocessing module, an anomaly screening emergency control module, an intelligent state judgment decision generation module, and an execution feedback closed-loop optimization module. The data perception preprocessing module collects condenser heat exchange perception data and performs time synchronization, noise separation, physical consistency verification, and normalization mapping processing on the condenser heat exchange perception data. The anomaly screening emergency control module performs two rounds of data screening on the sampling points based on the preprocessed condenser heat exchange perception data, enabling the issuance of emergency control commands and the processing of anomaly point data, thus constructing an effective condenser dynamic heat exchange control system. The system includes: a condenser operation dataset; an intelligent state judgment and decision generation module, used to evaluate the heat exchange adaptation of condenser 3 based on the effective condenser operation dataset, extract the chip temperature deviation ratio and the box air pressure deviation ratio, and identify the load state category; inputting the heat exchange adaptation evaluation results, load state category, and effective condenser operation data into the reinforcement learning decision model, and outputting control commands; and an execution feedback closed-loop optimization module, used to execute control commands and collect execution feedback data, evaluate the heat exchange performance of condenser 3, determine whether the current control command needs to be adjusted based on the evaluation results, and send the execution feedback data and evaluation results back for threshold updates and control strategy optimization.

[0063] like Figure 4As shown, saturated vapor pressure sensor 1 is used to collect the saturated vapor pressure of the enclosure, forming an important component of the condenser heat exchange sensing data. Cooling coil 2 is located inside condenser 3, responsible for heat exchange between the coolant inlet and outlet, and adjusting the heat exchange efficiency of condenser 3. Condenser 3 is positioned above the heat dissipation path of server 4, condensing the vapor generated by the chip into a liquid state in a timely manner, effectively releasing heat. Server 4 is the main heat source device; its chip continuously releases heat during operation, creating a heat exchange demand on condenser 3. Temperature sensors 5 are deployed at the inlet and outlet of cooling coil 2, collecting the inlet and outlet water temperatures to calculate the temperature difference between the inlet and outlet. Pump 6 drives the continuous circulation of coolant and adjusts the pump power according to control commands to achieve precise control of the target pump power. Coolant reservoir 7 stores coolant, ensuring the liquid supply and circulation stability of the cooling circuit.

[0064] In this implementation plan, a dynamic heat exchange control system for condensers is constructed, comprising a data sensing preprocessing module, an anomaly screening and emergency control module, an intelligent state judgment and decision generation module, and an execution feedback closed-loop optimization module. This system systematically covers the entire process of condenser heat exchange sensing data processing, anomaly identification and emergency response, intelligent identification of operating status and generation of control commands, execution effect feedback evaluation, and continuous strategy optimization. Each module has clearly defined functional boundaries. After time-series synchronization, noise separation, physical consistency verification, and normalization mapping, the condenser heat exchange sensing data enters the data filtering mechanism to form an effective condenser operating dataset. This dataset then undergoes load state identification and heat exchange adaptation evaluation, driving a reinforcement learning decision model to generate control commands. Finally, through heat exchange performance evaluation and execution feedback data feedback, the control strategy is continuously optimized, ensuring the accuracy, adaptability, and closed-loop robustness of the dynamic heat exchange control of condenser 3.

[0065] 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.

[0066] 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. Clearly, 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 dynamic heat transfer control method for condensers based on multi-dimensional sensing and AI, characterized in that, Includes the following steps: S1, collect condenser heat exchange sensing data, and perform time synchronization, noise separation, physical consistency verification and normalization mapping on the condenser heat exchange sensing data; The specific steps for collecting condenser heat exchange sensing data and performing time-series synchronization, noise separation, physical consistency verification, and normalization mapping on the condenser heat exchange sensing data are as follows: Collect condenser heat exchange sensing data, including chip temperature, chamber pressure, cooling coil inlet water temperature, cooling coil outlet water temperature, cooling water density, cooling water flow rate, cooling water specific heat capacity, pump power, and chamber saturated steam pressure. This paper employs a multi-channel timestamp comparison and sampling delay compensation method to perform cross-sensor time synchronization and unified sampling rate processing on condenser heat transfer sensing data. A multi-scale signal reconstruction method combining Kalman filtering and wavelet decomposition is used to eliminate measurement noise and separate periodic disturbances in the condenser heat transfer sensing data. A heat-flow integration consistency verification mechanism is established based on the energy conservation equation and the pressure-flow coupling model to identify and remove data from the condenser heat transfer sensing data that does not conform to physical laws. Data format standardization and interface compatibility conversion are performed on the condenser heat transfer sensing data through multi-dimensional feature encoding and adaptive encapsulation of transmission protocols. Finally, unit conversion and feature scale normalization are performed on the condenser heat transfer sensing data using physical quantity normalization mapping and dynamic interval stretching algorithms. S2, based on the pre-processed condenser heat exchange sensing data, performs two rounds of data filtering on the sampling points to realize the issuance of emergency control commands and the processing of abnormal point data, and constructs an effective condenser operation dataset; The specific steps for constructing an effective condenser operation dataset by performing two rounds of data filtering on sampling points based on preprocessed condenser heat exchange sensing data to achieve emergency control command issuance and anomaly data processing are as follows: Extract the pre-processed condenser heat exchange sensing data and perform the first round of data screening for each sampling point: when the chip temperature exceeds the upper limit threshold of chip temperature, the box pressure exceeds the upper limit threshold of pressure, or the cooling water flow rate is lower than the lower limit threshold of flow rate, the data is judged to be abnormal and emergency control is triggered. The inlet water temperature of the cold target cooling coil and the target pump power are adjusted to the emergency set value and sent to the actuator. The remaining sampling points enter the second screening. For the sampling points that enter the second screening, the temperature difference between the inlet and outlet of the cooling coil is calculated based on the difference between the inlet water temperature and the outlet water temperature of the cooling coil. The mean and standard deviation of the chip temperature, the box pressure, the cooling water flow rate and the temperature difference between the inlet and outlet of the cooling coil are calculated in the sliding window to assess the degree of abnormality of the condenser (3) and obtain the abnormality assessment value of the condenser. The system compares the condenser operating condition anomaly assessment value with the anomaly threshold in real time. When the condenser operating condition anomaly assessment value exceeds the anomaly scoring threshold, the corresponding sampling point is identified as an anomaly point. If it is an isolated anomaly point, the sliding average value of the condenser heat exchange sensing data from adjacent sampling points is used instead. If three or more consecutive sampling points are identified as anomaly points and the corresponding parameter acquisition source is the same sensor, the system switches to the backup sensor to collect the value. Otherwise, the corresponding sampling point is identified as a normal point, and the original data is retained. The condenser heat exchange sensing data retained after two rounds of data filtering are used to construct an effective condenser operation dataset according to the sampling point order. S3, evaluate the heat exchange adaptation of condenser (3) based on the effective condenser operation dataset, extract the chip temperature deviation ratio and the box pressure deviation ratio, and identify the load state category; input the heat exchange adaptation evaluation results, load state category and effective condenser operation data into the reinforcement learning decision model, and output the control command; S4, execute the control command and collect the execution feedback data, evaluate the heat exchange performance of the condenser (3), determine whether the current control command needs to be adjusted based on the evaluation results, and send back the execution feedback data and evaluation results for threshold update and control strategy optimization.

2. The condenser dynamic heat exchange control method based on multi-dimensional perception and AI according to claim 1, characterized in that: The specific steps for assessing the degree of abnormality in condenser operating conditions and obtaining the assessment value for abnormal condenser operating conditions are as follows: The temperature deviation term is obtained by dividing the square of the difference between the chip temperature and the mean chip temperature by the square of the standard deviation of the chip temperature plus one; the pressure deviation term is obtained by dividing the square of the difference between the chamber air pressure and the mean chamber air pressure by the square of the standard deviation of the chamber air pressure plus one; the flow deviation term is obtained by dividing the square of the difference between the cooling water flow rate and the mean cooling water flow rate by the square of the standard deviation of the cooling water flow rate plus one; the temperature difference deviation term is obtained by dividing the square of the difference between the inlet and outlet water temperature of the cooling coil and the mean by the square of the standard deviation of the temperature difference plus one. The sum of the temperature deviation, pressure deviation, flow rate deviation, and temperature difference deviation is added together and the square root is taken to obtain the normalized deviation sum of multiple parameters. The difference between the chip temperature and the target chip safe temperature is multiplied by the difference between the chamber pressure and the chamber saturated vapor pressure, and the absolute value is divided by the product of the target chip safe temperature and the chamber saturated vapor pressure to obtain the temperature and pressure interaction correction term. One is added to the temperature and pressure interaction correction term to obtain the correction factor. The normalized deviation sum of multiple parameters is multiplied by the correction factor to obtain the condenser operating condition anomaly assessment value.

3. The condenser dynamic heat exchange control method based on multi-dimensional perception and AI according to claim 1, characterized in that: The specific steps for evaluating the heat exchange adaptability of condenser (3) based on the effective condenser operation dataset, extracting the chip temperature deviation ratio and the box air pressure deviation ratio, and identifying the load state category are as follows: Extract the effective condenser operation dataset and evaluate the heat transfer adaptability of condenser (3): multiply the cooling water density, cooling water flow rate and cooling water specific heat capacity by the temperature difference between the inlet and outlet water of the cooling coil to obtain the heat transfer power component; take the larger value between the difference between the saturated steam pressure and the gas pressure of the box and the minimum term to obtain the pressure margin component; divide the heat transfer power component by the pressure margin component to obtain the pressure-corrected heat transfer power; divide the difference between the chip temperature and the target chip safe temperature by the target chip safe temperature to obtain the relative temperature deviation coefficient; multiply the pressure-corrected heat transfer power by the relative temperature deviation coefficient to obtain the condenser heat transfer adaptability evaluation value. Extract the average chip temperature and the average chamber pressure, and compare them with the chip temperature threshold and the chamber pressure threshold, respectively, to obtain the chip temperature deviation ratio and the chamber pressure deviation ratio; The condenser heat exchanger adaptation evaluation value, chip temperature deviation ratio, and box air pressure deviation ratio are combined to form a feature vector, which is then input into a BP neural network classifier to output the current load status category, including high load, medium load, and low load.

4. The condenser dynamic heat exchange control method based on multi-dimensional perception and AI according to claim 1, characterized in that: The specific steps for inputting the heat exchanger adaptation evaluation results, load state category, and effective condenser operation data into the reinforcement learning decision model and outputting control commands are as follows: Based on historical effective condenser operation data, a reinforcement learning decision model is constructed using the Q-Learning algorithm. The adjustment amount of the target cooling coil inlet water temperature and the target pump power are set as the action set, and the load state category, condenser heat exchange adaptation evaluation value, and effective condenser operation data are set as the state set. The reward function is to maximize heat dissipation efficiency under high load and minimize energy consumption under low load to complete the training of the reinforcement learning decision model. The current load state category, condenser heat exchange adaptation evaluation value, and effective condenser operation data are used as decision input parameters to input the reinforcement learning decision model, and the output control instructions are: under high load, reduce the target cooling coil inlet water temperature and increase the target pump power; under low load, increase the target cooling coil inlet water temperature and reduce the target pump power; under medium load, maintain the current target cooling coil inlet water temperature and target pump power.

5. The condenser dynamic heat exchange control method based on multi-dimensional perception and AI according to claim 1, characterized in that: The specific steps for executing control commands and collecting execution feedback data are as follows: Upon receiving control commands, the system sends the adjustment amount of the target cooling coil inlet water temperature to the cooling water temperature control execution unit and the adjustment amount of the target pump power to the pump speed control unit, and drives the control execution to perform real-time adjustment. During the control execution process, the system collects the actual values ​​of the cooling coil inlet water temperature, the actual values ​​of the cooling coil outlet water temperature, the actual values ​​of the pump power, the actual execution response time, and the actual total energy consumption within the control cycle after execution, and constructs an execution feedback dataset.

6. The condenser dynamic heat exchange control method based on multi-dimensional perception and AI according to claim 1, characterized in that: The specific steps for evaluating the heat exchange performance of the condenser are as follows: Based on the execution feedback dataset, the heat exchange power component is obtained by multiplying the cooling water density, cooling water flow rate, and cooling water specific heat capacity together, and then multiplying by the difference between the actual inlet water temperature and the actual outlet water temperature of the cooling coil after execution. The response time is then divided by the baseline response time and incremented by one to obtain the response time correction term. The absolute values ​​of the difference between the actual inlet water temperature and the target cooling coil after execution, and the absolute values ​​of the difference between the actual pump power and the target pump power after execution are calculated. The two absolute values ​​are then added together and incremented by one to obtain the execution deviation correction term. The total energy consumption, response time correction term, and execution deviation correction term within the control cycle are multiplied together to obtain the total corrected energy consumption. Finally, the heat exchange power component is divided by the total corrected energy consumption to obtain the heat exchange performance evaluation value.

7. The condenser dynamic heat exchange control method based on multi-dimensional perception and AI according to claim 1, characterized in that: The specific steps for determining whether to adjust the current control command based on the evaluation results, and for sending back the execution feedback data and evaluation results for threshold updating and control strategy optimization, are as follows: The system compares the heat exchange performance evaluation value with the performance threshold in real time. When the heat exchange performance evaluation value is greater than or equal to the performance threshold, the current control strategy is maintained for the next control cycle. When the heat exchange performance evaluation value is less than the performance threshold, the system determines the cooling coil inlet water temperature deviation and pump power deviation based on the correspondence between the actual value of the cooling coil inlet water temperature after execution and the target cooling coil inlet water temperature, and the actual value of the pump power after execution and the target pump power, as recorded in the execution feedback dataset. Based on the direction and magnitude of the deviation, the system generates corrected instructions for the cold target cooling coil inlet water temperature and the target pump power, and sends them to the actuator. Regardless of whether the strategy is corrected, the execution feedback dataset and the heat exchange performance evaluation value are returned to the data filtering stage and the intelligent decision-making stage, respectively. The temperature threshold, air pressure threshold, and flow threshold are updated in the data filtering stage, and the cold target cooling coil inlet water temperature adjustment strategy and the target pump power adjustment strategy are optimized in the intelligent decision-making stage to achieve closed-loop operation of dynamic heat exchange control.

8. A condenser dynamic heat exchange control system based on multi-dimensional sensing and AI, employing the condenser dynamic heat exchange control method based on multi-dimensional sensing and AI as described in any one of claims 1-7, characterized in that: include: The module comprises a data perception and preprocessing module, an anomaly screening and emergency control module, an intelligent state judgment and decision generation module, and an execution feedback closed-loop optimization module, among which: The data sensing preprocessing module is used to collect condenser heat exchange sensing data and perform time synchronization, noise separation, physical consistency verification and normalization mapping processing on the condenser heat exchange sensing data. The anomaly screening and emergency control module is used to perform two rounds of data screening on the sampling points based on the pre-processed condenser heat exchange sensing data, realize the issuance of emergency control commands and the processing of anomaly point data, and construct an effective condenser operation dataset. The intelligent state judgment decision generation module is used to evaluate the heat exchange adaptation degree of the condenser (3) based on the effective condenser operation dataset, extract the chip temperature deviation ratio and the box air pressure deviation ratio, and identify the load state category; input the heat exchange adaptation evaluation results, load state category and effective condenser operation data into the reinforcement learning decision model, and output the control command; The execution feedback closed-loop optimization module is used to execute control commands and collect execution feedback data, evaluate the heat exchange performance of the condenser (3), determine whether the current control command needs to be adjusted based on the evaluation results, and send back the execution feedback data and evaluation results for threshold update and control strategy optimization.

Images