Direct liquid cooling temperature control machine flow intelligent correction method and system

Abstract

This invention discloses a method and system for intelligent flow correction in a direct liquid-cooled temperature controller. The method includes: collecting and preprocessing multi-dimensional parameters from the circulating liquid, load, equipment, and environment, then inputting them into a prediction model to output a future liquid temperature trend curve and the theoretical optimal flow demand; fusing the anticipatory and real-time liquid temperature deviations to obtain coupled liquid temperature deviations and rates of change, and introducing fuzzy inference based on circulating liquid thermal properties, load power, and ambient temperature change factors to correct PID parameters and generate basic instructions; determining abnormal operating conditions, calculating compensation increments, and coupling the basic instructions and compensation increments to generate execution instructions to drive the pump and valves to regulate flow. The system includes six modules: data acquisition, prediction model, fuzzy PID, abnormal compensation, instruction coupling, and closed-loop feedback. This invention achieves anticipatory prediction, dynamic correction, abnormal compensation, and closed-loop iterative collaborative control, eliminating regulation lag, enhancing anti-interference capabilities, improving adaptability to abnormal operating conditions, and realizing precise, rapid, and stable flow regulation.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for direct liquid-cooled temperature control equipment, and in particular to an intelligent flow correction system and method for a direct liquid-cooled temperature controller. Background Technology

[0002] Direct liquid-cooled temperature controllers exchange heat with the customer's load through a closed-loop circulating fluid, making them core equipment for high-precision load thermal management. Their flow control accuracy directly determines the stability of the load's operating temperature. Currently, the mainstream flow control solution in the industry adopts a "real-time liquid temperature single-parameter feedback + fixed PID parameter adjustment" mode. This relies on real-time acquired liquid temperature signals to drive pump speed and valve opening adjustments to achieve closed-loop flow control. However, this control method relies solely on real-time liquid temperature feedback for adjustment, lacking proactive predictive capabilities. When load thermal characteristics fluctuate, the adjustment often lags behind liquid temperature changes, easily leading to repeated fluctuations of "overshoot-undershoot." Simultaneously, the density, specific heat capacity, viscosity, and other thermophysical parameters of the circulating fluid dynamically change with temperature and usage time. Fixed PID parameters cannot adapt to these changes, resulting in flow fluctuations and reduced heat exchange efficiency. Furthermore, when facing abnormal operating conditions such as heat exchanger scaling or sudden changes in load power, the lack of active compensation logic means the system can only passively adjust based on feedback, easily causing the liquid temperature to exceed the set range, threatening the load's operational safety.

[0003] Existing patent CN111163615B discloses an external liquid cooling system for a data center switch, including a switch enclosure with coolant pipes on both sides. These coolant pipes are connected to a water tank via a water pump. A motherboard is housed within the switch enclosure, and a device to be cooled is mounted on the motherboard. A copper pipe is mounted on the device, with one end of the copper pipe in contact with the coolant pipe and the other end passing through the switch enclosure and contacting the surface of the device to transfer heat from the device to the coolant pipe. However, this technical solution only adjusts based on real-time temperature deviation, lacking a predictive mechanism, resulting in significant adjustment lag. Its PID parameters are not dynamically corrected based on the drift of the circulating liquid's thermal properties, leading to weak anti-interference capabilities. Furthermore, it lacks a mechanism for identifying and actively compensating for abnormal operating conditions such as scaling and sudden load changes, passively adjusting according to temperature changes, which can easily cause the liquid temperature to exceed the set range under abnormal conditions. Summary of the Invention

[0004] Existing direct liquid-cooled temperature controllers suffer from problems such as significant adjustment lag, weak anti-interference capability, and poor adaptability to abnormal operating conditions.

[0005] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows: A method for intelligent flow correction in a direct liquid-cooled temperature controller includes the following steps: S1. Collect multi-dimensional parameters of circulating fluid, load, equipment and environment, and input them into the multi-parameter coupled prediction model after preprocessing to output the future liquid temperature trend curve and the theoretical optimal flow rate requirement. S2. Calculate the adaptive weighting coefficient to fuse the advanced liquid temperature deviation and the real-time liquid temperature deviation, and obtain the coupled liquid temperature deviation and its rate of change; at the same time, introduce the circulating liquid thermal properties, load power and ambient temperature change factors, and correct the proportional, integral and differential coefficients through fuzzy inference to generate the basic speed and opening command. S3. Determine if the current condition is abnormal. If so, calculate the corresponding compensation increment. Couple the basic command with the compensation increment to generate the final execution command, which drives the water pump and valve to regulate the flow.

[0006] Further improvements include the collection of multi-dimensional parameters in S1, such as circulating fluid supply and return fluid temperature, pipeline pressure difference, load power, and ambient temperature and humidity; preprocessing includes filtering out outliers using reasonable range thresholds, extracting the heat exchanger temperature difference change rate and load power change rate as coupling features, and mapping non-dimensional parameters to a unified numerical range for normalization.

[0007] A further improvement is that in S2, the advanced liquid temperature deviation is the set liquid temperature minus the predicted liquid temperature, and the real-time liquid temperature deviation is the set liquid temperature minus the actual liquid temperature; the coupled liquid temperature deviation is calculated in the following way: the product of the adaptive weighting coefficient and the advanced liquid temperature deviation is added to the product of the difference between the adaptive weighting coefficient and the real-time liquid temperature deviation, thereby fusing the deviation information of the advanced and real-time.

[0008] Further improvements are made in S2, where the coupling liquid temperature deviation and deviation change rate are fuzzified and converted into a fuzzy subset, and the circulating liquid thermal property change factor, load power change factor and ambient temperature change factor are introduced to optimize the weight allocation in the inference process; after fuzzy rule inference and declarative processing, the correction amount of the proportional, integral and differential coefficients is output; the corrected proportional coefficient is equal to the initial reference proportional coefficient plus the calculated proportional coefficient correction amount.

[0009] Further improvements are made in S3. If the temperature difference between the inlet and outlet of the heat exchanger is detected to be increasing continuously and exceeds the preset trend threshold, it is determined to be a heat exchanger scaling condition. If the sudden change in load power exceeds the preset proportional threshold, it is determined to be a load power sudden change condition. Under normal operating conditions, the compensation increment is set to zero. Under abnormal operating conditions, the compensation increment of the pump speed and valve opening is calculated based on the degree of parameter deviation, and the compensation increment is positively correlated with the degree of parameter deviation.

[0010] Further improvements include closed-loop feedback and iterative steps: real-time acquisition of adjusted liquid temperature and flow rate, comparison with the set liquid temperature range; if the liquid temperature exceeds the set range, the deviation data is fed back to the model and correction module for parameter fine-tuning; when the preset iteration cycle is reached, the effective data accumulated in the historical database is re-inputted into the prediction model for iterative training, and the fuzzy inference rule base and anomaly judgment rule base are updated simultaneously.

[0011] The present invention also provides a direct liquid-cooled temperature controller flow intelligent correction system, including a data acquisition module, a multi-parameter coupled prediction model module, a fuzzy PID intelligent correction module, an abnormal operating condition compensation module, an instruction coupled execution module, and a closed-loop feedback and iteration module.

[0012] The data acquisition module collects and preprocesses multi-dimensional parameters from the circulating fluid, load, equipment, and environment. The multi-parameter coupled prediction model module receives the preprocessed data based on a pre-trained model and outputs the future liquid temperature trend and the theoretical optimal flow rate requirement. The fuzzy PID intelligent correction module receives the prediction results, calculates the coupled liquid temperature deviation and rate of change, and, combined with adjustment factors such as circulating fluid thermophysical properties, load power, and ambient temperature, corrects the PID parameters through fuzzy inference to generate basic control commands. The abnormal operating condition compensation module monitors the system status, determines whether scaling or sudden load changes occur based on the rule base, and calculates the corresponding compensation increment. The command coupling execution module couples the basic commands with the compensation increment to generate the final execution commands to drive the pumps and valves. The closed-loop feedback and iteration module collects feedback data after execution to achieve real-time fine-tuning of control parameters and periodic updates to the model rule base.

[0013] The beneficial effects of this invention are as follows: Eliminating regulation lag: By using a multi-parameter coupled prediction model to predict the liquid temperature change trend in advance and generate the theoretical optimal flow demand, the flow rate can be adjusted in advance, which significantly shortens the liquid temperature stabilization response time and reduces the overshoot. Enhanced anti-interference capability: The PID parameters are dynamically fuzzy corrected by introducing circulating liquid thermal properties, load power and ambient temperature change factors, so that the control system can adapt to changes in fluid properties and environmental fluctuations, effectively suppressing flow fluctuations and improving heat exchange efficiency and temperature control accuracy. Enhanced adaptability to anomalies: Built-in abnormal operating condition judgment and active compensation logic can actively calculate the flow compensation increment for scenarios such as heat exchanger fouling and sudden changes in load power, quickly respond and offset the decrease in heat exchange efficiency or thermal shock, and ensure the safe operation of the load under extreme conditions. It has self-optimization capabilities: Through closed-loop feedback and periodic iteration mechanisms, it continuously optimizes the prediction model and rule base using newly added operating data, enabling the system to adapt to the performance degradation during long-term operation of equipment, reducing maintenance costs and extending the service life of equipment. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention.

[0015] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] Example 1: This example provides a method and system for intelligent flow correction of a direct liquid-cooled temperature controller. The direct liquid-cooled temperature controller exchanges heat with the customer's load through a closed-loop circulating liquid, making it a core device for high-end precision load thermal management. Currently, the mainstream flow control scheme in the industry adopts a "real-time liquid temperature single-parameter feedback + fixed PID parameter adjustment" mode, relying on real-time liquid temperature signals to drive pump speed and valve opening adjustments to achieve closed-loop flow control. This solution suffers from three major problems: First, significant lag in regulation. Relying solely on real-time liquid temperature feedback, it lacks proactive prediction capabilities. When load thermal characteristics fluctuate, the adjustment lags behind liquid temperature changes, easily leading to repeated overshoot and undershoot fluctuations, failing to quickly stabilize the liquid temperature, and affecting load operational stability. Second, weak anti-interference capability. The density, specific heat capacity, viscosity, and other thermophysical parameters of the circulating fluid dynamically change with temperature and usage time. Traditional fixed PID parameters cannot adapt to these changes, resulting in flow fluctuations, which in turn reduce heat exchange efficiency and compromise temperature control accuracy. Third, poor adaptability to abnormal operating conditions. Lacking proactive compensation logic, it can only passively adjust based on feedback when facing abnormal conditions such as heat exchanger fouling leading to decreased heat exchange efficiency or sudden changes in load power. This can easily cause the liquid temperature to exceed the set range, threatening load safety. To address these technical deficiencies, this embodiment proposes a flow intelligent correction scheme with proactive prediction, dynamic parameter correction, and proactive compensation capabilities for abnormal operating conditions. Through multi-parameter coupled control, it improves the temperature control accuracy, response speed, and operational stability of the direct liquid-cooled temperature controller, adapting to the needs of high-end precision loads.

[0018] The intelligent flow correction method for the direct liquid-cooled temperature controller in this embodiment includes the following steps: S1: Collect and preprocess multi-dimensional parameters of circulating fluid, load, equipment and environment, input them into the prediction model, and output the future liquid temperature trend curve and theoretical optimal flow demand; S2: Calculate the adaptive weighting coefficient to fuse the advanced liquid temperature deviation and the real-time liquid temperature deviation to obtain the coupled liquid temperature deviation and the rate of change. Combine the circulating liquid thermal properties, load power and ambient temperature change factor fuzzy reasoning correction ratio, integral and differential coefficients to generate the basic speed and opening command. S3: Determine abnormal operating conditions, calculate compensation increments, and couple basic instructions with compensation increments to generate execution instructions to drive water pumps and valves to regulate flow.

[0019] In step S1, the data acquisition module collects multi-dimensional parameters such as circulating fluid supply and return temperatures, pipeline pressure difference, real-time flow rate, load power, and ambient temperature and humidity at a preset frequency. Preprocessing includes: filtering out outliers using reasonable threshold ranges; extracting the heat exchanger temperature difference change rate and load power change rate as coupling features; and normalizing non-dimensional parameters by mapping them to a unified numerical range. The preprocessed data is then input into a multi-parameter coupled prediction model. This model is pre-trained based on historical liquid temperature, flow rate, and operating condition correlation data. It employs a time-series neural network structure such as LSTM or GRU, and through multi-parameter coupled inference, outputs the liquid temperature change trend curve for a preset future time period and the theoretical optimal flow rate demand corresponding to each time node, achieving advanced prediction. The input dimensions of this prediction model cover the preprocessed real-time multi-parameter features and historical time-series correlation data. Internally, it constructs a multi-parameter coupled mapping function, comprehensively considering the heat load, environmental heat dissipation effects, and the thermophysical properties of the fluid medium, thereby accurately calculating the optimal flow rate value that maintains thermal balance.

[0020] In step S2, the advanced liquid temperature deviation is the set liquid temperature minus the predicted liquid temperature, and the real-time liquid temperature deviation is the set liquid temperature minus the actual liquid temperature. The coupled liquid temperature deviation is equal to the adaptive weight coefficient multiplied by the advanced liquid temperature deviation, plus one minus the adaptive weight coefficient multiplied by the real-time liquid temperature deviation. The adaptive weight coefficient is between zero and one, and is dynamically adjusted according to the operating conditions. When the prediction reliability is high, the weight is increased to strengthen the advanced prediction; when the prediction deviation is large, the weight is decreased to strengthen the real-time feedback. The rate of change of the coupled liquid temperature deviation is further calculated, which is the difference between the coupled liquid temperature deviation at the current moment and the coupled liquid temperature deviation at the previous moment divided by the time interval between adjacent data acquisitions. The coupled liquid temperature deviation and its rate of change are fuzzified and converted into fuzzy subsets, including seven levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. At the same time, the circulating liquid thermal property change factor (γ1), the load power change factor (γ2), and the ambient temperature change factor (γ3) are introduced to optimize the weight allocation in the fuzzy inference process. The circulating fluid thermophysical property change factor (γ1) is obtained by dividing the absolute value of the deviation of the circulating fluid density from the standard state density by the standard state density, plus the absolute value of the deviation of the circulating fluid isobaric specific heat capacity from the standard state isobaric specific heat capacity divided by the standard state isobaric specific heat capacity, and summing the results. The load power change factor (γ2) is obtained by dividing the absolute value of the power difference between adjacent acquisition cycles by the rated power. The ambient temperature change factor (γ3) is obtained by dividing the absolute value of the deviation of the ambient temperature from the reference value by the ambient reference temperature, plus the absolute value of the deviation of the ambient humidity from the reference value by the ambient reference humidity, and summing the results. After inference using a preset fuzzy rule base, the centroid method is used to clarify the output of the proportional coefficient correction (ΔKp), integral coefficient correction (ΔKi), and differential coefficient correction (ΔKd). The corrected proportional coefficient (Kp) equals the initial baseline proportional coefficient (Kp0) plus the proportional coefficient correction (ΔKp); the corrected integral coefficient (Ki) equals the initial baseline integral coefficient (Ki0) plus the integral coefficient correction (ΔKi); and the corrected differential coefficient (Kd) equals the initial baseline differential coefficient (Kd0) plus the differential coefficient correction (ΔKd). Based on the theoretically optimal flow rate, the basic pump speed command (n_base) and basic valve opening command (θ_base) are generated.

[0021] In step S3, the abnormal operating condition judgment rule base includes: when the temperature difference between the inlet and outlet of the heat exchanger continuously increases beyond the preset rate of change threshold, it is judged as a scaling condition; when the load power sudden change exceeds the preset proportional threshold, it is judged as a load sudden change condition. Under normal operating conditions, the compensation increment is set to zero; under abnormal operating conditions, the pump speed compensation increment (Δn_comp) and valve opening compensation increment (Δθ_comp) are calculated based on the degree of parameter deviation, and the compensation increment is positively correlated with the degree of deviation. The instruction coupling execution module couples the basic instruction with the compensation increment, that is, the actual pump speed (n_act) is equal to the basic speed instruction (n_base) plus the speed compensation increment (Δn_comp), and the actual valve opening (θ_act) is equal to the basic opening instruction (θ_base) plus the opening compensation increment (Δθ_comp), driving the variable frequency pump and electric regulating valve to quickly adjust the circulating liquid flow rate to the target value.

[0022] After the command is executed, the closed-loop feedback and iteration module collects the real-time liquid temperature and actual flow rate, and compares them with the set liquid temperature range. If the real-time liquid temperature is within the set range, it returns to step S1 for continuous dynamic adjustment; if the real-time liquid temperature exceeds the set range, the deviation data is marked as optimization samples and fed back to the prediction model and fuzzy PID module to fine-tune parameters, optimize the model and rule base. When the preset iteration cycle is reached, valid data from the historical database is input into the prediction model for iterative training, and the fuzzy inference and anomaly detection rule base is updated synchronously to optimize factor weights and improve the long-term adaptability of the system.

[0023] This embodiment also provides a direct liquid-cooled temperature controller flow intelligent correction system for implementing the above method. The system includes a data acquisition module, a multi-parameter coupled prediction model module, a fuzzy PID intelligent correction module, an abnormal operating condition compensation module, an instruction coupled execution module, and a closed-loop feedback and iteration module. The data acquisition module collects and preprocesses multi-dimensional parameters before inputting them into the multi-parameter coupled prediction model module; the multi-parameter coupled prediction model module outputs the future liquid temperature trend curve and the theoretical optimal flow requirement to the fuzzy PID intelligent correction module; the fuzzy PID intelligent correction module calculates the coupled liquid temperature deviation and rate of change, and combines the adjustment factor correction ratio, integral and derivative coefficients to generate basic instructions, which are then sent to the instruction coupled execution module; the abnormal operating condition compensation module calculates the compensation increment and transmits it to the instruction coupled execution module; the instruction coupled execution module couples the basic instructions and the compensation increment to generate execution instructions to drive the water pump and valves. The closed-loop feedback and iteration module collects real-time liquid temperature and actual flow rate, and when the flow rate exceeds the set liquid temperature range, it feeds back the deviation data to the multi-parameter coupled prediction model module and the fuzzy PID intelligent correction module to fine-tune the parameters. When the iteration cycle is reached, the fuzzy inference and abnormal judgment rule base is updated. This embodiment solves the problems of lagging adjustment, weak anti-interference ability and poor anomaly adaptability in the prior art by adopting a four-layer collaborative architecture of advanced prediction, dynamic correction, anomaly compensation and closed-loop iteration.

[0024] Example 2: This example, as a preferred implementation of Example 1, further defines the specific training method, iterative optimization mechanism, and detailed operations of multi-dimensional parameter preprocessing for the multi-parameter coupled prediction model. By refining the data processing flow and model update strategy, it ensures that the prediction model can maintain high-precision output capability over a long period and adapt to performance changes throughout the entire lifecycle of the device.

[0025] In step S1, the multi-dimensional parameters include at least the circulating liquid supply temperature, return liquid temperature, pipeline differential pressure, load power, and ambient temperature and humidity. Preprocessing specifically includes three sub-steps: First, outlier removal, using reasonable threshold filtering to discard data points exceeding the sensor's range or physically impossible ranges; for example, supply temperatures exceeding 100 degrees Celsius or negative flow rates are considered outliers and replaced with values ​​from the previous time step. Second, coupled feature extraction, extracting the heat exchanger temperature difference change rate and load power change rate, i.e., the derivative of the heat exchanger inlet and outlet temperature difference with respect to time and the derivative of the load power with respect to time. Third, non-dimensional parameter normalization, mapping parameters of different dimensions to a uniform zero-to-one range using a mapping function.

[0026] The multi-parameter coupled prediction model is pre-trained based on historical liquid temperature, flow rate, and operating condition correlation data. The training data is collected by continuously recording multi-dimensional parameters during the normal operation of the temperature controller, covering various operating conditions such as steady-state operation, load step changes, and ambient temperature fluctuations. The model uses an LSTM network, with the input being the multi-parameter feature vectors of the past ten time steps and the output being the liquid temperature prediction sequence for the next twenty time steps. Training employs the mean squared error loss function and the Adam optimizer. After pre-training, the model parameters are stored in the embedded controller's storage unit and directly loaded and invoked upon system startup, eliminating the need for repeated training.

[0027] In actual operation, the system uses a historical database to store all collected and preprocessed valid data. When the cumulative runtime reaches a preset iteration cycle or the amount of new data reaches a preset threshold, the system automatically triggers model iterative training: it extracts recent valid data from the historical database, fine-tunes the original model parameters, trains for five to ten rounds with a relatively small learning rate, and replaces the old model with the updated model parameters. Simultaneously, the fuzzy inference rule base and the abnormal operating condition judgment rule base are updated, the weight allocation of adjustment factors is optimized, and the judgment threshold is adjusted. Through this iterative optimization, the system can adapt to changes such as circulating fluid aging, heat exchanger scaling, and long-term operating condition drift, maintaining high prediction accuracy and control performance.

[0028] Example 3: This example, as a preferred embodiment of the foregoing examples, further defines the complete inference process of fuzzy PID correction, the quantitative calculation method of abnormal operating condition compensation increment, and describes the overall application process of this technical solution in a direct liquid-cooled temperature controller. Through specific algorithm details and execution logic, it ensures that the system can achieve accurate flow control and rapid response under complex operating conditions.

[0029] In the fuzzy PID correction stage, the fuzzification of the coupled liquid temperature deviation (ΔT) and its rate of change (rΔT) adopts a triangular membership function, and the fuzzy subset is divided into seven levels: negative large, negative medium, negative small, zero, positive small, positive medium, and positive large. The specific calculation methods of the three auxiliary adjustment factors are as follows: the circulating liquid thermophysical property change factor (γ1) is obtained by dividing the absolute value of the circulating liquid density deviating from the standard state density by the standard state density, plus the absolute value of the circulating liquid isobaric specific heat capacity deviating from the standard state isobaric specific heat capacity divided by the standard state isobaric specific heat capacity, and summing them up; the load power change factor (γ2) is obtained by dividing the absolute value of the power difference between adjacent acquisition cycles by the rated power; the ambient temperature change factor (γ3) is obtained by dividing the absolute value of the ambient temperature deviating from the reference value by the ambient reference temperature, plus the absolute value of the ambient humidity deviating from the reference value by the ambient reference humidity, and summing them up. The fuzzy rule base contains forty-nine rules. The activation degree of each rule's consequent is multiplied by a weight coefficient (w). The weight coefficient (w) is equal to the first coefficient multiplied by the circulating liquid thermal property change factor (γ1), plus the second coefficient multiplied by the load power change factor (γ2), plus the third coefficient multiplied by the ambient temperature change factor (γ3). For example, if the coupled liquid temperature deviation (ΔT) is positive and the rate of change (rΔT) is positive, then the proportional coefficient correction (ΔKp) is positive, the integral coefficient correction (ΔKi) is negative, and the derivative coefficient correction (ΔKd) is positive. After defuzzification using the centroid method, the precise proportional coefficient correction (ΔKp), integral coefficient correction (ΔKi), and derivative coefficient correction (ΔKd) are obtained. After updating the PID parameters, the basic command is generated with the theoretical optimal flow rate as the target.

[0030] In the abnormal operating condition compensation stage, the criteria for determining scaling conditions are: the rate of change of the heat exchanger inlet and outlet temperature difference exceeds the preset rate of change threshold for multiple consecutive sampling cycles, and the load power and ambient temperature remain stable. The scaling severity coefficient (S_scale) is obtained by subtracting the difference between the current heat exchanger inlet and outlet temperature difference under normal operating conditions from the current heat exchanger inlet and outlet temperature difference, and then dividing by the normal heat exchanger inlet and outlet temperature difference. The pump speed compensation increment (Δn_comp) is equal to the scaling severity coefficient (S_scale) multiplied by the first proportional coefficient (K_scale1) and then multiplied by the base speed (n_base). The value of the first proportional coefficient (K_scale1) ranges from 0.1 to 0.5. The criteria for determining load change conditions are: the load power change rate exceeds 30% per second or the instantaneous change amplitude exceeds 15% of the rated power. Taking a sudden power surge as an example, the pump speed compensation increment (Δn_comp) equals the power surge increment divided by the rated power, multiplied by the second proportional coefficient (K_scale2), and then multiplied by the base speed (n_base). The value of the second proportional coefficient (K_scale2) ranges from 0.2 to 0.8. The valve opening compensation increment (Δθ_comp) and the pump speed compensation increment (Δn_comp) are calculated synchronously according to a coordinated ratio. The compensated execution command is output after amplitude limiting and first-order low-pass filtering, confining it to the safe operating range of the pump and valve to avoid system oscillation caused by violent actuator movements.

[0031] The overall application process of this technical solution in a direct liquid-cooled temperature controller is as follows: After the temperature controller is powered on, it first performs system initialization, loads the pre-trained multi-parameter coupled prediction model, fuzzy PID rule base and abnormal working condition judgment rule base, sets the data acquisition frequency and closed-loop feedback cycle, and completes the self-test of the sensor and actuator.

[0032] During continuous operation, the data acquisition module collects parameters in real time, including circulating fluid supply and return temperature, real-time flow rate, pipeline differential pressure, load power, pump speed, valve opening, and ambient temperature and humidity. The preprocessing module performs outlier removal, extracts the heat exchanger temperature difference change rate and load power change rate as coupling features, and normalizes all dimensionless parameters.

[0033] The preprocessed data is input into a multi-parameter coupled prediction model. The model combines historical time-series data to output the liquid temperature change trend curve for the next five to ten minutes and the theoretical optimal flow rate demand at each time point. Simultaneously, the fuzzy PID module calculates the coupled liquid temperature deviation (ΔT), that is, it integrates the leading deviation and the real-time deviation, and calculates the deviation change rate (rΔT). It introduces the circulating liquid thermal property change factor (γ1), the load power change factor (γ2), and the ambient temperature change factor (γ3) for fuzzy inference, dynamically corrects the proportional coefficient (Kp), integral coefficient (Ki), and derivative coefficient (Kd), and generates the basic pump speed command (n_base) and the basic valve opening command (θ_base) with the theoretical optimal flow rate as the target.

[0034] The abnormal operating condition compensation module matches the preprocessed parameters with the abnormal judgment rule base. If the condition is judged to be normal, the compensation increment is set to zero; if the condition is judged to be scaling or sudden load change, the pump speed compensation increment (Δn_comp) and valve opening compensation increment (Δθ_comp) are calculated based on the degree of deviation, and the compensation increment is positively correlated with the degree of deviation. The instruction coupling execution module couples the basic instruction with the compensation increment. The actual pump speed (n_act) is equal to the basic speed instruction (n_base) plus the speed compensation increment (Δn_comp), and the actual valve opening (θ_act) is equal to the basic opening instruction (θ_base) plus the opening compensation increment (Δθ_comp). This drives the variable frequency pump to adjust the speed and the electric regulating valve to adjust the opening, so that the circulating fluid flow rate quickly approaches the theoretical optimal flow rate.

[0035] The closed-loop feedback and iterative module collects the adjusted real-time liquid temperature (T_real) and actual flow rate (Q_act) and compares them with the set liquid temperature range. If the real-time liquid temperature is within the set range, the system returns to step S1 for continuous dynamic adjustment; if the real-time liquid temperature exceeds the set range, the deviation data is marked as optimization samples and fed back to the prediction model and fuzzy PID module to fine-tune parameters, optimize the model and rule base. Every 72 hours of operation or when the amount of new data reaches a threshold, the system automatically uses historical data to iteratively train the prediction model, while updating the fuzzy inference rule base and the abnormal operating condition judgment rule base to achieve long-term adaptive optimization. During the iterative optimization process, the system verifies the validity of new data, removes abnormal samples, and inputs them into the model for training to ensure the quality of training data.

[0036] In summary, this invention achieves advanced prediction of liquid temperature change trends through a multi-parameter coupled prediction model, effectively solving the regulation lag problem inherent in traditional feedback control. Utilizing fuzzy PID combined with a multi-factor dynamic correction mechanism, it successfully adapts to the thermal properties of the circulating liquid and environmental fluctuations, significantly enhancing the system's anti-interference capability. Through abnormal operating condition judgment and active compensation logic, it quickly responds to scaling and load surges, avoiding temperature runaway caused by passive regulation. Combined with closed-loop feedback and model iterative optimization, it ensures the long-term adaptability of the equipment. This solution achieves precise, rapid, and stable flow regulation without increasing hardware costs, greatly improving the temperature control accuracy and operational reliability of direct liquid-cooled temperature controllers, fully meeting the stringent thermal management requirements of high-end precision loads.

Claims

1. A method for intelligent flow correction in a direct liquid-cooled temperature controller, characterized in that, include: S1. Collect and preprocess multi-dimensional parameters of circulating fluid, load, equipment and environment, input them into the prediction model, and output the future liquid temperature trend curve and the theoretical optimal flow rate requirement. S2. Calculate the adaptive weighting coefficient to fuse the advanced liquid temperature deviation and the real-time liquid temperature deviation to obtain the coupled liquid temperature deviation and the rate of change. Combine the circulating liquid thermal properties, load power and ambient temperature change factor fuzzy reasoning correction ratio, integral and differential coefficients to generate the basic speed and opening command. S3. Determine abnormal operating conditions, calculate compensation increments, and couple basic instructions with compensation increments to generate execution instructions to drive water pumps and valves to regulate flow.

2. The intelligent flow correction method for a direct liquid-cooled temperature controller according to claim 1, characterized in that, The multi-dimensional parameters in S1 include circulating liquid supply and return liquid temperature, pipeline pressure difference, load power and ambient temperature and humidity. Preprocessing includes filtering out outliers within a reasonable range threshold, extracting the coupling characteristics of heat exchanger temperature difference change rate and load power change rate, and mapping non-dimensional parameters to a unified numerical range for normalization.

3. The intelligent flow correction method for a direct liquid-cooled temperature controller according to claim 1, characterized in that, In S2, the advanced liquid temperature deviation is the set liquid temperature minus the predicted liquid temperature, the real-time liquid temperature deviation is the set liquid temperature minus the actual liquid temperature, and the coupled liquid temperature deviation is the adaptive weighting coefficient multiplied by the advanced liquid temperature deviation plus one minus the adaptive weighting coefficient multiplied by the real-time liquid temperature deviation.

4. The intelligent flow correction method for a direct liquid-cooled temperature controller according to claim 1 or 3, characterized in that, In S2, the coupled liquid temperature deviation and rate of change are fuzzified into a fuzzy subset. The circulating liquid thermal properties, load power and ambient temperature change factors are introduced to optimize the inference weight allocation. After rule inference and clarification, the proportional, integral and differential coefficients are corrected. The corrected proportional coefficient is equal to the initial reference proportional coefficient plus the proportional coefficient correction.

5. The intelligent flow correction method for a direct liquid-cooled temperature controller according to claim 1, characterized in that, In S3, if the temperature difference between the inlet and outlet of the heat exchanger continues to increase beyond the preset trend, it is judged as a scaling condition. If the load power change exceeds the preset ratio, it is judged as a load change condition. The compensation increment for normal operation is set to zero. For abnormal operation, the pump speed and valve opening compensation increment are calculated based on the degree of parameter deviation, and the increment is positively correlated with the degree of deviation.

6. The intelligent flow correction method for a direct liquid-cooled temperature controller according to claim 1 or 5, characterized in that, In step S3, after the command is executed to drive the water pump and valve to adjust the flow rate, the real-time liquid temperature and actual flow rate are collected and compared with the set liquid temperature range. If the range is exceeded, the deviation data is fed back to the prediction model and the fuzzy PID module to fine-tune the parameters. When the cumulative runtime or the amount of new data reaches the preset iteration cycle, valid data from the historical database is input into the prediction model for iterative training, and the fuzzy inference rule base and anomaly judgment rule base are updated simultaneously.

7. A flow rate intelligent correction system for a direct liquid-cooled temperature controller, characterized in that, The data acquisition module collects multi-dimensional parameters, preprocesses them, and inputs them into the multi-parameter coupled prediction model module. The multi-parameter coupled prediction model module outputs the future liquid temperature trend curve and the theoretical optimal flow rate requirement to the fuzzy PID intelligent correction module. The fuzzy PID intelligent correction module calculates the coupled liquid temperature deviation and rate of change, and combines the adjustment factor correction ratio, integral and derivative coefficients to generate basic instructions, which are then sent to the instruction coupled execution module. The abnormal operating condition compensation module calculates the compensation increment and transmits it to the instruction coupled execution module. The instruction coupled execution module couples the basic instructions and the compensation increment to generate execution instructions to drive the water pump and valve.

8. The intelligent flow correction system for a direct liquid-cooled temperature controller according to claim 7, characterized in that, The closed-loop feedback and iteration module collects the real-time liquid temperature and actual flow rate after adjustment, and feeds back the deviation data that exceeds the set liquid temperature range to the multi-parameter coupled prediction model module and the fuzzy PID intelligent correction module respectively. This module also triggers the update of the fuzzy inference rule base and the anomaly judgment rule base when the preset iteration cycle is reached, and inputs newly added valid data from the historical database into the multi-parameter coupled prediction model module for iterative fine-tuning.

9. The intelligent flow correction system for a direct liquid-cooled temperature controller according to claim 7, characterized in that, The circulating fluid thermophysical property change factor is related to the circulating fluid density and isobaric specific heat capacity parameters; the load power change factor is related to the load power change; and the ambient temperature change factor is related to the degree of deviation of ambient temperature and humidity from the reference value.

10. The intelligent flow correction system for a direct liquid-cooled temperature controller according to claim 7, characterized in that, The multi-parameter coupled prediction model module is pre-trained based on historical liquid temperature and flow rate data. When the preset iteration cycle is reached, new valid data will be added to iteratively fine-tune the model parameters.

Images