Cascade waste heat recovery process coordinated control method and device based on liquid cooling data

Abstract

This invention discloses a collaborative control method and device for a cascade waste heat recovery process based on liquid cooling data, relating to the field of liquid cooling control technology. The method includes: S1, collecting liquid cooling heat recovery monitoring data and performing preprocessing on the data; S2, constructing a liquid cooling heat recovery state mapping model, generating heat recovery response prediction results, and generating cascade connection mapping data when the state transmission mismatch flag value is 0; S3, constructing a thermal field evolution prediction model to predict the thermal environment state for a future prediction period, obtaining thermal environment prediction data; S4, generating cascade connection classification results for the prediction period, calculating cascade collaborative adjustment values ​​according to liquid cooling branches, determining the collaborative control results, and generating corresponding control commands to be issued to the execution components for execution. This solves the problem of asynchronous state transmission between the cooling control side and the heat recovery side in existing liquid-cooled data center waste heat recovery processes, leading to inaccurate determination of cascade connection relationships and a disconnect between thermal environment prediction and waste heat distribution.

Description

Technical Field

[0001] This invention relates to the field of liquid cooling control technology, specifically to a method and apparatus for coordinated control of a cascade waste heat recovery process based on liquid cooling data. Background Technology

[0002] With the continuous development of high-density computing equipment, liquid cooling technology, and multi-scenario waste heat utilization technology in data centers, waste heat recovery and cascade utilization based on liquid cooling links have gradually become an important technical direction for energy saving, consumption reduction, and overall energy efficiency improvement in data centers. Existing technologies typically focus on waste heat recovery control, waste heat utilization regulation, environmental disturbance adaptation optimization, and multi-execution unit collaborative control. By collecting operating parameters such as temperature, flow rate, and pressure, and combining them with controllers or optimization algorithms, the waste heat recovery process is regulated to improve waste heat utilization, reduce energy loss, and enhance system operational stability.

[0003] For example, application CN104315916B discloses an electronic control system for waste heat recovery and utilization, including a waste heat recovery control unit, a waste heat utilization control unit, and a central control unit. The advantage of this invention lies in its ability to significantly improve the waste heat utilization rate by using multiple differential pressure sensors and temperature sensors to collect data, and by controlling the preheating process through first and second PLC controllers. This allows the waste heat to be applied to production and domestic hot water, achieving a utilization rate of 45% to 55%.

[0004] For example, application CN121578651A discloses an optimized control method and system for waste heat recovery. The method includes: obtaining an environmental mutation factor based on multiple environmental parameters of different dimensions; determining whether the environmental mutation factor is greater than an environmental mutation threshold; if so, triggering an emergency optimization thread to generate a control strategy that meets reliability constraints, thereby minimizing efficiency loss and equipment damage. This invention obtains an environmental mutation factor used to quantify the intensity of changes in environmental parameters within a preset time period. This factor can quantify the intensity of external disturbances, driving the system to dynamically balance efficiency and reliability. Compared to traditional methods for waste heat recovery based on static operating conditions and fixed control parameters, this invention triggers an emergency optimization thread based on the environmental mutation factor, resulting in a more sensitive response, minimizing efficiency loss and equipment damage, and improving the dynamic adaptability and overall recovery efficiency of the waste heat recovery system.

[0005] However, existing technologies mainly focus on process control, environmental disturbance response, or local efficiency optimization in general waste heat recovery systems. While these technologies can improve waste heat utilization and operational adaptability to some extent, they do not adequately address the collaborative control issues arising from the deep coupling of the "controlled cooling process" and the "regenerative process" in liquid-cooled data center scenarios. Especially when the heat load of different liquid cooling branches in a data center fluctuates continuously, the pump speed and valve position on the controlled cooling side are dynamically adjusted, the connection conditions of the regenerative side cascade branches change in real time, and multiple heat-consuming ends such as residential heat exchange stations, greenhouses, and swimming pool water replenishment exist simultaneously, existing technologies lack a system solution that can simultaneously characterize the state transmission on the controlled cooling side, determine the cascade connection relationship, predict future thermal environment evolution, and generate collaborative control commands based on liquid cooling monitoring data. This easily leads to problems such as asynchronous states between the controlled cooling side and the regenerative side, mismatched cascade connections, distorted heat distribution, and low collaborative control efficiency.

[0006] Therefore, in order to address the above problems, there is an urgent need for a collaborative control method and device for the cascade waste heat recovery process based on liquid cooling data. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a collaborative control method and device for a cascade waste heat recovery process based on liquid cooling data. This solves the problem of asynchronous state transmission between the controlled cooling side and the regenerating side in the existing liquid-cooled data center waste heat recovery process, which leads to inaccurate determination of the cascade connection relationship and a disconnect between thermal environment prediction and waste heat distribution.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: a collaborative control method for a cascade waste heat recovery process based on liquid cooling data, comprising: S1, collecting liquid cooling heat recovery monitoring data, performing preprocessing on the liquid cooling heat recovery monitoring data, and outputting preprocessed liquid cooling heat recovery monitoring data; S2, constructing a liquid cooling heat recovery time series sample and a liquid cooling heat recovery state mapping model based on the preprocessed liquid cooling heat recovery monitoring data, generating a heat recovery response prediction result; combining the change in the control cooling side adjustment amount and the prediction deviation on the heat recovery side to generate a state transmission mismatch marker, and in the state transmission... When the mismatch flag value is 0, cascade load-bearing mapping data is generated; S3, the preprocessed liquid cooling regeneration monitoring data and cascade load-bearing mapping data are read, a thermal field evolution prediction model is constructed, the thermal environment state of the future prediction period is predicted, and thermal environment prediction data is obtained; S4, based on the thermal environment prediction data and cascade load-bearing mapping data, the cascade load-bearing classification results for the prediction period are generated, the cascade coordinated adjustment value is calculated according to the liquid cooling branch, the coordinated control result is determined, and the corresponding control command is generated and sent to the execution unit for execution, thus completing the coordinated control of the cascade waste heat recovery process.

[0011] Further, the specific steps for collecting liquid cooling regeneration monitoring data and performing preprocessing on the liquid cooling regeneration monitoring data are as follows: Real-time collection of liquid cooling regeneration monitoring data, including the coolant temperature at the rack inlet, the coolant temperature at the rack outlet, the instantaneous flow rate of the liquid cooling branch, the surface temperature of the CPU chip, the surface temperature of the GPU chip, the internal temperature of the rack, the operating frequency of the variable frequency pump, the speed of the variable frequency pump, the opening degree of the branch electronic expansion valve, the valve position of the branch control valve, the primary side inlet temperature of the plate heat exchanger, the inlet water temperature of the heat pump, the outlet water temperature of the heat pump, the operating status of the heat pump, the supply water temperature of the first-stage branch, the supply water temperature of the second-stage branch, and the supply water temperature of the third-stage branch. The data includes the following parameters: supply water temperature, opening value of the electric regulating valve of the first-stage branch, opening value of the electric regulating valve of the second-stage branch, opening value of the electric regulating valve of the third-stage branch, return water temperature of the secondary network of the residential heat exchange station, air temperature of the greenhouse, and replenishment water temperature of the swimming pool. For the collected liquid-cooled heat recovery monitoring data, a network time protocol synchronization algorithm is used for multi-source time reference synchronization; a piecewise linear interpolation algorithm is used for missing data completion; a box plot anomaly detection algorithm is used for anomaly identification and removal; a Butterworth low-pass filtering algorithm is used for temporal noise suppression; and a Z-score normalization algorithm is used for numerical scaling unification, outputting the preprocessed liquid-cooled heat recovery monitoring data.

[0012] Furthermore, based on the preprocessed liquid cooling regeneration monitoring data, the specific steps for constructing liquid cooling regeneration time-series samples and liquid cooling regeneration state mapping models to generate regeneration response prediction results are as follows: Based on the preprocessed liquid cooling regeneration monitoring data, the continuous monitoring records are sorted according to the liquid cooling branch identifier and sampling time value, and a sliding time window is constructed with N consecutive sampling times; within each sliding time window, the coolant temperature value at the cabinet outlet, the instantaneous flow rate value of the liquid cooling branch, the operating frequency value of the variable frequency pump, the speed value of the variable frequency pump, the opening value of the branch electronic expansion valve, and the valve position value of the branch control valve are extracted as the input sequence for the cooling control side; the primary side inlet temperature value of the plate heat exchanger, the inlet water temperature value of the heat pump, the water supply temperature value of the first stage branch, the water supply temperature value of the second stage branch, and the... The water supply temperature values ​​of the three-stage branch circuits are used as the output sequence of the regenerating side. The input sequence of the controlled cooling side and the output sequence of the regenerating side are combined according to the sampling time to construct the liquid cooling regenerating time series sample. A gated recurrent unit network algorithm is used to construct a liquid cooling regenerating state mapping model. The input sequence of the controlled cooling side in the liquid cooling regenerating time series sample is used as the model input, and the output sequence of the regenerating side corresponding to the next sampling time is used as the model output. Iterative training is performed using the mean square error loss function. When the decrease in the validation set loss of two consecutive iterations is less than the convergence threshold, the model training is considered complete. In the real-time operation stage, the input sequence of the controlled cooling side corresponding to the current sliding time window is input into the trained liquid cooling regenerating state mapping model to obtain the regenerating response prediction result corresponding to the current sampling time.

[0013] Furthermore, the specific steps for generating a state transfer mismatch flag by combining the changes in the cooling-controlled adjustment and the prediction deviation on the regenerative side are as follows: The regenerative response prediction result is compared with the actual collected plate heat exchanger primary inlet temperature, heat pump inlet water temperature, first-stage branch water supply temperature, second-stage branch water supply temperature, and third-stage branch water supply temperature at the current sampling time, and the corresponding prediction deviation value is calculated. Then, the cabinet outlet coolant temperature, liquid cooling branch instantaneous flow rate, variable frequency pump operating frequency, variable frequency pump speed, branch electronic expansion valve opening, and branch control valve position are compared between the current sampling time and the previous sampling time. If at least one of these data changes exceeds the corresponding cooling-controlled action change threshold, and at least one prediction deviation value exceeds the corresponding dynamic fluctuation threshold, the state transfer mismatch flag value is recorded as 1; otherwise, the state transfer mismatch flag value is recorded as 0.

[0014] Further, the specific steps for generating tiered acceptance mapping data when the state transmission mismatch flag value is 0 are as follows: When the state transmission mismatch flag value is 0, tiered acceptance judgment is performed by combining the heat pump operating status value, heat pump outlet water temperature value, first-tier branch water supply temperature value, second-tier branch water supply temperature value, third-tier branch water supply temperature value, residential heat exchange station secondary network return water temperature value, greenhouse air temperature value, and swimming pool makeup water temperature value: When the heat pump operating status value is "on", and the difference between the heat pump outlet water temperature value and the residential heat exchange station secondary network return water temperature value is not less than the first-tier acceptance temperature difference threshold, and the difference between the first-tier branch water supply temperature value and the residential heat exchange station secondary network return water temperature value is not less than the first-tier delivery temperature difference threshold, a first-tier acceptance flag is generated; when the state transmission mismatch flag value is not less than 0, the first-tier acceptance flag is generated. When the conditions for first-level acceptance are met, and the difference between the water supply temperature of the second-level branch and the air temperature of the greenhouse is not less than the second-level acceptance temperature difference threshold, a second-level acceptance mark is generated. When the conditions for first-level and second-level acceptance are not met, and the difference between the water supply temperature of the third-level branch and the pool replenishment water temperature is not less than the third-level acceptance temperature difference threshold, a third-level acceptance mark is generated. The state transmission mismatch marks of each liquid cooling branch and the acceptance marks of each level are summarized. When a liquid cooling branch has generated a corresponding level acceptance mark, but the opening value of the electric regulating valve of the corresponding level branch is lower than the branch acceptance opening threshold, a branch acceptance mismatch mark is generated. Finally, the state transmission mismatch marks, the acceptance marks of each level, and the branch acceptance mismatch marks are integrated to output the level acceptance mapping data.

[0015] Further, the preprocessed liquid cooling regeneration monitoring data and cascade connection mapping data are read to construct a thermal field evolution prediction model, predicting the thermal environment state for future prediction periods. The specific steps to obtain the thermal environment prediction data are as follows: Read the preprocessed liquid cooling regeneration monitoring data and cascade connection mapping data; construct a thermal environment state matrix from the cabinet inlet coolant temperature, cabinet outlet coolant temperature, CPU chip surface temperature, GPU chip surface temperature, and cabinet internal space temperature; and construct a thermal environment state matrix from the instantaneous flow rate of the liquid cooling branch, the operating frequency of the variable frequency pump, and the variable frequency pump rotation speed. The speed value and branch electronic expansion valve opening value are used to construct a cooling control adjustment matrix. The heat pump operating status value, heat pump inlet water temperature value, heat pump outlet water temperature value, first-stage branch water supply temperature value, second-stage branch water supply temperature value, third-stage branch water supply temperature value, and stage connection mapping data are used to construct a regenerative connection matrix. Singular value decomposition is performed on the thermal environment state matrix to extract the dominant thermal field modes and construct a reduced-order basis for POD. The thermal environment state matrix at the current sampling time is projected onto the reduced-order basis to obtain the corresponding POD coefficient sequence. Then, the POD coefficient sequence is correlated with the cooling control adjustment... The matrix and the regeneration matrix are concatenated in the order of sampling time to construct the POD-GA-BPNN prediction sample. A backpropagation neural network algorithm optimized by genetic algorithm is used to construct a thermal field evolution prediction model based on the POD-GA-BPNN prediction sample. The current sample is used as the model input, and the POD coefficient sequence corresponding to the future prediction period is used as the model output. The initial weights and thresholds of the backpropagation neural network are optimized by genetic algorithm. Then, the optimized backpropagation neural network is used to predict the POD coefficients for the future prediction period. The predicted POD coefficient sequence is reconstructed with the POD reduced-order basis to obtain the thermal environment prediction data for the future prediction period. The thermal environment prediction data includes the predicted cabinet inlet coolant temperature, predicted cabinet outlet coolant temperature, predicted instantaneous flow rate of the liquid cooling branch, predicted CPU chip surface temperature, predicted GPU chip surface temperature, predicted cabinet internal space temperature, predicted heat pump inlet water temperature, predicted heat pump outlet water temperature, predicted first-stage branch water supply temperature, predicted second-stage branch water supply temperature, and predicted third-stage branch water supply temperature.

[0016] Furthermore, the specific steps for generating the cascade acceptance classification results for the prediction period based on thermal environment prediction data and cascade acceptance mapping data are as follows: Read the thermal environment prediction data and cascade acceptance mapping data, first remove liquid-cooled branches with a state transfer mismatch flag value of 1 or with a generated branch acceptance mismatch flag, then compare the predicted heat pump outlet water temperature value of the remaining liquid-cooled branches with the secondary network return water temperature value of the residential heat exchange station, the predicted water supply temperature value of the second cascade branch with the air temperature value of the greenhouse, and the predicted water supply temperature value of the third cascade branch with the swimming pool makeup water temperature value, retain the liquid-cooled branches that meet the corresponding cascade acceptance conditions, and generate the cascade acceptance classification results for the prediction period.

[0017] Further, the specific steps for calculating the cascade coordinated regulation value for each liquid cooling branch are as follows: Multiply the difference between the predicted coolant temperature at the cabinet outlet and the predicted coolant temperature at the cabinet inlet for the i-th liquid cooling branch, the square root of the instantaneous flow rate of the liquid cooling branch plus one, and the arctangent of the predicted heating temperature difference of the corresponding cascade plus one to obtain the regulation numerator; add the natural constant e to the difference between the predicted CPU chip surface temperature and the predicted internal temperature of the cabinet for the i-th liquid cooling branch and take the natural logarithm; add the natural constant e to the difference between the predicted GPU chip surface temperature and the predicted internal temperature of the cabinet for the i-th liquid cooling branch and take the natural logarithm; add both to the constant 1 to obtain the regulation denominator; divide the regulation numerator by the regulation denominator to obtain the cascade coordinated regulation value for the i-th liquid cooling branch.

[0018] Further, the specific steps for determining the coordinated control results and generating corresponding control commands to be issued to the execution components are as follows: The coordinated adjustment values ​​of the liquid-cooled branches within each stage are summarized to obtain the total adjustment value of the first, second, and third stages; then, the ratio of the coordinated adjustment value of each liquid-cooled branch to the corresponding total adjustment value is calculated to obtain the stage adjustment ratio for each liquid-cooled branch; the stage adjustment ratio is used as the basis for adjusting the opening of the electric regulating valve of the corresponding stage branch, the total adjustment value of the first stage is used as the basis for determining the start / stop of the heat pump, and the summarized result of the coordinated adjustment values ​​of all liquid-cooled branches is used as the operating frequency value, speed value, and branch electronic control value. The linkage adjustment of the expansion valve opening value is used to determine the coordinated control results within the predicted future period. Based on the coordinated control results, when the total adjustment value of the first stage exceeds the heat pump start-up judgment threshold and the heat pump operating status is off, a heat pump start-up command is generated; when the total adjustment value of the first stage is lower than the heat pump stop-down judgment threshold and the heat pump operating status is on, a heat pump stop-down command is generated. According to the stage adjustment ratio of each liquid cooling branch, corresponding stage branch electric regulating valve opening adjustment commands are generated, and variable frequency pump speed control commands and branch electronic expansion valve adjustment commands are generated according to the summary results of the stage coordinated adjustment values ​​of all liquid cooling branches. The generated commands are then sent to the corresponding execution components for execution to complete the coordinated distribution control.

[0019] The second aspect of this invention provides a collaborative control device for a cascade waste heat recovery process based on liquid cooling data, comprising: a data acquisition and processing module, a state mapping and determination module, a thermal field evolution prediction module, and a collaborative allocation and control module, wherein: the data acquisition and processing module is used to acquire liquid cooling heat recovery monitoring data, perform preprocessing on the liquid cooling heat recovery monitoring data, and output preprocessed liquid cooling heat recovery monitoring data; the state mapping and determination module is used to construct a liquid cooling heat recovery time series sample and a liquid cooling heat recovery state mapping model based on the preprocessed liquid cooling heat recovery monitoring data, generate a heat recovery response prediction result; and generate a state by combining the change in the control cooling side adjustment amount and the prediction deviation on the heat recovery side. The system transmits mismatch markers and generates cascade connection mapping data when the mismatch marker value is 0. The thermal field evolution prediction module reads the preprocessed liquid cooling regeneration monitoring data and cascade connection mapping data, constructs a thermal field evolution prediction model, predicts the thermal environment state for the future prediction period, and obtains thermal environment prediction data. The collaborative allocation control module generates cascade connection classification results for the prediction period based on the thermal environment prediction data and cascade connection mapping data, calculates the cascade collaborative adjustment value according to the liquid cooling branch, determines the collaborative control result, and generates corresponding control commands to be issued to the execution unit for execution, thus completing the collaborative control of the cascade waste heat recovery process.

[0020] Beneficial effects

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

[0022] (1) A collaborative control method and device for the cascade waste heat recovery process based on liquid cooling data. By constructing liquid cooling heat recovery time sequence samples and liquid cooling heat recovery state mapping models, and combining the change of the control cooling side adjustment amount with the prediction deviation of the heat recovery side to generate a state transmission mismatch mark, it is possible to identify the actual transmission of the control cooling side state change to the heat recovery side, thereby improving the accuracy and stability of the cascade connection determination of the liquid cooling branch.

[0023] (2) A collaborative control method and device for the cascade waste heat recovery process based on liquid cooling data, by combining the cascade receiving mapping data and the liquid cooling heat recovery monitoring data into a thermal field input matrix, and constructing a thermal environment state matrix, a cooling control adjustment matrix and a heat recovery receiving matrix, can directly introduce the cascade receiving relationship into the thermal field evolution prediction process, thereby improving the matching degree between the thermal environment prediction data and the collaborative control requirements in the future prediction period.

[0024] (3) A method and device for coordinated control of cascade waste heat recovery process based on liquid cooling data, which generates cascade acceptance classification results for the predicted period by combining thermal environment prediction data and cascade acceptance mapping data, and calculates cascade coordinated adjustment value according to liquid cooling branch, can integrate the predicted heating capacity of liquid cooling branch, chip thermal state constraints and user-side acceptance conditions into the cascade allocation process, thereby improving the precision of cascade waste heat allocation.

[0025] (4) A method and device for coordinated control of the cascade waste heat recovery process based on liquid cooling data, by using the cascade coordinated adjustment value, the total adjustment value of each cascade and the cascade adjustment ratio corresponding to each liquid cooling branch as the control basis, and generating heat pump start-stop command, cascade branch electric regulating valve opening adjustment command, variable frequency pump speed regulation command and branch electronic expansion valve adjustment command, can improve the overall coordination and control response capability of the liquid cooling control cooling process and the cascade waste heat recovery process. Attached Figure Description

[0026] Figure 1 The flowchart shows the collaborative control method for a cascade waste heat recovery process based on liquid cooling data.

[0027] Figure 2 This is a structural diagram of a collaborative control device for a cascade waste heat recovery process based on liquid cooling data.

[0028] Figure 3 Here is a flowchart of the thermal field evolution prediction based on POD-GA-BPNN;

[0029] Figure 4 This is the determination diagram for the coordinated adjustment of the first-stage liquid cooling branch. Detailed Implementation

[0030] 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.

[0031] Please see Figures 1-4 This invention provides a technical solution: a collaborative control method for a cascade waste heat recovery process based on liquid cooling data, comprising: S1, collecting liquid cooling heat recovery monitoring data, performing preprocessing on the liquid cooling heat recovery monitoring data, and outputting preprocessed liquid cooling heat recovery monitoring data; S2, constructing a liquid cooling heat recovery time series sample and a liquid cooling heat recovery state mapping model based on the preprocessed liquid cooling heat recovery monitoring data, generating heat recovery response prediction results; generating a state transmission mismatch flag by combining the change in the control cooling side adjustment amount and the prediction deviation on the heat recovery side, and generating cascade connection mapping data when the state transmission mismatch flag value is 0; S3, reading the preprocessed liquid cooling heat recovery monitoring data and cascade connection mapping data, constructing a thermal field evolution prediction model, predicting the thermal environment state for the future prediction period, and obtaining thermal environment prediction data; S4, generating cascade connection classification results for the prediction period based on the thermal environment prediction data and cascade connection mapping data, calculating cascade collaborative adjustment values ​​according to the liquid cooling branches, determining the collaborative control results, and generating corresponding control commands to be issued to the execution components for execution, thereby completing the collaborative control of the cascade waste heat recovery process.

[0032] Specifically, the steps for collecting and preprocessing liquid cooling regeneration monitoring data are as follows: Real-time collection of liquid cooling regeneration monitoring data, continuously collected at a fixed sampling period of 1 to 10 seconds. The data includes the following: cabinet inlet coolant temperature, cabinet outlet coolant temperature, instantaneous flow rate of the liquid cooling branch, CPU chip surface temperature, GPU chip surface temperature, cabinet internal temperature, variable frequency pump operating frequency, variable frequency pump speed, branch electronic expansion valve opening, branch control valve position, plate heat exchanger primary side inlet temperature, heat pump inlet water temperature, heat pump outlet water temperature, heat pump operating status, and first-stage branch water supply temperature. The data includes the water supply temperature values ​​of the second-stage branch, the water supply temperature values ​​of the third-stage branch, the opening value of the electric regulating valve of the first-stage branch, the opening value of the electric regulating valve of the second-stage branch, the opening value of the electric regulating valve of the third-stage branch, the return water temperature of the secondary network of the residential heat exchange station, the air temperature of the greenhouse, and the water replenishment temperature of the swimming pool. Specifically, the coolant temperature at the cabinet inlet is collected by a platinum resistance temperature sensor installed on the cabinet inlet pipe, the coolant temperature at the cabinet outlet is collected by a platinum resistance temperature sensor installed on the cabinet return pipe, the instantaneous flow rate of the liquid cooling branch is collected by an electromagnetic flowmeter installed on the main pipe of the liquid cooling branch, the CPU chip surface temperature is collected by a thermistor attached to the CPU package surface, and the GPU chip surface temperature is collected by... Temperature is collected by a thermistor attached to the GPU package surface; the internal temperature of the server rack is collected by an air temperature sensor installed in the airflow channel inside the rack; the operating frequency of the variable frequency pump is read from the feedback register of the variable frequency driver; the speed of the variable frequency pump is calculated from the feedback signal of the pump shaft encoder; the opening value of the branch electronic expansion valve is collected by the position feedback signal of the valve actuator; the valve position value of the branch control valve is collected by the valve position transmitter; the primary side inlet temperature of the plate heat exchanger is collected by a temperature sensor installed in the primary side inlet pipe section of the plate heat exchanger; the inlet water temperature of the heat pump is collected by a temperature sensor installed in the evaporator side inlet pipe section of the heat pump; the outlet water temperature of the heat pump is collected by a temperature sensor installed in the condenser side outlet pipe section of the heat pump. The operating status values ​​are read through the status register of the heat pump controller. The water supply temperature values ​​of the first-stage branch, the second-stage branch, and the third-stage branch are collected by temperature sensors installed in the corresponding water supply pipe sections. The opening values ​​of the electric regulating valves of the first-stage branch, the second-stage branch, and the third-stage branch are collected by the valve position feedback signals of the corresponding electric regulating valves. The return water temperature value of the secondary network of the residential heat exchange station is collected by temperature sensors installed in the return water pipe section of the secondary network of the residential heat exchange station. The air temperature value of the greenhouse is collected by air temperature sensors deployed near the crop layer in the greenhouse. The swimming pool makeup water temperature value is collected by temperature sensors installed in the swimming pool makeup water pipe section.During data acquisition, a unified time source distributes clock references to each acquisition terminal. The local clock of each acquisition terminal is compared with the time from the time source to correct sampling timestamp deviations. A network time protocol synchronization algorithm is used to perform multi-source time reference synchronization. The technical principle is to obtain the offset of the acquisition terminal's clock relative to the time source by estimating the round-trip time delay of messages, and then write the offset into the data acquisition time correction process. This ensures that the cabinet inlet coolant temperature, cabinet outlet coolant temperature, instantaneous flow rate of the liquid cooling branch, heat pump inlet water temperature, heat pump outlet water temperature, and water supply temperature of each cascade branch, all generated at the same physical moment, maintain a unified time reference. When the corrected sampling timestamp deviations of each acquisition terminal are small... When considering time synchronization tolerance, it is determined that multi-source time reference synchronization is complete, with a time synchronization tolerance ranging from 10 milliseconds to 100 milliseconds. For the synchronized liquid cooling regeneration monitoring data, an interpolation interval is first constructed using adjacent sampling times. A piecewise linear interpolation algorithm is then used to perform missing data completion processing. The technical principle is to reconstruct the missing value using the linear change relationship between two valid sampling points before and after the missing position, ensuring that the cabinet inlet coolant temperature, cabinet outlet coolant temperature, instantaneous flow rate of the liquid cooling branch, and the water supply temperature of the first, second, and third tier branches are continuous on the time axis. Then, the completed liquid cooling regeneration monitoring data is statistically analyzed using a sliding sampling window. Quantiles and interquartile ranges are used to perform anomaly detection and removal processing using a box plot anomaly detection algorithm. The technical principle is to classify and delete data records falling below the product of the lower quartile minus the anomaly discrimination factor and the interquartile range, and above the upper quartile plus the product of the anomaly discrimination factor and the interquartile range, as anomalies. This prevents transient spikes from entering the subsequent mapping and modeling process. Then, the liquid cooling regeneration monitoring data after anomaly removal is digitally filtered chronologically, and a Butterworth low-pass filter algorithm is used for temporal noise suppression. The sampling frequency is calculated based on the sampling period of the liquid cooling regeneration monitoring data, and the cutoff frequency is taken as 0.1 to 0.3 times the sampling frequency. The filter order ranges from 2nd to 4th. Its technical principle lies in using a filter with a smooth passband amplitude-frequency response and stable transition near the cutoff frequency to suppress high-frequency measurement jitter while preserving the effective low-frequency trends of cabinet thermal state changes, flow rate changes, valve position changes, and heat pump temperature changes. The filtered liquid-cooled regenerative monitoring data is then processed by calculating the mean and standard deviation according to the liquid-cooled branch identifier and data item, using the Z-score standardization algorithm for unified numerical scaling. Each data item's mean and standard deviation are calculated within a sliding time window corresponding to the liquid-cooled branch. The sliding time window length ranges from 20 to 100 consecutive sampling times, and the mean and standard deviation are updated in real-time with new sampled data.The technical principle lies in subtracting the corresponding mean from each data item and then dividing by the corresponding standard deviation. This eliminates the dimensional influence between different data items, compresses the numerical deviation between data of different magnitudes, and brings the cabinet outlet coolant temperature, liquid cooling branch instantaneous flow rate, variable frequency pump operating frequency, branch electronic expansion valve opening, heat pump outlet water temperature, first-stage branch water supply temperature, second-stage branch water supply temperature, and third-stage branch water supply temperature into a unified numerical scale. This facilitates the subsequent construction of liquid cooling recovery time-series samples, liquid cooling recovery state mapping models, and thermal field input matrices. After the above processing, the preprocessed liquid cooling recovery monitoring data is output, ensuring that the correspondence between liquid cooling branch identifiers, sampling times, temperature values, flow rates, valve position values, and heat pump operating status values ​​in each sampling record remains intact.

[0033] This implementation plan unifies and refines the data acquisition sources, time base, data continuity, anomaly constraints, noise control, and numerical scale of liquid cooling regeneration monitoring data. This ensures that the coolant temperature values ​​at the rack inlet, rack outlet, instantaneous flow rate of the liquid cooling branch, CPU chip surface temperature, GPU chip surface temperature, rack internal space temperature, heat pump inlet water temperature, heat pump outlet water temperature, and the water supply temperatures of the first, second, and third tier branches form a stable, comparable, and calculable data foundation under the same time series reference. This improves the accuracy of liquid cooling regeneration time series sample construction, enhances the reliability of liquid cooling regeneration state mapping results, reduces the risk of data offset during the thermal field input matrix reorganization process, and provides consistent data support for subsequent generation of tiered mapping data, acquisition of thermal environment prediction data, and calculation of tiered collaborative adjustment values.

[0034] Specifically, the steps for constructing liquid cooling regeneration time-series samples and a liquid cooling regeneration state mapping model based on preprocessed liquid cooling regeneration monitoring data, and generating regeneration response prediction results are as follows: Based on the preprocessed liquid cooling regeneration monitoring data, the continuous monitoring records are sorted according to the liquid cooling branch identifier and sampling time value, and a sliding time window is constructed with N consecutive sampling times, where N is an integer from 5 to 12. If the value of N is too small, it is difficult to cover the complete response process of the cooling side adjustment action transmitted to the regeneration side; if the value of N is too large, it will introduce an excessively long historical period and weaken the dominant role of the current state change in the prediction results. Within each sliding time window, the cabinet outlet coolant temperature value and the instantaneous flow rate of the liquid cooling branch are extracted. The values ​​of the variable frequency pump operating frequency, variable frequency pump speed, branch electronic expansion valve opening, and branch control valve position are used as the input sequence for the cooling side. The cabinet outlet coolant temperature characterizes the current heat dissipation output level of the liquid cooling branch; the instantaneous flow rate of the liquid cooling branch characterizes the heat transfer medium transport capacity of the branch; the variable frequency pump operating frequency and speed characterize the main circulation drive state; and the branch electronic expansion valve opening and control valve position characterize the heat exchange flux regulation state within the branch. By arranging these continuous time quantities in the order of sampling time, a serialized input reflecting the dynamic adjustment process of the cooling side can be formed. The primary inlet temperature of the plate heat exchanger, the heat pump inlet water temperature, and the first... The water supply temperatures of the first, second, and third tier branches are used as the output sequence on the regeneration side. The primary inlet temperature of the plate heat exchanger characterizes the initial thermal state when heat from the liquid-cooled branch enters the heat exchange link, the heat pump inlet water temperature characterizes the temperature level of the regeneration medium before boosting, and the water supply temperatures of the first, second, and third tier branches characterize the actual heating state of the regeneration process at different receiving levels. By arranging the controlled cooling side input sequence and the regeneration side output sequence in the order of sampling time, a serialized output reflecting the temperature response process on the regeneration side can be formed, thus constructing a liquid-cooled regeneration time-series sample. A gate-based approach is used. The gated recurrent unit network algorithm constructs a liquid cooling regenerative state mapping model. The input sequence of the cooling side in the liquid cooling regenerative time series samples is used as the model input, and the output sequence of the regenerative side corresponding to the next sampling time is used as the model output. Iterative training is performed using the mean square error loss function. The gated recurrent unit network controls the proportion of historical state writing through the reset gate and controls the fusion ratio of the current input and historical memory through the update gate. It can extract the time dependency between the change of the cooling side adjustment and the temperature response of the regenerative side in the continuous sampling sequence. The mean square error loss function constrains the model parameters by calculating the mean of the square difference between the predicted value and the actual value, so that the regenerative response prediction results keep a continuous approximation in terms of temperature magnitude.During training, the liquid-cooled regenerative time-series samples are divided into training and validation sample segments according to time sequence. The validation set loss is recorded after each iteration. When the decrease in validation set loss for two consecutive iterations is less than the convergence threshold, the model training is considered complete. The convergence threshold is set to 0.001 to 0.01 to avoid ineffective iterations when parameters fluctuate at low amplitudes. In the real-time operation phase, the controlled-cooling side input sequence corresponding to the current sliding time window is input into the trained liquid-cooled regenerative state mapping model to obtain the regenerative response prediction results corresponding to the current sampling time. The regenerative response prediction results include the predicted values ​​of the plate heat exchanger primary side inlet temperature, heat pump inlet water temperature, first-stage branch water supply temperature, second-stage branch water supply temperature, and third-stage branch water supply temperature at the current sampling time. These values ​​characterize the temperature response trend after the current control state on the controlled-cooling side is transferred to the regenerative side. The regenerative response prediction results are further used in the state transfer mismatch marker generation process. The prediction baseline input is compared with the actual collected values ​​of the plate heat exchanger primary inlet temperature, heat pump inlet water temperature, and first-stage branch water supply temperature, second-stage branch water supply temperature, and third-stage branch water supply temperature at the current sampling time. This comparison forms the prediction deviation value, which participates in the state transfer mismatch flag determination. When the state transfer mismatch flag value is 0, the predicted values ​​of the first-stage branch water supply temperature, second-stage branch water supply temperature, and third-stage branch water supply temperature corresponding to the regenerative response prediction result continue to enter the cascade bearing mapping data generation process. Subsequently, this process enters the thermal environment prediction data generation process, the cascade bearing classification result generation process for the prediction period, the cascade collaborative adjustment value calculation process, and the collaborative control result determination process. This ensures that the output result of the liquid-cooled regenerative state mapping model is continuously transmitted along the chain of state transfer verification, bearing mapping constraints, future prediction adjustment, and execution linkage control. Therefore, the model output is not only used for temperature prediction but also directly participates in subsequent heat rearrangement and valve-pump linkage decisions.

[0035] In this implementation scheme, by organizing the coolant temperature at the cabinet outlet, the instantaneous flow rate of the liquid cooling branch, the operating frequency of the variable frequency pump, the speed of the variable frequency pump, the opening degree of the branch electronic expansion valve, the valve position of the branch control valve, and the primary side inlet temperature of the plate heat exchanger, the inlet water temperature of the heat pump, the water supply temperature of the first-stage branch, the water supply temperature of the second-stage branch, and the water supply temperature of the third-stage branch into liquid cooling regeneration time series samples and constructing a liquid cooling regeneration state mapping model, a stable time coupling relationship can be established between the cooling-controlled side adjustment behavior and the regeneration side temperature response. This improves the fit between the regeneration response prediction results and the actual heat transfer link, enhances the accuracy of state transfer mismatch marker generation, and provides a more reliable state mapping foundation for subsequent generation of cascade connection mapping data, acquisition of thermal environment prediction data, generation of cascade connection classification results, calculation of cascade collaborative adjustment values, and determination of collaborative control results.

[0036] Specifically, the steps for generating a state transfer mismatch marker by combining the changes in the controlled cooling side adjustment and the predicted deviation on the regenerative side are as follows: The predicted regenerative response is compared with the actual collected values ​​at the current sampling time: the primary inlet temperature of the plate heat exchanger, the inlet water temperature of the heat pump, the supply water temperature of the first-stage branch, the supply water temperature of the second-stage branch, and the supply water temperature of the third-stage branch. The corresponding prediction deviation value is then calculated. The difference between the predicted primary inlet temperature of the plate heat exchanger and the actual primary inlet temperature of the plate heat exchanger is used to characterize the inlet offset degree when heat from the liquid-cooled branch enters the regenerative link. The difference between the predicted heat pump inlet water temperature and the actual heat pump inlet water temperature is used to characterize... To improve the degree of temperature response deviation of the heat transfer medium, the differences between the predicted water supply temperatures of the first, second, and third-level branches and their corresponding actual collected values ​​are used to characterize the degree of temperature response deviation of each branch. The calculation first subtracts the corresponding actual collected value from each predicted value, and then takes the absolute value to form the prediction deviation value. The principle of using absolute value processing is to eliminate the mutual cancellation effect of positive and negative deviations in sign, so that the temperature response deviation can enter the subsequent mismatch judgment process with a unified scale. Then, the cabinet outlet coolant temperature and liquid cooling branch temperature are compared between the current sampling time and the previous sampling time. For the instantaneous flow rate of the liquid cooling branch, the operating frequency of the variable frequency pump, the speed of the variable frequency pump, the opening degree of the branch electronic expansion valve, and the valve position of the branch control valve, if the change in at least one of these data exceeds the corresponding cooling control action change threshold, and at least one of the prediction deviation values ​​exceeds the corresponding dynamic fluctuation threshold, the state transmission mismatch flag value will be recorded as 1. Specifically, the change in the cabinet outlet coolant temperature value is obtained by subtracting the previous sampling time value from the current sampling time value and taking the absolute value; the change in the instantaneous flow rate of the liquid cooling branch is obtained by subtracting the previous sampling time value from the current sampling time value and taking the absolute value; the change in the operating frequency of the variable frequency pump and the change in the speed of the variable frequency pump are also recorded. The changes in the volume of chemical flux, the opening value of the branch electronic expansion valve, and the valve position value of the branch control valve are all calculated in the same way. The threshold values ​​for changes in cooling action are set separately for each data item. The threshold value corresponding to the cabinet outlet coolant temperature is used to determine whether the thermal state of the heat dissipation outlet has changed significantly. The threshold value corresponding to the instantaneous flow rate of the liquid cooling branch is used to determine whether the heat transfer medium transport capacity of the branch has changed significantly. The threshold values ​​corresponding to the variable frequency pump operating frequency and the variable frequency pump speed are used to determine whether the main circulation drive state has changed significantly. The threshold values ​​corresponding to the opening value of the branch electronic expansion valve and the valve position value of the branch control valve are used to determine whether the flow state of the branch has changed significantly.The dynamic fluctuation threshold is established separately for the primary side inlet temperature of the plate heat exchanger, the inlet water temperature of the heat pump, the supply water temperature of the first-stage branch, the supply water temperature of the second-stage branch, and the supply water temperature of the third-stage branch. During establishment, the corresponding historical values ​​from 5 to 20 consecutive sampling times prior to the current sampling time are read. The absolute value sequence of differences between adjacent sampling times is calculated, and the median of the absolute value sequence is multiplied by the fluctuation amplification to form the corresponding dynamic fluctuation threshold. The technical principle of this processing method is to use historical fluctuation levels to characterize the temperature fluctuation range of the current liquid cooling branch under normal operating conditions, so that mismatch judgment is based on the historical fluctuation background of the liquid cooling branch itself, avoiding false triggering by minor measurement fluctuations. The judgment is as follows: At least one data change exceeding the corresponding cooling control action threshold is used as the criterion for determining that effective adjustment has occurred on the cooling control side; at least one prediction deviation exceeding the corresponding dynamic fluctuation threshold is used as the criterion for determining that the temperature response on the regenerating side has not stably followed the predicted trajectory. When both conditions are met, it indicates that the cooling control side adjustment has occurred and the actual response on the regenerating side deviates from the predicted trajectory to an out-of-bounds level. In this case, the state transmission mismatch flag is recorded as 1; otherwise, the state transmission mismatch flag is recorded as 0. When the state transmission mismatch flag is recorded as 0, it indicates that the data transmission relationship from the cooling control side to the regenerating side in the current liquid cooling branch remains within the allowable fluctuation range, and it can proceed to the subsequent cascade acceptance judgment process.

[0037] In this implementation scheme, by jointly constraining the deviation between the predicted regenerative response and the primary inlet temperature of the plate heat exchanger, the inlet water temperature of the heat pump, the supply water temperature of the first, second, and third stage branches, and the temporal variation intensity between the cabinet outlet coolant temperature, the instantaneous flow rate of the liquid cooling branch, the operating frequency of the variable frequency pump, the speed of the variable frequency pump, the opening degree of the branch electronic expansion valve, and the valve position of the branch control valve, the actual adjustment behavior of the cooling side and the actual response behavior of the regenerative side can be identified under the same discrimination basis. This improves the ability of the state transmission mismatch marker to distinguish the transmission anomalies of the liquid cooling branch, reduces the risk of misjudgment caused by simply relying on temperature deviation, and provides a more reliable basis for mismatch screening for the subsequent generation of stage-by-stage mapping data.

[0038] Specifically, the steps for generating cascade connection mapping data when the state transmission mismatch flag value is 0 are as follows: When the state transmission mismatch flag value is 0, cascade connection discrimination is performed by combining the heat pump operating status value, heat pump outlet water temperature value, first-stage branch water supply temperature value, second-stage branch water supply temperature value, third-stage branch water supply temperature value, residential heat exchange station secondary network return water temperature value, greenhouse air temperature value, and swimming pool makeup water temperature value; among which, the heat pump operating status value is used to characterize whether the booster link is currently in an effective heating state, the heat pump outlet water temperature value is used to characterize the available heating temperature after boosting, the first-stage branch water supply temperature value is used to characterize the actual delivery temperature of the first-stage branch to the residential heat exchange station, and the second-stage branch water supply temperature value is used to characterize the second-stage branch water supply temperature. The actual delivery temperature of the cascade branch lines to the end heat exchanger of the greenhouse is used to characterize the actual delivery temperature of the third-level branch line to the pool makeup water. The return water temperature of the secondary network of the residential heat exchange station is used to characterize the current base temperature of the return water of the residential heat exchange station. The air temperature of the greenhouse is used to characterize the current base temperature of the air in the heated space of the greenhouse. The pool makeup water temperature is used to characterize the current base temperature of the pool makeup water. When determining the cascade connection, the differences between the heat pump outlet water temperature and the return water temperature of the secondary network of the residential heat exchange station, the differences between the supply water temperature of the first-level branch line and the return water temperature of the secondary network of the residential heat exchange station, the differences between the supply water temperature of the second-level branch line and the air temperature of the greenhouse, and the differences between the supply water temperature of the third-level branch line and the pool makeup water temperature are calculated separately. The differences in temperature values ​​are obtained by subtracting the latter from the former. Specifically, the difference between the heat pump outlet water temperature and the return water temperature of the secondary network of the residential heat exchange station characterizes the carrying capacity of the upgraded heat medium relative to the baseline temperature of the return water of the residential heat exchange station; the difference between the supply water temperature of the first-stage branch and the return water temperature of the secondary network of the residential heat exchange station characterizes the temperature coverage capacity when the first-stage branch delivers heat to the residential heat exchange station; the difference between the supply water temperature of the second-stage branch and the air temperature of the greenhouse characterizes the temperature driving capacity when the second-stage branch transfers heat to the air inside the greenhouse via the end heat exchange device; and the difference between the supply water temperature of the third-stage branch and the pool makeup water temperature characterizes the heat replenishment driving capacity when the third-stage branch releases heat to the pool makeup water. When the heat pump operating status value... To enable the system, a first-stage acceptance marker is generated when the temperature difference between the heat pump outlet water and the return water temperature of the secondary network of the residential heat exchange station is not less than the first-stage acceptance temperature difference threshold, and the temperature difference between the supply water temperature of the first-stage branch and the return water temperature of the secondary network of the residential heat exchange station is not less than the first-stage delivery temperature difference threshold. The first-stage acceptance temperature difference threshold is used to constrain the outlet temperature of the heat medium after the heat pump is boosted to be higher than the current base temperature of the return water of the residential heat exchange station to reach the acceptance lower limit. The first-stage delivery temperature difference threshold is used to constrain the supply water temperature of the first-stage branch to maintain a sufficient temperature difference margin before being delivered to the residential heat exchange station. Only when the temperature on the heat pump outlet side meets the acceptance requirements and the temperature on the supply water side of the first-stage branch meets the delivery requirements is the current liquid-cooled branch determined to have the first-stage acceptance capability.When the first-tier acceptance conditions are not met, and the difference between the water supply temperature of the second-tier branch and the air temperature of the greenhouse is not less than the second-tier acceptance temperature difference threshold, a second-tier acceptance mark is generated. The second-tier acceptance temperature difference threshold is used to constrain the water supply temperature of the second-tier branch to maintain a minimum heat exchange drive difference relative to the greenhouse air temperature, ensuring that the heat from the second-tier branch is continuously released into the greenhouse air after passing through the greenhouse terminal heat exchange device. When the first and second-tier acceptance conditions are not met, and the difference between the water supply temperature of the third-tier branch and the pool makeup water temperature is not less than the third-tier acceptance temperature difference threshold, a third-tier acceptance mark is generated. The temperature difference threshold is used to constrain the water supply temperature of the third-level branch to maintain a minimum heat supply temperature difference relative to the pool makeup water temperature, ensuring that heat from the third-level branch can enter the pool makeup water heating process. The above judgment process uses a step-by-step judgment method according to the tiered order. The technical principle is to first determine the capacity of the first tier with the highest temperature requirement, then the second tier with the second highest temperature requirement, and finally the third tier with the lowest temperature requirement. This ensures that the same liquid-cooled branch enters only one capacity level at the same sampling time, avoiding duplicate classification and tiered mapping conflicts. The status transmission mismatch flags of each liquid-cooled branch and the capacity flags of each tier are summarized. The summary is first performed by liquid-cooled branch. A branch-level discrimination record is established, and then the state transmission mismatch mark, the first-level acceptance mark, the second-level acceptance mark, and the third-level acceptance mark are written into the corresponding liquid-cooled branch record, so that each liquid-cooled branch forms a unique acceptance state combination at the current sampling time. When a liquid-cooled branch has generated a corresponding level acceptance mark, but the opening value of the corresponding level branch's electric regulating valve is lower than the branch acceptance opening threshold, a branch acceptance mismatch mark is generated. Among them, the branch acceptance opening threshold is used to constrain the corresponding level branch's electric regulating valve to reach the minimum flow opening. If the liquid-cooled branch has the acceptance capability at the temperature level discrimination level, but the opening value of the corresponding level branch's electric regulating valve is still lower than the minimum flow opening, a mismatch mark will be generated. This indicates that the current valve position execution status is not consistent with the receiving capacity. A branch receiving mismatch marker is used to identify this type of liquid-cooled branch. Finally, the status transmission mismatch marker, each stage receiving marker, and the branch receiving mismatch marker are integrated to output stage receiving mapping data. During integration, a branch-by-branch mapping result record is formed according to the liquid-cooled branch identifier and the sampling time value. This ensures that the stage receiving mapping data simultaneously retains whether the liquid-cooled branch currently has an abnormal status transmission, which stage receiving capacity it currently possesses, and whether there is a valve position receiving mismatch. This provides a directly callable receiving mapping basis for subsequent thermal environment prediction data constraint input, generation of stage receiving classification results for the prediction period, and calculation of stage coordinated adjustment values.

[0039] In this implementation plan, the connection relationships between the heat pump operating status value, heat pump outlet water temperature value, first-stage branch water supply temperature value, second-stage branch water supply temperature value, third-stage branch water supply temperature value, and the secondary network return water temperature value of the residential heat exchange station, the air temperature value of the greenhouse, and the swimming pool makeup water temperature value are uniformly mapped. The status transmission mismatch mark, each stage connection mark, and branch connection mismatch mark are all written into the same liquid-cooled branch discrimination result. This enables the temperature level adaptation capability, connection level attribution, and valve position execution consistency of the liquid-cooled branch to be expressed in a closed loop under the same discrimination chain. This improves the completeness of the characterization of the liquid-cooled branch's acceptable state by the stage connection mapping data, reduces the risk of mapping distortion caused by the disconnect between stage connection judgment and valve position execution state, and provides a stable connection basis for subsequent thermal environment prediction data constraint input, generation of stage connection classification results for the prediction period, and calculation of stage coordinated adjustment value.

[0040] Specifically, the following steps are taken to read the preprocessed liquid cooling regeneration monitoring data and cascade connection mapping data, construct a thermal field evolution prediction model, and predict the thermal environment state for the future prediction period to obtain the thermal environment prediction data: Figure 3As shown, the preprocessed liquid cooling heat recovery monitoring data and cascade bearing mapping data are read, and a thermal environment state matrix is ​​constructed from the cabinet inlet coolant temperature, cabinet outlet coolant temperature, CPU chip surface temperature, GPU chip surface temperature, and cabinet interior temperature. The cabinet inlet coolant temperature characterizes the initial heat transfer temperature of the liquid cooling medium entering the cabinet, the cabinet outlet coolant temperature characterizes the heat transfer temperature of the liquid cooling medium after it flows out of the cabinet, the CPU chip surface temperature and GPU chip surface temperature characterize the core heat source surface temperature, and the cabinet interior temperature characterizes the thermal background of the cabinet interior environment. This data is then arranged according to the liquid cooling branch identifier and sampling time value. A multidimensional time series can form a thermal environment state matrix reflecting the thermal state evolution of the liquid cooling branch; the instantaneous flow rate of the liquid cooling branch, the operating frequency of the variable frequency pump, the speed of the variable frequency pump, and the opening value of the branch electronic expansion valve are used to construct a cooling control adjustment matrix. The instantaneous flow rate of the liquid cooling branch is used to characterize the transport capacity of the liquid cooling medium in the liquid cooling branch; the operating frequency and speed of the variable frequency pump are used to characterize the change in the main circulation driving force; and the opening value of the branch electronic expansion valve is used to characterize the change in the branch flow cross-sectional area. By synchronously writing the above adjustment data into the same time series according to the sampling time values, a cooling control adjustment matrix reflecting the change process of the cooling control side can be formed; the heat pump operating status value, heat pump inlet water temperature value, heat pump outlet water temperature value, and... The water supply temperatures of the first, second, and third tier branches, along with the tiered connection mapping data, are used to construct a regenerative connection matrix. Here, the heat pump operating status value characterizes whether the booster link is operational; the heat pump inlet water temperature characterizes the base temperature of the heat medium before boosting; the heat pump outlet water temperature characterizes the output temperature of the heat medium after boosting; the water supply temperatures of the first, second, and third tier branches characterize the regenerative transport status at different connection levels; and the tiered connection mapping data characterizes the current connection eligibility, status transmission mismatch, and branch connection mismatch of each liquid-cooled branch. This matrix is ​​constructed by combining the heat medium status data with the connection judgment results using the same sampling method. Synchronous organization of sample time values ​​can form a regeneration connection matrix that reflects changes in the connection relationship during the regeneration process. The technical principle of incorporating the cascade connection mapping data into the regeneration connection matrix is ​​that the future temperature change on the regeneration side is not only determined by the heat pump operating status value, heat pump inlet water temperature value, heat pump outlet water temperature value, first-stage branch water supply temperature value, second-stage branch water supply temperature value, and third-stage branch water supply temperature value, but also by whether the current liquid cooling branch is qualified to receive the connection, whether there is a mismatch in state transmission, and whether there is a branch connection mismatch constraint. If the prediction input is constructed solely based on continuous temperature values, the thermal environment state during the future prediction period will only reflect the thermal state evolution trend and will be difficult to reflect the impact of changes in connection constraints on the regeneration link.After incorporating the cascaded connection mapping data into the regenerative connection matrix, the thermal field evolution prediction model can simultaneously learn the coupling relationships between changes in the thermal state of the liquid cooling branch, changes in the control cooling side regulation, and changes in connection constraints. This improves the prediction stability of the thermal environment prediction data at connection switching moments, mismatch removal moments, and valve position mismatch correction moments during future prediction periods. Singular value decomposition (SVD) is performed on the thermal environment state matrix to extract the dominant thermal field modes and construct a reduced-order basis (POD). The principle of SVD is to decompose the original high-dimensional thermal environment state matrix into a combination of singular vectors and singular values, and then retain the main modes according to the magnitude of the singular value contribution. This extracts the dominant thermal field change structure in the time and space dimensions, reducing the dimensionality of subsequent prediction calculations while retaining the main thermal field features. The principle of intrinsic orthogonal decomposition (IOD) is that the thermal environment state matrix consists of the coolant temperature values ​​at the rack inlet, rack outlet, CPU chip surface, GPU chip surface, and rack interior, exhibiting multi-point, multi-time, and strong correlation characteristics. Directly converting the high-dimensional thermal environment state matrix into a single matrix would significantly reduce the overall thermal field characteristics. The input thermal field evolution prediction model will increase the parameter dimension, amplify the noise disturbance between samples, and weaken the convergence stability of subsequent training. After compressing the high-dimensional thermal environment state into a small number of dominant thermal field modes through intrinsic orthogonal decomposition, the prediction input can be focused on the main structural components in the thermal state change of the liquid cooling branch, reducing the interference of local measurement fluctuations on the thermal environment prediction data of the future prediction period, thereby improving the availability of subsequent prediction results in the control chain. The number of modes retained by intrinsic orthogonal decomposition is denoted as K, where K is an integer from 3 to 15, determined according to the cumulative energy contribution rate threshold, which is 90% to 99%. When the cumulative energy contribution rate after accumulating the singular values ​​from largest to smallest first reaches the cumulative energy contribution rate threshold, the corresponding first K dominant thermal field modes are retained. The thermal environment state matrix at the current sampling moment is projected onto the reduced-order basis of POD to obtain the corresponding POD coefficient sequence. The projection process calculates the projection weights of the current thermal environment state matrix on each dominant thermal field mode, converting the high-dimensional temperature distribution into a low-dimensional coefficient expression, so that the thermal field state at the current sampling moment can be characterized by a set of finite-dimensional POD coefficient sequences. Then, the POD coefficient sequence is concatenated with the cooling control matrix and the heat recovery matrix in the order of sampling moment to construct the POD-GA-BPNN prediction sample. Here, POD-GA-BPNN represents an intrinsic orthogonal decomposition-genetic algorithm optimized backpropagation neural network, and the POD-GA-BPNN prediction sample represents the prediction sample formed by concatenating the POD coefficient sequence, the cooling control matrix, and the heat recovery matrix in the order of sampling moment, which is used for subsequent training of the thermal field evolution prediction model and prediction of the thermal environment state in future prediction periods.A backpropagation neural network algorithm optimized by a genetic algorithm is used to construct a thermal field evolution prediction model based on POD-GA-BPNN prediction samples. The current sample is used as the model input, and the POD coefficient sequence corresponding to the future prediction period is used as the model output. The backpropagation neural network establishes a nonlinear correspondence between the current sample and the future POD coefficient sequence through multi-layer weight mapping. The genetic algorithm performs global optimization of the initial weights and thresholds of the backpropagation neural network through population initialization, fitness calculation, selection, crossover, and mutation processes, avoiding the local optima stagnation caused by traditional random initialization. The length of the future prediction period is denoted as H, which takes 3 to 12 sampling times. If the value of H is too small, it is difficult to cover the complete evolution range of the liquid cooling branch's thermal state to the heat return side. If the value of H is too large, the prediction error in the long term will accumulate and weaken the timeliness of the control command's response to the current adjustment demand. A genetic algorithm is used to optimize the initial weights and thresholds of the backpropagation neural network. The optimized backpropagation neural network is then used to predict the POD coefficients for the future prediction period. During genetic algorithm optimization, the population size is set to 20-100, the maximum number of iterations to 50-300, the crossover probability to 0.6-0.9, and the mutation probability to 0.01-0.2. The population size controls the number of candidate weight and threshold combinations, the maximum number of iterations controls the optimization depth, the crossover probability controls the reconstruction intensity of superior individual combinations, and the mutation probability controls the amplitude of random perturbations during the search process. Limiting these parameters within the aforementioned ranges ensures sufficient search of the weight space while maintaining controllable optimization time, avoiding large prediction errors due to insufficient optimization and increased training burden due to excessive parameter search. The predicted POD coefficient sequence is then reconstructed with the reduced-order POD basis. This reconstruction process remaps the predicted POD coefficient sequence back to the thermal environment state space, restoring the thermal field distribution of the liquid cooling branch during the future prediction period, thus converting the low-dimensional prediction results back into directly callable physical quantity prediction results.The predicted thermal environment data for the future forecast period is obtained. This data includes the predicted coolant temperature at the rack inlet, rack outlet, and instantaneous flow rate of each liquid cooling branch, as well as the predicted CPU chip surface temperature, GPU chip surface temperature, rack internal temperature, heat pump inlet and outlet water temperatures, and the predicted water supply temperatures for the first, second, and third tier branches. The predicted coolant temperature at the rack inlet and outlet, and the predicted flow rate of each liquid cooling branch are also included. The instantaneous flow rate of the liquid-cooled branch is used to characterize the heating drive capacity of the liquid-cooled branch during the future forecast period. The predicted surface temperature values ​​of the CPU chip, GPU chip, and internal space of the cabinet are used to characterize the thermal constraint state during the future forecast period. The predicted inlet water temperature, outlet water temperature, supply water temperature of the first-stage branch, supply water temperature of the second-stage branch, and supply water temperature of the third-stage branch are used to characterize the cascade support capacity during the future forecast period. This provides a directly accessible data foundation for generating subsequent cascade support classification results, calculating cascade coordinated adjustment values, and determining coordinated control results during the subsequent forecast period.

[0041] In this implementation scheme, the following parameters are organized into a thermal environment state matrix: the coolant temperature at the rack inlet, the coolant temperature at the rack outlet, the CPU chip surface temperature, the GPU chip surface temperature, and the internal temperature of the rack. The instantaneous flow rate of the liquid cooling branch, the operating frequency of the variable frequency pump, the speed of the variable frequency pump, and the opening degree of the branch electronic expansion valve are organized into a cooling control matrix. The following parameters are also organized into a cooling regulation matrix: the heat pump operating status value, the heat pump inlet water temperature, the heat pump outlet water temperature, the first-stage branch water supply temperature, the second-stage branch water supply temperature, and the third-stage branch water supply temperature. The water temperature values ​​and cascade load mapping data are organized into a regenerative load matrix, which can unify the thermal state of the liquid cooling side, the regulation state of the controlled cooling side, and the load state of the regenerative side into the same prediction basis. This improves the completeness of the thermal environment prediction data in representing the future thermal state change trend, load capacity change trend, and heating drive change trend of the liquid cooling branch. It also reduces the risk of state fragmentation caused by prediction based solely on local temperature values, and provides a consistent prediction basis for the generation of cascade load classification results, calculation of cascade coordinated regulation values, and determination of coordinated control results in subsequent prediction periods.

[0042] Specifically, the steps for generating the cascade load-bearing classification results for the predicted time period based on thermal environment prediction data and cascade load-bearing mapping data are as follows: First, read the thermal environment prediction data and cascade load-bearing mapping data. Then, remove liquid-cooled branches with a state transmission mismatch flag value of 1 or those with generated branch load-bearing mismatch flags. A state transmission mismatch flag value of 1 indicates that, given the cooling-side adjustment has already occurred, the actual response on the regenerator side deviates significantly from the predicted response. If this branch continues to participate in cascade load-bearing judgment, subsequent heat rearrangement will be based on branch states with distorted transmission relationships. A branch load-bearing mismatch flag indicates that the current liquid-cooled branch already possesses cascade load-bearing capability. However, the opening value of the electric regulating valve of the corresponding cascade branch did not reach the branch's acceptance threshold. If it continues to participate in the cascade classification during the prediction period, the temperature matching relationship will become disconnected from the valve position execution status. Therefore, all branches are first mismatched and incorrectly matched according to the liquid-cooled branch identification, and only liquid-cooled branches with stable transmission relationships and acceptable valve positions are retained. Then, the predicted heat pump outlet water temperature value of the remaining liquid-cooled branches is compared with the return water temperature value of the secondary network of the residential heat exchange station, the predicted water supply temperature value of the second cascade branch is compared with the air temperature value of the greenhouse, and the predicted water supply temperature value of the third cascade branch is compared with the swimming pool makeup water temperature value. When comparing, the predicted heat pump outlet water temperature value is first calculated by subtracting the return water temperature value of the secondary network of the residential heat exchange station. The return water temperature forms the first-tier heating temperature difference. The predicted second-tier branch water supply temperature is then subtracted from the greenhouse air temperature to form the second-tier heating temperature difference. Finally, the predicted third-tier branch water supply temperature is subtracted from the pool makeup water temperature to form the third-tier heating temperature difference. Each tier heating temperature difference represents the current liquid-cooled branch's temperature coverage capability for the corresponding receiving object during the future prediction period. Liquid-cooled branches that meet the corresponding tier receiving conditions are retained. The first-tier receiving condition constrains the output temperature of the heat pump after boosting to reach the lower limit of the receiving temperature relative to the secondary network return water temperature of the residential heat exchange station. The second-tier receiving condition constrains the supply water temperature of the second-tier branch... For greenhouse air temperature, it must reach the lower limit of heating. The third-tier acceptance condition is used to constrain the water supply temperature of the third-tier branch to reach the lower limit of heating relative to the pool water supply temperature. When retaining, each liquid cooling branch is written into the tier assignment result. If a liquid cooling branch meets the first-tier acceptance condition, it is written into the first-tier candidate set. If a liquid cooling branch does not meet the first-tier acceptance condition but meets the second-tier acceptance condition, it is written into the second-tier candidate set. If a liquid cooling branch does not meet the first-tier acceptance condition and does not meet the second-tier acceptance condition but meets the third-tier acceptance condition, it is written into the third-tier candidate set.The technical principle of the above processing lies in first using the state transfer mismatch marker value and the branch acceptance mismatch marker to exclude liquid-cooled branches that lack a stable acceptance basis, and then using the temperature coverage capability of each liquid-cooled branch for different acceptance objects during the future prediction period to complete the tiered rearrangement. This ensures that the heat destination of the liquid-cooled branches during the prediction period is based on the joint constraints of stable transfer relationships, effective acceptance qualifications, and future temperature capacity, ultimately generating the tiered acceptance classification result for the prediction period.

[0043] In this implementation plan, by incorporating the state transfer mismatch marker, branch connection mismatch marker, predicted heat pump outlet water temperature, predicted second-stage branch water supply temperature, predicted third-stage branch water supply temperature, and the secondary network return water temperature of the residential heat exchange station, the air temperature of the greenhouse, and the water replenishment temperature of the swimming pool into the same rearrangement and discrimination process, the classification results of the cascade connection during the prediction period can simultaneously reflect the transfer stability, connection effectiveness, and future temperature coverage of the liquid-cooled branch. This improves the adaptability of the liquid-cooled branch cascade assignment results to the future regeneration process, reduces the risk of heat rearrangement deviation caused by mistakenly including abnormal transfer branches and valve position mismatch branches into the cascade connection sequence, and provides a more stable branch screening basis for subsequent calculation of cascade coordinated regulation values ​​and determination of coordinated control results.

[0044] Specifically, the steps for calculating the cascade coordinated adjustment value for each liquid-cooled branch are as follows: Multiply the difference between the predicted coolant temperature at the cabinet outlet and the predicted coolant temperature at the cabinet inlet for the i-th liquid-cooled branch, the square root of the instantaneous flow rate of the liquid-cooled branch plus one, and the arctangent of the predicted heating temperature difference of the corresponding cascade plus one to obtain the adjustment numerator; where the difference between the predicted coolant temperature at the cabinet outlet and the predicted coolant temperature at the cabinet inlet is used to characterize the heat-carrying temperature rise of the i-th liquid-cooled branch in the future predicted period. The larger the temperature rise, the higher the heat carried by the unit fluid after passing through the cabinet; the square root of the instantaneous flow rate of the liquid-cooled branch plus one is used... To characterize the transport capacity of the liquid cooling medium, square root compression can maintain the trend of increased flow rate leading to enhanced heating capacity while suppressing excessive amplification of the calculation results by large flow rate values. The arctangent value of the predicted heating temperature difference for the corresponding cascade stage, plus one, is used to characterize the temperature coverage capability of the current liquid cooling branch towards the target receiving object. After arctangent transformation, the gain within a large temperature difference range gradually slows down, avoiding the dominance of a single temperature factor in the cascade coordinated adjustment value when the temperature difference is too large. The predicted heating temperature difference for the corresponding cascade stage is taken as the difference between the predicted heat pump outlet water temperature and the secondary network return water temperature of the residential heat exchange station when the i-th liquid cooling branch enters the first cascade stage. In the second stage, the difference between the predicted water supply temperature of the second-stage branch and the air temperature of the greenhouse is used. When the i-th liquid cooling branch enters the third stage, the difference between the predicted water supply temperature of the third-stage branch and the pool replenishment water temperature is used. The difference between the predicted CPU chip surface temperature and the predicted internal temperature of the server rack for the i-th liquid cooling branch is added to the natural constant e and then the natural logarithm is taken. The difference between the predicted GPU chip surface temperature and the predicted internal temperature of the server rack for the i-th liquid cooling branch is added to the natural constant e and then the natural logarithm is taken. Both are added to the constant 1 to obtain the adjustment denominator. Among them, the difference between the predicted CPU chip surface temperature and the predicted internal temperature of the server rack is... The difference between the predicted GPU chip surface temperature and the predicted internal temperature of the cabinet is used to characterize the temperature difference constraint strength of the CPU heat source relative to the internal thermal background of the cabinet. Introducing a natural constant e after the temperature difference term and then taking the natural logarithm ensures that the growth effect of the denominator term in the higher temperature difference range continues to increase, while avoiding excessively rapid compression of the value in the lower temperature difference range. The constant 1 is used to maintain the lower bound of the adjustment denominator term to prevent the denominator term from being too small when the temperature difference constraint is low, which would cause abnormal amplification of the tiered synergistic adjustment value. Dividing the adjustment numerator term by the adjustment denominator term yields the tiered synergistic adjustment value of the i-th liquid cooling branch.The tiered coordinated regulation value is used to comprehensively characterize the heating drive capacity and thermal state constraint strength of the i-th liquid cooling branch during the future prediction period. After the above calculation, liquid cooling branches with larger temperature rise, stronger liquid cooling medium transport capacity, sufficient target temperature difference, and more stable constraints on the CPU chip surface temperature and GPU chip surface temperature relative to the internal temperature of the cabinet will obtain higher tiered coordinated regulation values, thus entering the subsequent process of summarizing the total tiered regulation value, calculating the tiered regulation ratio, and determining the coordinated control result. The predicted cabinet outlet coolant temperature, predicted cabinet inlet coolant temperature, predicted instantaneous flow rate of the liquid cooling branch, predicted CPU chip surface temperature, predicted GPU chip surface temperature, predicted internal temperature of the cabinet, predicted heat pump outlet water temperature, predicted water supply temperature of the second tiered branch, and predicted water supply temperature of the third tiered branch are all prediction results obtained based on preprocessed dimensionless data.

[0045] The specific formula for calculating the tiered coordinated regulation value is as follows:

[0046] ;

[0047] In the formula, This represents the cascade coordinated adjustment value of the i-th liquid-cooled branch. This represents the predicted coolant temperature at the cabinet outlet for the i-th liquid-cooled branch. This represents the predicted cabinet inlet coolant temperature for the i-th liquid-cooled branch. This represents the predicted instantaneous flow rate of the i-th liquid-cooled branch. This represents the predicted CPU chip surface temperature value for the i-th liquid cooling branch. This represents the predicted surface temperature of the GPU chip for the i-th liquid cooling branch. This represents the predicted internal temperature value of the cabinet for the i-th liquid-cooled branch. This represents the predicted heating temperature difference for the i-th liquid-cooled branch corresponding to the first cascade. Take the difference between the predicted heat pump outlet water temperature and the return water temperature of the secondary network of the residential heat exchange station; when the i-th liquid-cooled branch belongs to the second stage, Take the difference between the predicted water supply temperature of the second-level branch and the air temperature of the greenhouse; when the i-th liquid cooling branch belongs to the third level, Take the difference between the predicted water supply temperature of the third-level branch and the water replenishment temperature of the swimming pool.

[0048] In this implementation plan, by incorporating the predicted coolant temperature values ​​at the rack outlet, rack inlet, instantaneous flow rate of the liquid cooling branch, CPU chip surface temperature, GPU chip surface temperature, rack interior temperature, heat pump outlet water temperature, second-level branch water supply temperature, and third-level branch water supply temperature into the same tiered collaborative regulation value calculation process, the heat carrying capacity, temperature adaptability, and thermal constraint strength of the liquid cooling branch during the predicted future period can be uniformly quantified into directly comparable regulation characteristics. This improves the distinguishability of regulation capabilities between different liquid cooling branches, reduces the risk of deviation caused by relying solely on a single temperature difference or flow rate value for regulation judgment, and provides a stable quantitative basis for subsequent calculations of the first, second, and third tiered total regulation values, determination of tiered regulation ratios, and generation of collaborative control results.

[0049] In this embodiment, Table 1 shows the adjustment numerator, adjustment denominator, and stage-wide coordinated adjustment value for the five liquid cooling branches in the first stage. Specifically: the adjustment numerator for liquid cooling branch 1 is 19.0445, the adjustment denominator is 8.3955, and the calculated cascade synergistic adjustment value is 2.2684; the adjustment numerator for liquid cooling branch 2 is 19.8222, the adjustment denominator is 8.4228, and the calculated cascade synergistic adjustment value is 2.3534; the adjustment numerator for liquid cooling branch 3 is 21.0463, the adjustment denominator is 8.4445, and the calculated cascade synergistic adjustment value is 2.4923; the adjustment numerator for liquid cooling branch 4 is 23.2138, the adjustment denominator is 8.4758, and the calculated cascade synergistic adjustment value is 2.7388; the adjustment numerator for liquid cooling branch 5 is 24.6913, the adjustment denominator is 8.4991, and the calculated cascade synergistic adjustment value is 2.9052. The above data is used to quantitatively compare the cascade coordinated adjustment capabilities of different liquid-cooled branches within the first stage under the combined effect of heating drive capability and thermal state constraints, and to provide direct calculation basis for subsequent summarization of cascade coordinated adjustment values ​​of each liquid-cooled branch, calculation of the total adjustment value of the first stage, and generation of opening adjustment commands for the electric regulating valves of the first stage branches.

[0050] Table 1. Data on the first-level tiered coordinated adjustment values.

[0051]

[0052] like Figure 4As shown in the figure, the horizontal axis represents the adjustment denominator for each liquid cooling branch, and the vertical axis represents the adjustment numerator for each liquid cooling branch. The five scattered points in the figure correspond to liquid cooling branches 1 to 5 in the first stage, and the corresponding stage-wide coordinated adjustment value of the liquid cooling branch is marked next to each scattered point. Contour lines with stage-wide coordinated adjustment values ​​of 2.2, 2.4, 2.6, 2.8, and 3.0 are set from bottom to top in the figure to represent different stages of coordinated adjustment capability. Among them, the arrow of "enhanced heating drive direction" indicates the enhancing trend of stage-wide coordinated adjustment capability of liquid cooling branches when the adjustment numerator increases, and the arrow of "enhanced thermal constraint direction" indicates the changing trend of thermal constraint degree of liquid cooling branches when the adjustment denominator increases. As can be clearly seen from the figure, liquid-cooled branches 1 to 5 are distributed along contour lines of different cascade coordinated regulation values. Liquid-cooled branches 4 and 5 are located in the higher cascade coordinated regulation value range, indicating that their heating drive capacity in the first stage is relatively stronger, capable of handling a larger cascade support and valve position adjustment ratio. Liquid-cooled branches 1 and 2 are located in the lower cascade coordinated regulation value range, indicating that they are more suitable as auxiliary support branches participating in the first stage coordinated control. This figure clearly demonstrates that the present invention does not adjust the liquid-cooled branches solely based on a single temperature or flow rate value, but rather calculates the cascade coordinated regulation value through the coupling relationship between the numerator and denominator terms. Based on the cascade coordinated regulation value, the coordinated regulation capacity of different liquid-cooled branches in the first stage is distinguished. This provides a graphical basis for subsequently determining the total regulation value of the first stage, the cascade regulation ratio, and the opening degree of the corresponding electric regulating valves of the cascade branches, verifying the feasibility of the coordinated control logic of the cascade waste heat recovery process of the present invention.

[0053] Specifically, the steps for determining the coordinated control results and generating corresponding control commands to be issued to the execution components are as follows: The coordinated adjustment values ​​of the liquid-cooled branches within each stage are summarized to obtain the total adjustment value for the first stage, the total adjustment value for the second stage, and the total adjustment value for the third stage. The total adjustment value for the first stage is obtained by summing the coordinated adjustment values ​​corresponding to all liquid-cooled branches that generated the first stage acceptance mark; the total adjustment value for the second stage is obtained by summing the coordinated adjustment values ​​corresponding to all liquid-cooled branches that generated the second stage acceptance mark; and the total adjustment value for the third stage is obtained by summing the coordinated adjustment values ​​corresponding to all liquid-cooled branches that generated the third stage acceptance mark. This process of summarizing by stage is employed. This method can uniformly convert the future heating drive capacity of liquid-cooled branches within different receiving levels into the total regulation intensity of each level, which is used to characterize the overall regulation demand level of each level during the current forecast period. Then, the ratio of the level-specific coordinated regulation value of each liquid-cooled branch to the corresponding total regulation value of the level is calculated to obtain the level-specific regulation ratio for each liquid-cooled branch. Specifically, when a liquid-cooled branch generates a first-level receiving mark, the level-specific coordinated regulation value of that liquid-cooled branch is divided by the total regulation value of the first level to obtain the first-level regulation ratio; when a liquid-cooled branch generates a second-level receiving mark, the level-specific coordinated regulation value of that liquid-cooled branch is divided by the total regulation value of the second level to obtain the second-level regulation ratio; when a liquid-cooled branch generates a third-level receiving mark... The third-level regulation ratio is obtained by dividing the cascade coordinated regulation value of the liquid-cooled branch by the total regulation value of the third level. The cascade regulation ratio is used to characterize the relative regulation share that a single liquid-cooled branch should bear within the corresponding receiving level, so that the regulation allocation of different liquid-cooled branches is based on the unified expression of the total regulation intensity within the same level. The cascade regulation ratio is used as the basis for adjusting the opening of the electric regulating valve of the corresponding cascade branch, the total regulation value of the first level is used as the basis for determining the start and stop of the heat pump, and the sum of the cascade coordinated regulation values ​​of all liquid-cooled branches is used as the basis for the linkage adjustment of the variable frequency pump operating frequency value, variable frequency pump speed value, and branch electronic expansion valve opening value to determine the coordinated control results in the future prediction period. Among them, the adjustment of the opening of the electric regulating valve of the first level branch... The adjustment of the first-level regulation ratio corresponding to each liquid-cooled branch is obtained by mapping it to the target opening of the first-level branch electric regulating valve. The adjustment of the second-level branch electric regulating valve opening is obtained by mapping it to the target opening of the second-level regulation ratio corresponding to each liquid-cooled branch. The adjustment of the third-level branch electric regulating valve opening is obtained by mapping it to the target opening of the third-level regulation ratio corresponding to each liquid-cooled branch. The total regulation value of the first level is used to characterize the overall heating demand intensity of the heat pump booster link in the future forecast period. When the total regulation value of the first level continues to increase, it indicates that the demand corresponding to the residential heat exchange station is increasing, and the heating temperature needs to be increased by putting the heat pump into operation.The aggregated results of the cascade coordinated regulation values ​​of all liquid-cooled branches are used to characterize the overall regulation load of the liquid-cooled regeneration process during the current forecast period. When the aggregated results increase, it indicates that the overall heating drive demand of the liquid-cooled branches is enhanced, requiring a simultaneous increase in the operating frequency, speed, and opening of the branch electronic expansion valves to improve the liquid-cooled medium transport capacity, heat flux release capacity, and regeneration side temperature stability. When determining the coordinated control results, the execution constraints of the heat pump start-up and shutdown status, the target opening of the corresponding cascade branch electric regulating valves, the target operating frequency, the target speed, and the target opening of the branch electronic expansion valves are first checked. Among them, a minimum start-up and shutdown interval is set when checking the heat pump start-up and shutdown status, and the minimum start-up and shutdown interval is taken as three consecutive sampling intervals. From the initial sampling time to 12 consecutive sampling times, the heat pump's current operating state remains unchanged within the minimum start-stop interval to avoid frequent switching between adjacent prediction periods. When verifying the target opening of the electric regulating valve for the corresponding cascade branch, a lower valve position limit, an upper valve position limit, and a valve position adjustment dead zone are set. The lower valve position limit is set to 10% to 20%, the upper valve position limit to 85% to 100%, and the valve position adjustment dead zone to 1% to 5%. The lower valve position limit constrains the minimum flow opening of the electric regulating valve for the corresponding cascade branch, the upper valve position limit constrains the maximum allowable opening, and the valve position adjustment dead zone constrains the issuance of no new adjustment action when the difference between the target opening and the current opening is within the dead zone range, to avoid frequent repetition of the valve position due to minor fluctuations. The variable frequency pump... When verifying the target operating frequency value, a frequency change rate limit is set, ranging from 0.5 Hz to 2 Hz per sampling moment. When verifying the target speed value of the variable frequency pump, a speed change rate limit is set, ranging from 50 rpm to 200 rpm per sampling moment. When verifying the target opening degree of the branch electronic expansion valve, an opening change rate limit is set, ranging from 2% to 8% per sampling moment. This ensures that the variable frequency pump operating frequency value, variable frequency pump speed value, and branch electronic expansion valve opening value change at a limited slope between adjacent sampling moments, avoiding shocks in liquid cooling medium transport, amplified heat exchange fluctuations, and oscillations in the regenerator side temperature due to sudden changes in target values. Based on the coordinated control results, when the total adjustment value of the first stage exceeds the heat pump start-up judgment... When a threshold is set and the heat pump's operating status is off, a heat pump start command is generated; when the total regulation value of the first stage is lower than the heat pump shutdown judgment threshold and the heat pump's operating status is on, a heat pump shutdown command is generated. The heat pump start judgment threshold is used to constrain the heat pump's start-up conditions, and the heat pump shutdown judgment threshold is used to constrain the heat pump's shutdown conditions. By setting the start judgment threshold higher than the shutdown judgment threshold, a start-stop / rebound interval is formed, preventing frequent heat pump start-ups and shutdowns caused by fluctuations in the total regulation value of the first stage near the threshold. According to the tiered regulation ratio of each liquid cooling branch, corresponding tiered branch electric regulating valve opening adjustment commands are generated, and variable frequency pump speed control commands and branch electronic expansion valve adjustment commands are generated based on the aggregated results of the tiered coordinated regulation values ​​of all liquid cooling branches.Specifically, the corresponding cascade branch electric regulating valve opening adjustment command is obtained by converting the cascade adjustment ratio into the target valve position adjustment range; the variable frequency pump speed control command is obtained by mapping the aggregated results of the cascade coordinated adjustment values ​​of all liquid cooling branches to the target operating frequency range and target speed range; and the branch electronic expansion valve adjustment command is obtained by mapping the aggregated results of the cascade coordinated adjustment values ​​of all liquid cooling branches to the target opening range. Then, the generated heat pump start command and heat pump stop command are sent to the heat pump for execution, the corresponding cascade branch electric regulating valve opening adjustment command is sent to the corresponding cascade branch electric regulating valve for execution, the variable frequency pump speed control command is sent to the variable frequency pump for execution, and the branch electronic expansion valve adjustment command is sent to the branch electronic expansion valve for execution, completing the coordinated distribution control. The technical principle of the above processing is to convert the cascade adjustment ratio of each liquid cooling branch within the predicted time period into the target valve position adjustment range. The local regulation capacity is first aggregated into the total regulation intensity of the cascade. Then, the proportion of the local regulation capacity relative to the total regulation intensity of the cascade is converted into the valve position regulation share. Simultaneously, the total regulation value of the first cascade is converted into the basis for heat pump start-up and shutdown. The aggregated result of the coordinated regulation values ​​of all liquid-cooled branch cascades is converted into the basis for liquid-cooled side drive intensity. Before execution, start-stop / shutdown mechanisms, minimum start-stop intervals, valve position limits, valve position adjustment dead zones, frequency change rate limits, speed change rate limits, and opening change rate limits are introduced. This ensures that the heat pump operating status value, variable frequency pump operating frequency value, variable frequency pump speed value, branch electronic expansion valve opening value, and corresponding cascade branch electric regulating valve opening value change in a unified regulation logic within the same prediction period. This guarantees that the cascade waste heat recovery process maintains coordination and consistency between the heat source side, transport side, and receiving side, while maintaining stable and controllable execution.

[0054] In this implementation scheme, by further converting the cascade coordinated adjustment value into the first cascade total adjustment value, the second cascade total adjustment value, the third cascade total adjustment value, and the cascade adjustment ratio, and then mapping them to the heat pump operating status value, the variable frequency pump operating frequency value, the variable frequency pump speed value, the branch electronic expansion valve opening value, and the corresponding cascade branch electric regulating valve opening value in the control determination process, the local adjustment capability of the liquid cooling branch and the overall adjustment demand of each cascade can be uniformly quantified and expressed under the same control chain. This improves the consistency of the coordinated control results in responding to changes in the load undertaken by the cascade in the future prediction period, reduces the risk of control deviation caused by the independent adjustment between the heat pump start-up and shutdown state, the liquid cooling side drive state, and the cascade branch valve position state, and provides a more stable linkage control basis for coordinated distribution control.

[0055] like Figure 2As shown, the second aspect of the present invention provides a collaborative control device for a cascade waste heat recovery process based on liquid cooling data, comprising: a data acquisition and processing module, a state mapping determination module, a thermal field evolution prediction module, and a collaborative allocation control module, wherein: the data acquisition and processing module is used to acquire liquid cooling heat recovery monitoring data, perform preprocessing on the liquid cooling heat recovery monitoring data, and output preprocessed liquid cooling heat recovery monitoring data; the state mapping determination module is used to construct liquid cooling heat recovery time series samples and liquid cooling heat recovery state mapping models based on the preprocessed liquid cooling heat recovery monitoring data, generate heat recovery response prediction results; and generate a prediction result by combining the change in the control cooling side adjustment amount and the heat recovery side prediction deviation. The system includes a state transfer mismatch marker, which generates cascade connection mapping data when the state transfer mismatch marker value is 0; a thermal field evolution prediction module, which reads the preprocessed liquid cooling regeneration monitoring data and cascade connection mapping data, constructs a thermal field evolution prediction model, predicts the thermal environment state for the future prediction period, and obtains thermal environment prediction data; and a collaborative allocation control module, which generates cascade connection classification results for the prediction period based on the thermal environment prediction data and cascade connection mapping data, calculates cascade collaborative adjustment values ​​according to the liquid cooling branches, determines the collaborative control results, and generates corresponding control commands to be sent to the execution components for execution, thus completing the collaborative control of the cascade waste heat recovery process.

[0056] In this implementation plan, by incorporating liquid cooling regeneration monitoring data preprocessing, liquid cooling regeneration state mapping, cascade connection mapping, thermal environment prediction, cascade heat rearrangement, cascade collaborative adjustment value calculation, and collaborative control result determination into the same technical chain, the system enables continuous processing of the following values ​​under unified control logic: cabinet inlet coolant temperature, cabinet outlet coolant temperature, instantaneous flow rate of liquid cooling branch, CPU chip surface temperature, GPU chip surface temperature, heat pump inlet water temperature, heat pump outlet water temperature, first-stage branch water supply temperature, second-stage branch water supply temperature, and third-stage branch water supply temperature. This improves the comprehensive response capability of the cascade waste heat recovery process to changes in the liquid cooling side thermal state, changes in the regeneration side connection state, and changes in the thermal environment during the future prediction period. It also reduces the risk of control instability caused by the disconnect between liquid cooling branch connection determination, thermal environment prediction, and control command generation, resulting in higher consistency, continuity, and executability of the collaborative control of the cascade waste heat recovery process.

[0057] 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.

[0058] 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 method for coordinated control of a cascade waste heat recovery process based on liquid cooling data, characterized in that, Includes the following steps: S1: Collect liquid cooling regeneration monitoring data, perform preprocessing on the liquid cooling regeneration monitoring data, and output the preprocessed liquid cooling regeneration monitoring data. S2, Based on the preprocessed liquid cooling regeneration monitoring data, construct the liquid cooling regeneration time series sample and liquid cooling regeneration state mapping model, and generate regeneration response prediction results; The predicted result of the heat recovery response is compared with the actual value collected at the current sampling time to calculate the corresponding prediction deviation value. Then, the change in the cooling side data at the current sampling time and the previous sampling time are compared. If the change in at least one of the data exceeds the corresponding cooling action change threshold and at least one prediction deviation value exceeds the corresponding dynamic fluctuation threshold, the state transmission mismatch flag value is recorded as 1; otherwise, the state transmission mismatch flag value is recorded as 0. A state transfer mismatch marker is generated by combining the change in the cooling side adjustment and the prediction deviation on the regenerating side, and cascade support mapping data is generated when the state transfer mismatch marker value is 0. When the state transmission mismatch flag value is 0, the cascade acceptance judgment is performed by combining the heat pump operating status value, heat pump outlet water temperature value, first-stage branch water supply temperature value, second-stage branch water supply temperature value, third-stage branch water supply temperature value, residential heat exchange station secondary network return water temperature value, greenhouse air temperature value, and swimming pool makeup water temperature value. The system summarizes the state transmission mismatch flags and the stage connection flags of each liquid cooling branch. When a liquid cooling branch has generated a corresponding stage connection flag, but the opening value of the electric regulating valve of the corresponding stage branch is lower than the branch connection opening threshold, a branch connection mismatch flag is generated. Finally, the state transmission mismatch flags, stage connection flags, and branch connection mismatch flags are integrated to output stage connection mapping data. S3: Read the preprocessed liquid cooling regeneration monitoring data and cascade bearing mapping data, construct a thermal field evolution prediction model, predict the thermal environment state for the future prediction period, and obtain thermal environment prediction data. S4 generates the classification results of cascade load-bearing for the predicted period based on thermal environment prediction data and cascade load-bearing mapping data, calculates the cascade coordinated adjustment value according to the liquid cooling branch, determines the coordinated control result, and generates the corresponding control command to be sent to the execution unit for execution, thus completing the coordinated control of the cascade waste heat recovery process.

2. The liquid-cooled data-based cascade waste heat recovery process coordinated control method according to claim 1, characterized in that: The specific steps for collecting liquid cooling regeneration monitoring data and performing preprocessing on the liquid cooling regeneration monitoring data are as follows: Real-time collection of liquid cooling heat recovery monitoring data includes: cabinet inlet coolant temperature, cabinet outlet coolant temperature, instantaneous flow rate of liquid cooling branch, CPU chip surface temperature, GPU chip surface temperature, cabinet internal space temperature, variable frequency pump operating frequency, variable frequency pump speed, branch electronic expansion valve opening, branch control valve position, plate heat exchanger primary side inlet temperature, heat pump inlet water temperature, heat pump outlet water temperature, heat pump operating status, first-stage branch water supply temperature, second-stage branch water supply temperature, third-stage branch water supply temperature, first-stage branch electric regulating valve opening, second-stage branch electric regulating valve opening, third-stage branch electric regulating valve opening, residential heat exchange station secondary network return water temperature, greenhouse air temperature, and swimming pool makeup water temperature. For the collected liquid cooling regeneration monitoring data, the network time protocol synchronization algorithm is used to perform multi-source time reference synchronization processing; the piecewise linear interpolation algorithm is used to perform missing data completion processing; the box plot anomaly detection algorithm is used to perform anomaly data identification and removal processing; the Butterworth low-pass filtering algorithm is used to perform time series noise suppression processing; and the Z-fraction normalization algorithm is used to perform numerical scale unification processing, outputting the preprocessed liquid cooling regeneration monitoring data.

3. The liquid-cooled data-based cascade waste heat recovery process coordinated control method according to claim 2, characterized in that: The specific steps for constructing a liquid cooling regeneration time-series sample and a liquid cooling regeneration state mapping model based on the preprocessed liquid cooling regeneration monitoring data, and generating regeneration response prediction results are as follows: Based on the preprocessed liquid cooling regeneration monitoring data, the continuous monitoring records are sorted according to the liquid cooling branch identifier and sampling time value, and a sliding time window is constructed with N consecutive sampling times. Within each sliding time window, the coolant temperature value at the cabinet outlet, the instantaneous flow rate value of the liquid cooling branch, the operating frequency value of the variable frequency pump, the speed value of the variable frequency pump, the opening value of the branch electronic expansion valve, and the valve position value of the branch control valve are extracted as the input sequence for the cooling control side. The primary side inlet temperature value of the plate heat exchanger, the inlet water temperature value of the heat pump, the water supply temperature value of the first-stage branch, the water supply temperature value of the second-stage branch, and the water supply temperature value of the third-stage branch are extracted as the output sequence for the regeneration side. The input sequence for the cooling control side and the output sequence for the regeneration side are combined according to the corresponding sampling time to construct the liquid cooling regeneration time series sample. A gated recurrent unit network algorithm is used to construct a liquid cooling regeneration state mapping model. The input sequence of the controlled cooling side in the liquid cooling regeneration time series sample is used as the model input, and the output sequence of the regeneration side corresponding to the next sampling time is used as the model output. Iterative training is performed using the mean square error loss function. The model is considered to have completed training when the decrease in validation set loss is less than the convergence threshold in two consecutive iterations. During the real-time operation phase, the input sequence of the controlled cooling side corresponding to the current sliding time window is input into the trained liquid cooling regenerative state mapping model to obtain the regenerative response prediction result corresponding to the current sampling time.

4. The liquid-cooled data-based cascade waste heat recovery process coordinated control method according to claim 3, characterized in that: The specific steps for comparing the predicted heat recovery response with the actual collected value at the current sampling time to calculate the corresponding prediction deviation value, and then comparing the change in air-cooled side data between the current sampling time and the previous sampling time, are as follows: The predicted heat recovery response is compared with the actual plate heat exchanger primary side inlet temperature, heat pump inlet water temperature, first-stage branch water supply temperature, second-stage branch water supply temperature, and third-stage branch water supply temperature collected at the current sampling time, and the corresponding prediction deviation is calculated. Then compare the cabinet outlet coolant temperature, liquid cooling branch instantaneous flow rate, variable frequency pump operating frequency, variable frequency pump speed, branch electronic expansion valve opening degree, and branch control valve position value with the current sampling time and the previous sampling time.

5. The liquid-cooled data-based cascade waste heat recovery process coordinated control method according to claim 4, characterized in that: The specific steps for performing the tiered acceptance judgment by combining the heat pump operating status value, heat pump outlet water temperature value, first-tier branch water supply temperature value, second-tier branch water supply temperature value, third-tier branch water supply temperature value, residential heat exchange station secondary network return water temperature value, greenhouse air temperature value, and swimming pool makeup water temperature value are as follows: When the heat pump is in the "on" state, and the difference between the heat pump outlet water temperature and the return water temperature of the secondary network of the residential heat exchange station is not less than the first-level acceptance temperature difference threshold, and the difference between the supply water temperature of the first-level branch and the return water temperature of the secondary network of the residential heat exchange station is not less than the first-level delivery temperature difference threshold, a first-level acceptance mark is generated. When the first-level acceptance conditions are not met, and the difference between the supply water temperature of the second-level branch and the air temperature of the greenhouse is not less than the second-level acceptance temperature difference threshold, a second-level acceptance mark is generated. When the first-level and second-level acceptance conditions are not met, and the difference between the supply water temperature of the third-level branch and the swimming pool makeup water temperature is not less than the third-level acceptance temperature difference threshold, a third-level acceptance mark is generated.

6. The collaborative control method for a cascade waste heat recovery process based on liquid cooling data according to claim 5, characterized in that: The specific steps for reading the preprocessed liquid cooling regeneration monitoring data and cascade bearing mapping data, constructing a thermal field evolution prediction model, and predicting the thermal environment state for the future prediction period to obtain thermal environment prediction data are as follows: Read the pre-processed liquid cooling regeneration monitoring data and cascade connection mapping data, construct a thermal environment state matrix by taking the cabinet inlet coolant temperature value, cabinet outlet coolant temperature value, CPU chip surface temperature value, GPU chip surface temperature value, and cabinet internal space temperature value, construct a cooling control adjustment matrix by taking the instantaneous flow value of liquid cooling branch, variable frequency pump operating frequency value, variable frequency pump speed value, and branch electronic expansion valve opening value, and construct a regeneration connection matrix by taking the heat pump operating status value, heat pump inlet water temperature value, heat pump outlet water temperature value, first cascade branch water supply temperature value, second cascade branch water supply temperature value, third cascade branch water supply temperature value, and cascade connection mapping data; Singular value decomposition is performed on the thermal environment state matrix to extract the dominant thermal field modes and construct a reduced POD basis. The thermal environment state matrix at the current sampling time is projected onto the reduced POD basis to obtain the corresponding POD coefficient sequence. The POD coefficient sequence is then concatenated with the cooling control matrix and the heat recovery matrix in the order of sampling time to construct the POD-GA-BPNN prediction sample. A backpropagation neural network algorithm optimized by a genetic algorithm is used to construct a thermal field evolution prediction model based on POD-GA-BPNN prediction samples. The current time sample is used as the model input, and the POD coefficient sequence corresponding to the future prediction period is used as the model output. The initial weights and thresholds of the backpropagation neural network are optimized using a genetic algorithm. Then, the optimized backpropagation neural network is used to predict the POD coefficients for the future prediction period. The predicted POD coefficient sequence is reconstructed with the POD reduced-order basis to obtain the thermal environment prediction data for the future prediction period. The thermal environment prediction data includes the predicted cabinet inlet coolant temperature, predicted cabinet outlet coolant temperature, predicted instantaneous flow rate of the liquid cooling branch, predicted CPU chip surface temperature, predicted GPU chip surface temperature, predicted cabinet internal space temperature, predicted heat pump inlet water temperature, predicted heat pump outlet water temperature, predicted first-stage branch water supply temperature, predicted second-stage branch water supply temperature, and predicted third-stage branch water supply temperature.

7. The liquid-cooled data-based cascade waste heat recovery process coordinated control method according to claim 6, characterized in that: The specific steps for generating the cascade land use classification results for the prediction period based on thermal environment prediction data and cascade land use mapping data are as follows: Read the thermal environment prediction data and cascade connection mapping data. First, remove liquid-cooled branches with a state transfer mismatch flag value of 1 or a generated branch connection mismatch flag. Then, compare the predicted heat pump outlet water temperature of the remaining liquid-cooled branches with the secondary network return water temperature of the residential heat exchange station, the predicted water supply temperature of the second cascade branch with the air temperature of the greenhouse, and the predicted water supply temperature of the third cascade branch with the swimming pool makeup water temperature. Retain the liquid-cooled branches that meet the corresponding cascade connection conditions and generate the cascade connection classification results for the prediction period.

8. The liquid-cooled data-based cascade waste heat recovery process coordinated control method according to claim 6, characterized in that: The specific steps for calculating the cascade coordinated adjustment value based on the liquid-cooled branch are as follows: Multiply the difference between the predicted coolant temperature at the cabinet outlet and the predicted coolant temperature at the cabinet inlet for the i-th liquid cooling branch, the square root of the instantaneous flow rate of the liquid cooling branch plus one, and the arctangent of the predicted heating temperature difference of the corresponding cascade plus one to obtain the adjustment numerator. Add the natural constant e to the difference between the predicted CPU chip surface temperature and the predicted internal temperature of the cabinet for the i-th liquid cooling branch and take the natural logarithm. Add the natural constant e to the difference between the predicted GPU chip surface temperature and the predicted internal temperature of the cabinet for the i-th liquid cooling branch and take the natural logarithm. Add both of these to the constant 1 to obtain the adjustment denominator. Divide the adjustment numerator by the adjustment denominator to obtain the cascade coordinated adjustment value of the i-th liquid cooling branch.

9. The collaborative control method for a cascade waste heat recovery process based on liquid cooling data according to claim 8, characterized in that: The specific steps for determining the collaborative control result and generating corresponding control commands to be issued to the execution unit are as follows: The coordinated regulation values ​​of the liquid cooling branches in each stage are summarized to obtain the total regulation value of the first stage, the total regulation value of the second stage, and the total regulation value of the third stage. Then, the ratio of the coordinated regulation value of each liquid cooling branch to the corresponding total regulation value of the stage is calculated to obtain the stage regulation ratio of each liquid cooling branch. The percentage of cascade regulation is used as the basis for adjusting the opening of the electric regulating valve of the corresponding cascade branch. The total regulation value of the first cascade is used as the basis for determining the start and stop of the heat pump. The sum of the cascade coordinated regulation values ​​of all liquid cooling branches is used as the basis for the linkage adjustment of the variable frequency pump operating frequency value, variable frequency pump speed value and branch electronic expansion valve opening value, so as to determine the coordinated control results in the future prediction period. Based on the results of the coordinated control, when the total regulation value of the first stage exceeds the heat pump start-up judgment threshold and the heat pump operating status is off, a heat pump start-up command is generated. When the total adjustment value of the first stage is lower than the heat pump shutdown judgment threshold and the heat pump operating status is on, a heat pump shutdown command is generated; according to the stage adjustment ratio of each liquid cooling branch, corresponding stage branch electric regulating valve opening adjustment commands are generated, and variable frequency pump speed control commands and branch electronic expansion valve adjustment commands are generated according to the summary result of the stage coordinated adjustment values ​​of all liquid cooling branches; then the generated commands are sent to the corresponding execution components for execution to complete the coordinated distribution control.

10. A liquid cooled data based cascade waste heat recovery process coordinated control apparatus applying the liquid cooled data based cascade waste heat recovery process coordinated control method as claimed in any one of claims 1 to 9, characterized by, The device includes: The data acquisition and processing module is used to acquire liquid cooling heat recovery monitoring data, perform preprocessing on the liquid cooling heat recovery monitoring data, and output the preprocessed liquid cooling heat recovery monitoring data. The state mapping determination module is used to construct liquid cooling recovery time series samples and liquid cooling recovery state mapping models based on preprocessed liquid cooling recovery monitoring data, and generate recovery response prediction results; it combines the changes in the control cooling side adjustment amount and the prediction deviation of the recovery side to generate a state transfer mismatch mark, and generates cascaded transfer mapping data when the state transfer mismatch mark value is 0. The thermal field evolution prediction module is used to read the preprocessed liquid cooling regeneration monitoring data and cascade bearing mapping data, construct a thermal field evolution prediction model, predict the thermal environment state for the future prediction period, and obtain thermal environment prediction data. The collaborative allocation control module is used to generate the classification results of cascade acceptance for the predicted period based on thermal environment prediction data and cascade acceptance mapping data, calculate the cascade collaborative adjustment value according to the liquid cooling branch, determine the collaborative control result, and generate corresponding control commands to be sent to the execution unit for execution, so as to complete the collaborative control of the cascade waste heat recovery process.

Images