A multi-device parallel cooling method for a cryogenic BOG reliquefaction device

Abstract

The application discloses a kind of deep cold BOG re-liquefaction device multi-device parallel cooling method, it is related to industrial cooling system technical field, comprising: fusion adaptive fuzzy PID control and graph convolutional neural network, real-time monitoring and predicting the heat load of each cooling equipment of deep cold BOG re-liquefaction device, generates dynamic cooling demand atlas;Based on heat load prediction result, dynamically adjusts the cooling water flow distribution strategy of each parallel branch.The application realizes high-precision real-time prediction and modeling to the dynamic heat load of multi-device by fusion adaptive fuzzy PID control and graph convolutional neural network, forms dynamic cooling demand atlas, solves the uneven cooling problem under the traditional fixed-flow distribution mode, intelligently generates cooling water flow distribution strategy according to the real-time heat load difference of each device, realizes accurate cooling on demand, effectively avoids local overheating or overcooling phenomenon, significantly improves equipment operation efficiency and service life, guarantees long-term stable operation of system.

Description

Technical Field

[0001] This invention relates to the field of industrial cooling system technology, specifically to a method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit. Background Technology

[0002] In the liquefied natural gas (LNG) industry, BOG (boiling gas) refers to the phenomenon where some liquefied natural gas evaporates into gas during liquid storage due to temperature and pressure changes. This gas needs to be effectively treated to avoid excessive pressure during storage and transportation. This requires cryogenic BOG reliquefaction units to improve the efficiency and flexibility of the gas reliquefaction process. During operation, the core equipment (multi-stage compressors, motors, frequency converters, etc.) of the cryogenic BOG reliquefaction unit generates a large amount of heat. If heat cannot be dissipated effectively in a timely manner, it will lead to decreased equipment operating efficiency, accelerated component wear, and even safety failures. Therefore, it is necessary to design an efficient, stable, and energy-saving cooling water system for the equipment requiring cooling (shell-and-tube heat exchangers after the two compression stages, shell-and-tube heat exchangers for motor cooling gas, main unit casing, and frequency converter cabinet) to ensure the long-term safe and reliable operation of the unit.

[0003] During the operation of the cooling water system of a cryogenic BOG reliquefaction unit, the dynamic heat loads of different equipment requiring cooling vary significantly and fluctuate frequently. This leads to uneven cooling of each branch under the traditional constant flow distribution mode, causing local overheating or undercooling, affecting overall energy efficiency and equipment lifespan. Furthermore, the frequent system adjustments cause pipeline pressure pulsation, generating water hammer vibration and noise, which threatens pipeline sealing and the operational stability of precision equipment. At the same time, during pressure pulsation adjustments, differences in heat exchange efficiency and flow rate among the branches still result in uneven return water temperature, causing continuous fluctuations in the recooling load of the cooling tower, reducing system heat recovery efficiency, and increasing overall energy consumption. Therefore, a multi-equipment parallel cooling method for cryogenic BOG reliquefaction units is proposed to solve the above-mentioned problems. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit, comprising the following steps:

[0005] Step 1: Integrate adaptive fuzzy PID control with graph convolutional neural network to monitor and predict the heat load of each cooling equipment in the cryogenic BOG reliquefaction unit in real time, generate a dynamic cooling demand map, provide a basis for flow allocation, and thus accurately predict the instantaneous cooling demand of each equipment, and realize the decision basis for on-demand cooling.

[0006] Step 2: Based on the heat load prediction results, dynamically adjust the cooling water flow distribution strategy of each parallel branch, and implement on-demand distribution and precise cooling strategy through intelligent valves to eliminate local overheating or overcooling, thereby ensuring uniform temperature field of all equipment and effectively extending the service life of core components.

[0007] Step 3: Introduce a pressure wave suppression frequency domain analysis model and a reinforcement learning adaptive damping algorithm to predict water hammer vibration caused by flow regulation in real time, dynamically adjust valve action and pipeline damping, smooth pressure fluctuations, reduce noise and vibration, thereby significantly reducing pipeline impact and mechanical noise, and improving system operation stability and quietness.

[0008] Step 4: By deploying a group of temperature sensors at the return water inlets of each branch, a distributed temperature field reconstruction algorithm is used to calculate the return water temperature distribution in real time, identify areas with uneven temperature distribution, and dynamically adjust the flow compensation between branches based on a co-evolution strategy to achieve equalization of return water temperature, making the return water temperature highly uniform and providing a stable heat load input for the downstream cooling tower.

[0009] Step 5: Based on the return water temperature balance result, dynamically adjust the cooling tower fan speed and system bypass ratio to maximize waste heat recovery, stabilize the cooling tower heat load, reduce overall energy consumption, significantly improve the overall system energy efficiency, and achieve synergistic optimization of waste heat recovery and energy consumption.

[0010] Step 6: Establish a digital twin model of the cooling system of the cryogenic BOG reliquefaction unit, evaluate the system's energy efficiency ratio and stability in real time, form a closed-loop control system, continuously iterate and optimize the cooling strategy, ensure the long-term stable and efficient operation of the system, promote the system to achieve continuous intelligent evolution of self-learning and self-optimization, and ensure long-term efficient and reliable operation.

[0011] Preferably, step 1 specifically includes:

[0012] Temperature sensors and flow meters are installed at the inlet, outlet, and key heat exchange surfaces of each cooling equipment in the cryogenic BOG reliquefaction unit to collect real-time heat load-related time-series data, establish an equipment operation status monitoring network, and achieve continuous and comprehensive perception of the entire system's operation status.

[0013] A graph structure is constructed based on the collected heat load related time series data, where nodes represent equipment or measuring points and edges represent heat transfer relationships. The thermal coupling characteristics between equipment are extracted through graph convolutional neural networks, and the short-term heat load change trend is predicted. The heat load prediction value is output, accurately capturing the thermal coupling relationship between equipment and improving the accuracy of heat load prediction.

[0014] The prediction results are input into the adaptive fuzzy PID controller, which, combined with the real-time operating conditions of the equipment and environmental parameters, generates a dynamic cooling demand map. This provides a feedforward control basis for the flow distribution of each branch, making the cooling control strategy forward-looking and significantly improving the timeliness of response.

[0015] Preferably, step 1 further includes:

[0016] Based on the heat load prediction value output by the graph convolutional neural network, combined with the equipment heat capacity parameters and heat exchanger efficiency model, the theoretical cooling water demand of each equipment requiring cooling under the current operating conditions is calculated, which significantly improves the accuracy of cooling water volume calculation and ensures cooling efficiency from the source.

[0017] Based on an adaptive fuzzy rule base, the prediction error is corrected online, the PID control parameters are optimized, the heat load fluctuations are quickly tracked and suppressed, the system overshoot and lag are effectively suppressed, and the temperature control response is rapid and stable.

[0018] The revised cooling demand is output in the form of a spatial distribution map, namely a dynamic cooling demand map. At the same time, the target flow rate and temperature difference of each branch are marked in the dynamic cooling demand map and associated with the intelligent valve control command queue to realize the visualization and command of the cooling strategy, thereby improving the intuitiveness of system control and the consistency of execution.

[0019] Preferably, step 2 specifically includes:

[0020] Receive the target flow signals of each branch from the dynamic cooling demand map, and calculate the target adjustment amount of each smart valve by combining the current pipeline pressure and valve opening status, so as to ensure the accuracy of flow distribution and effectively improve the matching degree of cooling supply of each branch.

[0021] The PLC sends control commands to the intelligent valves of each branch, and the valve opening is adjusted by a segmented easing strategy to achieve smooth flow transition, effectively avoid sudden flow changes, and ensure the stability of the hydraulic process of the system and the safety of the equipment.

[0022] During the flow regulation process, the inlet water temperature and temperature difference of each cooling device are monitored in real time. The valve position is finely adjusted through closed-loop feedback to ensure that the cooling effect of each cooling device meets the set temperature control range. The cooling effect is continuously corrected to ensure that the key equipment is always in the optimal operating temperature range.

[0023] Preferably, step 3 specifically includes:

[0024] A frequency domain analysis model for pressure wave suppression, which includes pipeline geometric parameters, fluid properties, and valve characteristics, is established to simulate the pressure response spectrum under different flow step changes. This model can accurately predict the dynamic pressure response characteristics of the system under different operating conditions.

[0025] Based on real-time flow regulation commands, the frequency and amplitude of water hammer caused by pressure wave suppression are predicted through a frequency domain analysis model, identifying high-risk periods of pressure fluctuations, achieving accurate early warning of potential water hammer risks, and providing a key time window for proactive intervention;

[0026] By combining reinforcement learning adaptive damping algorithm, the action curve of smart valve is pre-adjusted before the pressure peak is predicted, and the impedance parameters of hydraulic damper are dynamically adjusted at key nodes in the pipeline, which effectively reduces the peak pressure fluctuation and vibration energy, and significantly improves the stability of system operation.

[0027] Preferably, step 3 further includes:

[0028] Using pressure fluctuation amplitude, vibration acceleration and noise level as evaluation indicators, a damping control reward function is constructed. A damping adjustment strategy is trained through a deep Q-learning network, integrating multiple physical quantities into a single evaluation value, which significantly improves the comprehensive optimization capability of the vibration suppression strategy.

[0029] Before and after each valve action, pipeline pressure and vibration sensor data are collected, input into a trained strategy network, and the optimal damper adjustment command is output to achieve adaptive damping adjustment based on real-time status, effectively coping with transient water hammer conditions.

[0030] The strategy network parameters are updated based on the real-time adjustment effect, enabling the damping system to adaptively optimize for varying operating conditions, continuously suppressing water hammer vibration and structural noise. Through an online learning mechanism, it can continuously evolve over time and maintain excellent vibration suppression and noise reduction performance over the long term.

[0031] Preferably, step 4 specifically includes:

[0032] High-density temperature sensor arrays are deployed at the return water inlets of each branch to collect spatiotemporal distribution data of return water temperature, construct a two-dimensional temperature field observation matrix, obtain the return water temperature distribution in real time, and provide a high-resolution data foundation for full-field temperature monitoring.

[0033] A distributed temperature field reconstruction algorithm based on Kalman filtering and radial basis function interpolation is used to reconstruct the temperature field of the return water with high precision. It identifies regions with uneven temperature distribution and their distribution characteristics. These regions are local high-temperature and low-temperature areas. It accurately identifies areas with abnormal temperatures and provides reliable spatial temperature information for targeted regulation.

[0034] By combining a co-evolution strategy, a temperature compensation mechanism is established between adjacent branches. By dynamically adjusting the flow distribution weight, the return water temperature field tends to be more uniformly distributed, thereby achieving coordinated temperature regulation between branches and effectively improving the overall uniformity of the return water temperature field.

[0035] Preferably, step 4 further includes:

[0036] Each branch is treated as a population, and the uniformity index of the return water temperature field is used as the fitness function. The flow compensation coefficient of each branch is optimized by a genetic algorithm to improve the overall optimization efficiency and quickly approach the optimal temperature field distribution.

[0037] Within each control cycle, fitness is calculated based on the temperature field reconstruction results. Superior individuals are selected for crossover and mutation operations to generate a new generation of flow allocation schemes. Superior individuals are branch combinations, which enhance the algorithm's global search capability and avoid getting trapped in local optima.

[0038] The optimized flow compensation command is sent to the intelligent valves of each branch, which quickly converges and maintains the return water temperature field in a steady state, thereby achieving dynamic balance and stable control of the temperature field and improving the thermal management quality of the system.

[0039] Preferably, step 5 specifically includes:

[0040] The total residual heat of the system is calculated based on the balanced return water temperature field. Combined with the cooling tower performance curve and the current ambient wet-bulb temperature, the theoretical heat dissipation requirements are determined, which effectively avoids energy waste and equipment damage caused by overcooling or overheating of the cooling tower, and significantly improves heat dissipation efficiency.

[0041] By adjusting the cooling tower fan speed using a frequency converter, the actual heat dissipation capacity is matched with the theoretical requirement. At the same time, the real-time relationship between fan energy consumption and heat dissipation efficiency is monitored. Under the premise of ensuring that the heat dissipation requirement is met, the operating energy consumption of the cooling tower fan is directly reduced by optimizing the fan speed, thereby achieving the economic efficiency of system operation.

[0042] Based on the total system load and return water temperature distribution, the opening of the bypass pipeline valve from the motor cooling heat exchanger to the main unit casing is dynamically adjusted to achieve the cascade utilization of waste heat and the active balance of system heat load. By actively scheduling waste heat, the starting frequency and operating load of the cooling tower are reduced under low load conditions, further reducing the overall energy consumption of the system and improving the comprehensive utilization rate of thermal energy.

[0043] Preferably, step 6 specifically includes:

[0044] Based on physical modeling and data-driven methods, a full-element digital twin model of the cooling system of a cryogenic BOG reliquefaction unit, including the pipeline network, heat exchange equipment, pumps, valves and control system, is constructed to establish a high-precision virtual system and provide a reliable simulation platform for prediction and optimization.

[0045] By synchronizing on-site operational data to the digital twin model in real time via the MQTT protocol, the consistency and dynamic mapping between the virtual system and the actual system are achieved, ensuring real-time consistency between the digital model and the actual system and improving the accuracy of monitoring and diagnosis.

[0046] The model predictive control algorithm is run in a twin environment to simulate the system energy efficiency and stability indicators under different cooling strategies. The optimized control parameters are then deployed online to the actual control system to form a closed-loop optimization system, thereby realizing the pre-verification and online optimization of the control strategy and continuously improving the system energy efficiency and stability.

[0047] This invention provides a method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit. It offers the following advantages:

[0048] (I) This method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit achieves high-precision real-time prediction and modeling of the dynamic heat load of multiple devices by integrating adaptive fuzzy PID control and graph convolutional neural network, forming a dynamic cooling demand map, solving the problem of uneven cooling in the traditional constant flow distribution mode. Based on the real-time heat load difference of each device, a cooling water flow distribution strategy is intelligently generated to achieve precise cooling on demand, effectively avoiding local overheating or overcooling, significantly improving equipment operating efficiency and service life, and ensuring long-term stable operation of the system.

[0049] (II) The parallel cooling method for multiple devices in a cryogenic BOG reliquefaction unit introduces a pressure wave suppression frequency domain analysis model and a reinforcement learning adaptive damping adjustment mechanism. It can actively predict and suppress water hammer vibration and pressure pulsation caused by flow regulation. Through dynamic collaborative optimization of valve action curves and pipeline damping, it can effectively smooth pressure fluctuations, reduce noise and vibration, thereby protecting pipeline sealing and equipment structural integrity, and improving the system's operational stability and safety under frequent adjustment conditions.

[0050] (III) The multi-equipment parallel cooling method of the cryogenic BOG reliquefaction device adopts a distributed temperature field reconstruction and co-evolution optimization strategy to achieve real-time monitoring and high-precision balanced control of the return water temperature field. Through the fusion of sensor array and intelligent algorithm, it can dynamically identify areas with uneven temperature distribution and optimize the flow compensation between branches based on genetic algorithm, so that the return water temperature can quickly converge to the set range. This not only improves the stability of the cooling tower heat load, but also creates uniform temperature conditions for waste heat recovery. Attached Figure Description

[0051] Figure 1 This is a control logic timing diagram of a multi-equipment parallel cooling method for a cryogenic BOG reliquefaction device according to the present invention.

[0052] Figure 2 This is a schematic diagram of the working process of a multi-equipment parallel cooling method for a cryogenic BOG reliquefaction device according to the present invention.

[0053] Figure 3 This is a schematic diagram of the process flow for a multi-device parallel cooling method in a cryogenic BOG reliquefaction unit according to the present invention. Detailed Implementation

[0054] 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.

[0055] Example 1, please refer to Figures 1 to 3 This invention provides a technical solution: a method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit, comprising the following steps:

[0056] Step 1: Integrating adaptive fuzzy PID control and graph convolutional neural network, the heat load of each cooling device in the cryogenic BOG reliquefaction unit is monitored and predicted in real time, generating a dynamic cooling demand map to provide a basis for flow allocation. This allows for accurate prediction of the instantaneous cooling demand of each device, providing a basis for on-demand cooling decisions. Temperature sensors and flow meters are installed at the inlet, outlet, and key heat exchange surfaces of each cooling device in the cryogenic BOG reliquefaction unit to collect real-time heat load-related time-series data and establish a device operation status monitoring network. This enables continuous and comprehensive perception of the entire system's operating status. Based on the collected heat load-related time-series data, a graph structure is constructed, where nodes represent devices or measuring points, and edges represent heat transfer relationships. The graph convolutional neural network extracts the thermal coupling characteristics between devices and predicts short-term heat load change trends, outputting predicted heat load values. This accurately captures the thermal coupling relationship between devices, improving the accuracy of heat load prediction. The prediction results are input into the adaptive fuzzy PID controller, combined with real-time device operating conditions and environmental parameters, to generate a dynamic cooling demand map. This provides a feedforward control basis for flow allocation in each branch, making the cooling control strategy forward-looking and significantly improving response timeliness.

[0057] Furthermore, in the cooling water system of the cryogenic BOG reliquefaction unit, Pt100 temperature sensors and electromagnetic flow meters are installed at the cooling water inlet and outlet of each cooling device, including the shell-and-tube heat exchanger after the first-stage compressor, the shell-and-tube heat exchanger after the second-stage compressor, the motor cooling gas shell-and-tube heat exchanger, the main unit casing, and the frequency converter cabinet. Additional temperature measurement points are also deployed on key heat exchange surfaces (such as heat exchanger tube sheets and high-temperature areas of the main unit casing), forming a monitoring network covering the entire system's equipment operation status. The data acquisition module collects real-time temperature and flow time-series data from each measurement point at a frequency of 1Hz and transmits it to the central control unit via industrial Ethernet. Based on the real-time collected temperature and flow time-series data, a graph structure is constructed with equipment and measurement points as nodes and heat transfer relationships as edges. The input dimension of the graph convolutional neural network model is the number of node features (including real-time temperature, flow rate, equipment power, and ring power). The model extracts thermal coupling features between devices using a three-layer graph convolutional layer (based on ambient temperature and humidity), outputting predicted heat load values ​​for each device within the next 10 minutes. The prediction mean square error is controlled within 5%. The model is trained using historical operating datasets, covering various typical operating conditions such as high temperatures in summer, low temperatures in winter, and transitional seasons, ensuring that the prediction model has good generalization ability and robustness. The predicted heat load values ​​output by the graph convolutional neural network are input into an adaptive fuzzy PID controller. The controller combines real-time operating parameters of the devices and environmental parameters (ambient temperature and relative humidity), and dynamically adjusts the proportional, integral, and derivative coefficients according to a preset fuzzy rule base containing 49 rules. The controller outputs the target flow rate and inlet / outlet water temperature difference of cooling water for each branch with a control cycle of 10 seconds, and generates a dynamic cooling demand map in the form of a spatial distribution map. This map serves as the basis for feedforward control and is sent to the intelligent valve controllers of each branch via the OPCUA protocol.

[0058] In addition, step 1 also includes: based on the heat load prediction value output by the graph convolutional neural network, combined with the equipment heat capacity parameters and heat exchanger efficiency model, calculating the theoretical cooling water demand of each cooling equipment under the current operating conditions, significantly improving the accuracy of cooling water quantity calculation, ensuring cooling efficiency from the source, correcting prediction errors online based on an adaptive fuzzy rule base, optimizing PID control parameters, quickly tracking and suppressing heat load fluctuations, effectively suppressing system overshoot and lag, ensuring rapid and stable temperature control response, and outputting the corrected cooling demand in the form of a spatial distribution map, i.e., a dynamic cooling demand map. At the same time, marking the target flow rate and temperature difference of each branch in the dynamic cooling demand map and associating it with the intelligent valve control command queue, realizing the visualization and command of the cooling strategy, and improving the intuitiveness and execution consistency of system control;

[0059] Furthermore, during actual system operation, the central control unit calculates the theoretical cooling water demand of each cooling device under current operating conditions based on the predicted heat load values ​​of each cooling device output by the graph convolutional neural network model, combined with the known heat capacity parameters and heat exchanger efficiency model of each cooling device. Specifically, based on the fundamental heat transfer formula Q=3600P / (cρΔt), where Q is the cooling water flow rate (m³ / s). 3 / h), P is the cooling load (kW), c is the specific heat capacity of water (4.186 kJ / kg·℃), and ρ is the density of water (1000 kg / m³). 3 Δt represents the temperature difference between the inlet and outlet of the cooling water (10℃). After the theoretical cooling water demand of each piece of equipment requiring cooling is calculated, the result is used as the basis for cooling water flow allocation. Combined with the real-time collected equipment inlet and outlet temperature and pressure data, a preliminary cooling strategy is formed. After the theoretical cooling water demand is calculated, the prediction error is corrected online through an adaptive fuzzy rule base to optimize the PID control parameters and achieve rapid tracking and suppression of heat load fluctuations. During operation, the central control unit continuously compares the short-term prediction results (10-minute prediction cycle) of the graph convolutional neural network with the actual collected equipment heat load data (sampling frequency 1Hz) to calculate the real-time prediction error. Based on the fuzzy rule base consisting of 49 rules, the proportional, integral, and derivative coefficients of the PID controller are dynamically adjusted with an adjustment cycle of 10 seconds. When the prediction error is detected to exceed the set threshold (±5%), the integral coefficient is adjusted first to eliminate the error. Steady-state error; when a rapid change in heat load is detected, the proportional coefficient and derivative coefficient are adjusted accordingly to improve the response speed and reduce cooling lag or overcooling caused by sudden changes in equipment operating conditions; after error correction and parameter optimization, the final and more accurate cooling demand is generated as a dynamic cooling demand map in the form of a spatial distribution map. This map uses the system topology as the base map and clearly marks the target flow rate and target inlet and outlet temperature difference of each cooling branch (including the five main branches of the first-stage compressor, the second-stage compressor, motor cooling, the main unit casing, and the inverter cabinet). The map information is sent to the intelligent valve controllers on each branch in real time through the OPCUA protocol and directly written into the controller's instruction queue. The intelligent valves perform precise opening adjustment based on the received target flow rate value, combined with their built-in flow-opening curve and current pipeline pressure feedback, to achieve distributed and refined closed-loop control of the system's cooling water flow rate;

[0060] It should be noted that the basic design parameters are shown in Table 1.

[0061] Table 1 Basic Design Parameters:

[0062] .

[0063] Step 2: Based on the heat load prediction results, dynamically adjust the cooling water flow distribution strategy of each parallel branch. Intelligent valves execute on-demand distribution and precise cooling strategies to eliminate local overheating or overcooling, thereby ensuring uniform temperature across all equipment and effectively extending the service life of core components. Receive target flow signals from each branch based on the dynamic cooling demand map, and calculate the target adjustment amount of each intelligent valve based on the current pipeline pressure and valve opening status to ensure the accuracy of flow distribution and effectively improve the matching degree of cooling supply to each branch. Send control commands to the intelligent valves of each branch via PLC, and use a segmented easing strategy to adjust the valve opening to achieve smooth flow transition, effectively avoiding sudden flow changes and ensuring the stability of the system's hydraulic process and equipment safety. During flow adjustment, monitor the inlet water temperature and temperature difference of each equipment requiring cooling in real time, and fine-tune the valve position through closed-loop feedback to ensure that the cooling effect of each equipment meets the set temperature control range. Continuously correct the cooling effect to ensure that key equipment is always in the optimal operating temperature range.

[0064] Furthermore, after receiving the target flow signals for each branch from the dynamic cooling demand map, the central control unit calculates the target adjustment amount for each intelligent valve based on the current actual operating parameters of each branch. Specifically, it synchronously reads the real-time opening feedback signals of the intelligent valves, the current flow values ​​of each branch (from electromagnetic flowmeters), and the pressure values ​​(pressure transmitters) of key nodes in the main pipeline. Based on the inherent flow characteristic curves of the valves, the real-time resistance characteristics of the pipeline, and the difference between the target flow and the current flow, it calculates the target adjustment amount. Simultaneously, it compensates for the impact of pipeline pressure fluctuations on the valve flow capacity, ensuring the accuracy of the target adjustment command. The calculation process is completed within each control cycle (set to 10 seconds). A set of instructions, including the target opening degree and desired operating speed of each valve, is generated and written to a designated data register of the central PLC via industrial Ethernet, preparing for instruction transmission. After acquiring the target adjustment instruction set, the central PLC executes a segmented easing control strategy to achieve a smooth flow transition. The core of this segmented easing control strategy is to decompose the entire valve's movement process into several sub-stages, setting different operating speeds and dwell times for each stage. Specifically, for cases where the target opening degree change exceeds 20%, the movement process is divided into three stages: acceleration, constant speed, and deceleration. The duration of each stage is set to 30%, 40%, and 30% of the total movement time, respectively, based on the valve's mechanical characteristics. The constant speed stage speed is set to... The control command, operating at 60%-80% of the valve's maximum permissible speed, is transmitted to each intelligent valve positioner via the Profibus-DP fieldbus protocol. Simultaneously, the PLC continuously monitors the valve opening feedback signal and compares it in real-time with the command value. If the deviation between the feedback value and the target value exceeds the set dead zone (±1% of the full stroke), a deviation alarm is triggered, and compensation adjustments are made according to preset logic to ensure the valve moves precisely along the predetermined trajectory. During flow regulation, the cooling effect is verified and fine-tuned through a real-time monitoring network. Pt100 temperature sensors at the cooling water inlet and outlet of each cooling device continuously collect water temperature data at a frequency of 1Hz. The LC calculates the real-time temperature difference, which is compared with the target temperature difference range set in the dynamic cooling demand map. When the measured temperature difference deviates from the target range for more than a preset time threshold, it is determined that the initial flow distribution has not fully met the cooling demand. Then, the fine-tuning program is started. The fine-tuning program takes the deviation between the measured temperature difference and the target value as input and calculates the small compensation amount for the opening of the corresponding branch smart valve through an independent proportional-integral controller. The adjustment range is limited to ±2% of the opening. Then, the compensation command is superimposed on the original target opening command to finely correct the valve position. The program continues to run until the measured temperature difference at the inlet and outlet of all equipment stably falls within their respective set target ranges.

[0065] Step 3 introduces a pressure wave suppression frequency domain analysis model and a reinforcement learning adaptive damping algorithm to predict water hammer vibration caused by flow regulation in real time, dynamically adjust valve actions and pipeline damping, smooth pressure fluctuations, and reduce noise and vibration. This significantly reduces pipeline impact and mechanical noise, improves system stability and quietness, and establishes a pressure wave suppression frequency domain analysis model that includes pipeline geometric parameters, fluid properties, and valve characteristics. It simulates the pressure response spectrum under different flow step changes and can accurately predict the dynamic pressure response characteristics of the system under different operating conditions. Based on real-time flow regulation commands, the water hammer frequency and amplitude are predicted by the pressure wave suppression frequency domain analysis model, identifying high-risk periods of pressure fluctuations and achieving accurate early warning of potential water hammer risks. This provides a key time window for proactive intervention. Combined with the reinforcement learning adaptive damping algorithm, the intelligent valve action curve is pre-adjusted before the pressure wave peak is predicted, and the hydraulic damper impedance parameters are dynamically adjusted at key pipeline nodes, effectively reducing the pressure fluctuation peak and vibration energy, and significantly improving the stability of system operation.

[0066] Furthermore, by integrating the actual geometric parameters of the pipeline network, the physical properties of the cooling water, and the dynamic characteristics of the intelligent valves, a frequency domain analysis model for pressure wave suppression is constructed. In practice, precise input is required, including the actual length, diameter, wall thickness, material (Q235B welded steel pipe, 304 stainless steel pipe), and layout of the main water supply pipeline (DN250) and each branch pipeline (DN50-DN150). The total equivalent length of the pipeline system is calculated, and fluid properties are set according to the design conditions: cooling water density is taken as 998 kg / m³. 3 (Based on 20℃), kinematic viscosity is taken as 1.0×10⁻⁶. -6 m² / s, with the bulk modulus of the fluid taken as 2.2 GPa, the valve characteristics are determined according to the technical manual of the selected intelligent control valve. Its flow coefficient, nominal diameter, stroke time and flow characteristic curve are input, and the one-dimensional unsteady flow equations are solved. The pressure transient response of each key node (pump outlet, heat exchanger inlet, pipe bend, etc.) under different flow step commands is simulated, and its pressure fluctuation spectrum is output to identify the inherent dangerous resonant frequency of the system.

[0067] The formula for calculating the resonant frequency of the pressure fluctuation spectrum is as follows:

[0068] or ;

[0069] ;

[0070] In the formula: Represents the system's first The first resonant frequency (or natural frequency) is determined when the external excitation frequency is close to or equal to a certain value. When this happens, the system will resonate, and the amplitude of pressure fluctuations will be significantly amplified, which is the most dangerous frequency region. The order of the resonant frequency. , which represents the half-wave number of the standing wave formed by the pressure wave in the pipeline; The propagation speed of pressure waves (water hammer waves) in a fluid-filled pipe is also known as water hammer wave velocity. The bulk modulus of the fluid; The density of the cooling water; This refers to the inner diameter of the pipe. The elastic modulus of the pipe wall material; The thickness of the pipe wall; To generate the characteristic tube length for resonance; using or As the denominator, depending on the boundary conditions of the pipeline, when one end is a closed end (or the valve closes quickly) and the other end is a reservoir (constant pressure end), the resonant frequency formula is: At this time This refers to the length of the pipe from the valve to the reservoir (or to another reflecting boundary); when both ends are closed (or both are fast-closing valves), the resonant frequency formula is... At this time This refers to the pipe length between closed ends. During system operation, the central control unit, based on real-time flow regulation commands (derived from dynamic cooling demand maps), drives the pressure wave suppression frequency domain analysis model to perform online forward simulation. The model uses the current steady-state operating conditions as initial conditions and the planned valve action curve as input excitation. After calculation, the model predicts in real time the main frequency components and peak amplitude of the water hammer pressure wave that may be induced at a specific location in the pipeline by this regulation action. Then, it compares the prediction results with the preset safety threshold to identify high-risk periods of pressure fluctuation, i.e., the time window when the predicted pressure peak exceeds the threshold limit by 70% or the predicted frequency falls into the system resonance band. This high-risk period information, along with the specific valve identification and the predicted peak arrival time, generates an early warning message and pushes it to subsequent processes to provide accurate time targets and tuning targets. After obtaining the water hammer risk prediction, it immediately starts... The reinforcement learning adaptive damping algorithm intervenes to minimize the pressure fluctuation amplitude, valve action smoothness, and energy consumption as a comprehensive reward objective. It pre-optimizes the original valve action curve to be executed: before the predicted pressure wave peak arrives, the valve action speed curve is finely adjusted to change the excitation phase of the pressure wave. At the same time, a real-time impedance parameter adjustment command is sent to the adjustable hydraulic damper deployed at the key nodes of the pipeline. According to the command, the adjustable hydraulic damper dynamically adjusts its flow channel impedance from the default state to the optimal value during high-risk periods to absorb the pressure fluctuation energy of specific frequencies. The entire adjustment process is completed within a single control cycle (10 seconds). The valve action curve and damper parameters are dynamically generated based on the real-time prediction results. After the action is completed, the reinforcement learning strategy network is updated online based on the fluctuation data fed back by the actual pressure sensor to achieve adaptive evolution of the vibration suppression strategy.

[0071] In addition, step 3 also includes: constructing a damping control reward function using pressure fluctuation amplitude, vibration acceleration, and noise level as evaluation indicators; training a damping adjustment strategy through a deep Q-learning network; integrating multiple physical quantities into a single evaluation value; significantly improving the comprehensive optimization capability of the vibration suppression strategy; collecting pipeline pressure and vibration sensor data before and after each valve action; inputting the trained strategy network; outputting the optimal damper adjustment command; realizing adaptive damping adjustment based on real-time status; effectively coping with transient water hammer conditions; updating strategy network parameters according to real-time adjustment effects; realizing adaptive optimization of the damping system for changing conditions; continuously suppressing water hammer vibration and structural noise; and through an online learning mechanism, continuously evolving over time and maintaining excellent vibration suppression and noise reduction performance over the long term.

[0072] Furthermore, a damping control reward function is constructed, using pressure fluctuation amplitude, pipeline vibration acceleration, and noise level as comprehensive evaluation indicators, to quantify the effect of each control action. Specifically, the reward function is defined in the central control unit. for: ,in, To control the deviation between the maximum pressure fluctuation amplitude and the steady-state value at key monitoring points within the control period, its weighting coefficient Set it to 0.5; The root mean square acceleration values ​​measured by vibration sensors at the same location, weighted by a coefficient. Set it to 0.3; The A-weighted sound pressure level measured at a distance of 1 meter from the device, with weighting coefficients. Set it to 0.1; The weighting factor represents the estimated energy savings relative to the baseline strategy due to optimized valve action curves. The value is set to 0.1 to encourage energy conservation. The negative sign indicates that the first three terms are penalty terms that need to be minimized. Subsequently, a deep Q-learning network is used for offline training. The network input state is... This is a 48-dimensional vector containing the current pipeline pressure distribution (6 measuring points), valve pre-action commands, and system flow requirements; the output action... To correct the impedance setpoint (discretized into 20 levels) of the adjustable hydraulic damper and the valve speed curve (0.8 to 1.2), a preliminary damping adjustment strategy network model was obtained by using historical operation datasets, exploring with an ε-greedy strategy, and training the network using experience replay until the Q-value converged. During actual operation, this damping control strategy is triggered whenever a valve adjustment command is generated based on the dynamic cooling demand map. 100 milliseconds before the valve action begins, the central control unit collects real-time status data, including readings from six pressure transmitters on the main water supply pipeline after the pump, before each main branch branch bifurcation point, and at the inlet of the target heat exchanger, as well as data from a 3-axis vibration acceleration sensor mounted on the pipeline support. After preprocessing (including a 1Hz high-pass filter to remove DC components and a 5th-order Butterworth low-pass filter to eliminate high-frequency noise), the real-time status data, together with the valve command to be executed, constitutes a real-time status vector. The vector is input into the pre-trained deep Q-network preliminary damping adjustment strategy network model for forward computation. From all possible action combinations, the set of actions with the highest predicted Q value is selected as the optimal output. The output instruction set specifically includes: the target impedance values ​​specified for the two adjustable hydraulic dampers on the relevant pipe section (set to 350 Pa·s / m respectively). 3 and 180 Pa·s / m 3The instructions include a speed correction coefficient for the main control valve's action curve (adjusting the constant speed segment speed to 70% of the maximum allowable speed). Each instruction is sent to the corresponding actuator via the Profibus-DP bus before the valve action begins, enabling proactive intervention. An online learning mechanism is designed to collect pressure, vibration, and noise data from the same measuring points again within a complete evaluation window after each valve control action (10 seconds after the action ends), calculating the actual reward for that action. In specific calculations, Take the maximum overshoot during the pressure stabilization process after the action. Take the root mean square value of the vibration acceleration during the action. Take the average sound pressure level during the action, and simultaneously record the actual actions performed. and the new state after the action , quadruple A circular replay buffer with a capacity of 10,000 experience points is stored. Every 100 control cycles (approximately 1,000 seconds), the background learning thread of the central control unit randomly samples a batch (size 64) of historical experience data from the buffer and incrementally updates the parameters of the deep Q-network. The update adopts double Q-learning and target network technology to stabilize the training process. The learning rate is set to 0.0001. Through continuous online learning, the policy network gradually adapts to the latest dynamic characteristics of the system, enabling the damping adjustment command to continuously optimize itself and achieve long-term, adaptive, and efficient suppression of water hammer vibration and structural noise.

[0073] Step 4: By deploying a group of temperature sensors at the return water inlets of each branch, a distributed temperature field reconstruction algorithm is used to calculate the return water temperature distribution in real time, identify areas with uneven temperature distribution, and dynamically adjust the flow compensation between branches based on a co-evolution strategy to achieve equalization of return water temperature, making the return water temperature highly uniform and providing a stable heat load input for the downstream cooling tower.

[0074] Step 5: Based on the return water temperature balance result, dynamically adjust the cooling tower fan speed and system bypass ratio to maximize waste heat recovery, stabilize the cooling tower heat load, reduce overall energy consumption, significantly improve the overall system energy efficiency, and achieve synergistic optimization of waste heat recovery and energy consumption.

[0075] Step 6: Establish a digital twin model of the cooling system of the cryogenic BOG reliquefaction unit, evaluate the system's energy efficiency ratio and stability in real time, form a closed-loop control system, continuously iterate and optimize the cooling strategy, ensure the long-term stable and efficient operation of the system, promote the system to achieve continuous intelligent evolution of self-learning and self-optimization, and ensure long-term efficient and reliable operation.

[0076] Example 2, as Figures 1 to 3As shown, based on Embodiment 1, the present invention provides a technical solution: Step 4 specifically includes: arranging a high-density temperature sensor array at the return water inlet of each branch, collecting spatiotemporal distribution data of return water temperature, constructing a two-dimensional temperature field observation matrix, obtaining the return water temperature distribution in real time, providing a high-resolution data foundation for full-field temperature monitoring, using a distributed temperature field reconstruction algorithm based on Kalman filtering and radial basis function interpolation to perform high-precision reconstruction of the return water temperature field, identifying uneven temperature distribution areas and their distribution characteristics, the uneven temperature distribution areas being local high-temperature areas and low-temperature areas, accurately identifying abnormal temperature areas, providing reliable spatial temperature information for targeted adjustment, and combining a co-evolution strategy to establish a temperature compensation mechanism between adjacent branches, by dynamically adjusting the flow distribution weight, promoting the return water temperature field to tend towards a uniform distribution, realizing coordinated temperature adjustment between branches, and effectively improving the overall uniformity of the return water temperature field;

[0077] Furthermore, to achieve accurate monitoring of the return water temperature field of each branch of the cooling system of the cryogenic BOG reliquefaction unit, a high-density temperature sensor array was established on the return water collection pipe sections of five key cooling branches (first-stage compression stage, second-stage compression stage, motor cooling, main unit casing, and inverter cabinet). At least three Pt100 platinum resistance temperature sensors were arranged at equal intervals on the vertical pipe section of each branch's return water inlet. The installation depth ensured that the sensing elements were completely immersed in the fluid and insulated from the pipe wall. The sensor array synchronously collected temperature data with a 50-millisecond sampling period, transmitted it to the distributed I / O module via shielded twisted-pair cable, and uploaded it to the central control unit via the Modbus RTU protocol, forming a raw observation dataset with time series and spatial location dimensions. The data preprocessing stage used a moving average filter (1-second window width) to eliminate random interference, and sensor calibration compensation was performed according to GB / T17614.1 standard. Finally, a two-dimensional real-time temperature field observation matrix with a time resolution of 1 second and a spatial resolution of 3 measuring points per branch was constructed. Based on... Real-time collected discrete temperature observation data are used to reconstruct the full-field temperature using a distributed algorithm combining Kalman filtering and radial basis function interpolation. A system state-space model is established with the return water temperature of each branch as the state variable. Based on the pipeline thermal inertia parameters, the process noise covariance matrix Q is set to an identity matrix with diagonal elements of 0.01℃². Based on the sensor accuracy calibration, the observation noise covariance matrix R is set to an identity matrix with diagonal elements of 0.0225℃². The Kalman filtering prediction-correction loop is iteratively run with a 100-millisecond cycle to effectively suppress measurement noise and achieve temporal smoothing of the temperature field. On this basis, a spatial interpolation model is constructed using thin-plate spline radial basis functions. The kernel function width parameter is adaptively adjusted to 0.15 meters according to the sensor spacing. The interpolation grid resolution is set to 0.05 meters × 0.05 meters. The reconstruction algorithm outputs a complete two-dimensional temperature distribution cloud map, which can identify areas with uneven temperature distribution exceeding a set threshold and automatically label the spatial coordinates and temperature gradient characteristics of high-temperature areas (>42℃) and low-temperature areas (<38℃).

[0078] Based on the process parameters and equipment technical data of the device, and combined with the principle of heat transfer, the cooling load and cooling water flow rate of each equipment requiring cooling are calculated as shown in Table 2 below (considering a 10% safety margin).

[0079] Table 2. Calculation of Cooling Load and Cooling Water Flow Rate for Each Equipment Requiring Cooling:

[0080] .

[0081] In addition, step 4 also includes: treating each branch as a population, using the return water temperature field uniformity index as the fitness function, optimizing the flow compensation coefficient of each branch through a genetic algorithm to improve the overall optimization efficiency and quickly approach the optimal temperature field distribution state. In each control cycle, the fitness is calculated based on the temperature field reconstruction results, and excellent individuals are selected for crossover and mutation operations to generate a new generation of flow allocation schemes. The excellent individuals are the branch combinations, which enhance the global search capability of the algorithm and avoid getting trapped in local optima. The optimized flow compensation command is sent to the intelligent valves of each branch to quickly converge and maintain the return water temperature field in a steady state, thereby achieving dynamic equilibrium and stable control of the temperature field and improving the thermal management quality of the system.

[0082] Furthermore, at the beginning of each control cycle (set to 10 seconds), the return water temperature field reconstruction algorithm is invoked. Based on real-time data from the high-density sensor array, a two-dimensional temperature distribution cloud map of the current moment is generated. Based on this two-dimensional temperature distribution cloud map, the core optimization index of this control cycle, namely the return water temperature field uniformity index, is calculated. This index is quantified using the root mean square error formula. The specific calculation process is as follows: the temperature field of the reconstruction area is discretized into a grid with a resolution of 0.05 m × 0.05 m, and the sum of the squares of the deviations of the temperature values ​​of all nodes in the grid from the target return water temperature (40℃) set by the system is calculated. The mean value is then taken and the square root is taken. The resulting uniformity index of the return water temperature field will be used as the fitness benchmark value of the current population of the genetic algorithm. At the same time, the genetic algorithm parameters are initialized and the population size is set to 20 individuals. Each individual is composed of a set of real number vector codes containing five elements, which correspond to the flow compensation coefficients of the five cooling branches (primary compression, secondary compression, motor cooling, main unit casing, and inverter cabinet). The initial values ​​of each flow compensation coefficient are randomly generated within the range of [-0.15, +0.15]. This range corresponds to the engineering design constraint that the flow adjustment range of the branch is limited to ±15% of the rated flow.

[0083] The expression for the uniformity index of the return water temperature field is as follows:

[0084] ;

[0085] In the formula: This is the uniformity index of the return water temperature field. The smaller the value, the closer the temperature of all grid points is to the target temperature, and the better the uniformity of the entire temperature field. Ideally, This indicates that the overall temperature is exactly equal to the target temperature. This represents the total number of discrete grid points obtained through the reconstruction algorithm, used to calculate the uniformity index. The first one is calculated using the Kalman filter and radial basis function interpolation reconstruction algorithm. Node temperature values ​​at each grid point; The target return water temperature set for the system;

[0086] Based on the calculated fitness values, a genetic algorithm is used to perform selection, crossover, and mutation operations to evolve the population. The selection process employs a tournament selection method, randomly selecting three individuals from the population each time, retaining the individual with the best fitness value (i.e., the return water temperature field uniformity index) as the parent. This process is repeated until a sufficient number of parent individuals are selected. Then, arithmetic crossover is performed on the selected parent individuals with a preset crossover probability of 0.8 to generate offspring individuals. The offspring individuals generated by crossover are then subjected to Gaussian perturbation of any coefficient in their encoding vector with a mutation probability of 0.05. The standard deviation of the perturbation is set to 0.03 to ensure search diversity and local fine-tuning capability. The evolutionary process continues, with a maximum of 50 generations. After each generation, the fitness of newly generated individuals is re-evaluated. Evolution terminates when the algorithm iterates to the maximum number of generations, or when the fitness value of the optimal individual in the population changes by less than 0.05℃ within 10 consecutive generations. At this point, the individual with the best fitness value in the current population, which represents the flow compensation coefficient vector that ensures the most uniform predicted temperature field, is selected. The optimal flow allocation scheme for this control cycle is determined. After the optimal flow allocation scheme is determined, it is converted into specific execution instructions. The central control unit sends the five flow compensation coefficients in the scheme to the intelligent regulating valve controllers on the corresponding branches through the OPCUA communication protocol. After each controller receives its exclusive flow compensation coefficient k, it calculates and sets the new target position of the valve in real time according to the formula: target opening degree = reference opening degree × (1+k). The reference opening degree is the original instruction determined according to the dynamic cooling demand map. The valve positioner drives the actuator to move to the target opening degree according to the preset speed curve to complete the flow regulation. The entire instruction sending and execution process must be completed within the 10-second control cycle. Then, the next control cycle is entered, and the closed-loop process of monitoring-reconstruction-evaluation-optimization-execution is repeated. Through periodic online rolling optimization, the changes in operating conditions are continuously tracked, and the return water temperature of each branch is driven to converge quickly and remain in a steady state within the allowable fluctuation range of the target temperature of 40℃±0.8℃, so as to achieve long-term, high-precision balanced control of the return water temperature field.

[0087] Step 5 specifically includes: calculating the total waste heat of the system based on the balanced return water temperature field; determining the theoretical heat dissipation requirements by combining the cooling tower performance curve and the current ambient wet-bulb temperature; effectively avoiding energy waste and equipment damage caused by overcooling or overheating of the cooling tower; significantly improving heat dissipation efficiency; adjusting the cooling tower fan speed through the frequency converter to match the actual heat dissipation with the theoretical requirements; monitoring the real-time relationship between fan energy consumption and heat dissipation efficiency; and directly reducing the operating energy consumption of the cooling tower fan by optimizing the fan speed to achieve system operation economy, based on the total system load and return water temperature distribution; dynamically adjusting the valve opening of the bypass pipeline from the motor cooling heat exchanger to the main unit casing to achieve cascade utilization of waste heat and active balance of system heat load, and reducing the starting frequency and operating load of the cooling tower under low load conditions by actively scheduling waste heat, thereby further reducing the overall system energy consumption and improving the comprehensive utilization rate of thermal energy.

[0088] Furthermore, within each control cycle, the central control unit calculates the total residual heat of the system based on the real-time reconstructed equalized return water temperature field. Specifically, this process involves integrating the return water flow and temperature data from all cooling branches and applying heat transfer formulas. Calculations are performed, in which, The density of the cooling water is taken as the design value of 998 kg / m³. 3 (20℃ operating condition) The specific heat capacity of water at constant pressure is taken as 4.18 kJ / (kg·℃). The measured flow rate (m³) of each branch is given. 3 / h), This is the difference between the return water temperature of this branch and the set reference supply water temperature (32℃). The system's total waste heat is determined. Subsequently, the current ambient wet-bulb temperature sensor data is read, and the cooling tower performance curve database is accessed. This performance curve uses wet-bulb temperature, fan speed, and cooling water flow rate as independent variables, and heat dissipation as the dependent variable. A three-dimensional interpolation algorithm is used to determine the current total return water flow rate (approximately 180 m³ / h). 3The theoretical heat dissipation required for the cooling tower to reach theoretical heat dissipation balance (i.e., return water temperature drops to 32℃) at the current wet-bulb temperature is calculated. After obtaining the theoretical heat dissipation, the fan speed is dynamically controlled by adjusting the output frequency of the cooling tower fan inverter. The built-in closed-loop control algorithm takes the deviation between the actual heat dissipation (indirectly reflected by the total residual heat of the system) and the theoretical heat dissipation as input, and uses a proportional-integral controller to output frequency commands. The adjustment range is 30% to 100% of the fan's rated frequency. Increasing the fan speed enhances the heat exchange between the air and cooling water, thereby increasing the heat dissipation; conversely, decreasing the fan speed reduces the heat dissipation. During the actual adjustment process, the real-time operating power of the fan is monitored simultaneously, and its ratio to the heat dissipation is calculated, i.e., the real-time energy efficiency ratio (COP). The control strategy, under the premise of meeting the heat dissipation requirements, appropriately balances the fan speed and energy efficiency ratio. Through a search algorithm, within the feasible speed range that meets the theoretical heat dissipation, the operating point that results in a relatively high real-time energy efficiency ratio is selected to achieve energy-saving operation. Simultaneously, based on the real-time calculation of the total waste heat of the system and the uniformity of the return water temperature field, the bypass pipeline from the outlet of the motor cooling heat exchanger to the cooling circuit of the main unit casing is coordinated and adjusted. An electric regulating valve is installed on this bypass pipeline, and its opening is dynamically set by the controller according to a preset optimization strategy. The core of the strategy is: when the total system load is high and the uniformity of the return water temperature field is good, the opening of the bypass valve is appropriately reduced to allow more high-temperature return water to enter the cooling tower for heat dissipation; when the total waste heat of the system is low and there is a temperature rise requirement in the main unit casing area, the opening of the bypass valve is increased to directly guide the waste heat generated by the motor cooling to the main unit casing for auxiliary heating, reducing the heat dissipation burden of the cooling tower. The change of bypass flow rate is fed back to the total flow and temperature calculation in real time to form a closed loop, and the adjustment rate of the valve opening is limited to within 2% of the stroke per second to avoid hydraulic impact on the main system. Through dynamic bypass adjustment, the cascade utilization of waste heat inside the system and the active balance of the overall heat load are achieved.

[0089] Step 6 specifically includes: Based on physical modeling and data-driven methods, constructing a full-element digital twin model of the cooling system of the cryogenic BOG reliquefaction unit, including the pipeline network, heat exchange equipment, pumps, valves, and control system; establishing a high-precision virtual system to provide a reliable simulation platform for predictive optimization; synchronizing field operation data to the digital twin model in real time via the MQTT protocol to achieve consistency and dynamic mapping between the virtual and actual systems, ensuring real-time consistency between the digital model and the actual system, improving monitoring and diagnostic accuracy; running model predictive control algorithms in the twin environment to simulate system energy efficiency and stability indicators under different cooling strategies; and deploying the optimized control parameters online to the actual control system to form a closed-loop optimization system, realizing pre-verification and online optimization of control strategies, and continuously improving system energy efficiency and stability.

[0090] Furthermore, based on the actual structure and operating parameters of the cooling system of the cryogenic BOG reliquefaction unit, a full-element digital twin model was constructed, including the pipeline network, heat exchange equipment, pumps, valves, and control system. The model, based on system design drawings and actual installation parameters, accurately models the main water supply pipeline (DN250 welded steel pipe), branch pipelines (DN50-DN150 galvanized steel pipes and stainless steel pipes), shell-and-tube heat exchangers after the two-stage compression stages, motor cooling heat exchangers, main unit casing cooling circuits, and inverter cabinet cooling coils, among other physical components. Fluid property parameters were set according to GB50050-2017 standards, with the cooling water density set to 998 kg / m³. 3 The specific heat capacity is taken as 4.18 kJ / (kg·℃). Pump and valve model integrated selection parameters include the main circulating water pump (ISG250-315(I), flow rate 200 m³ / h). 3 The flow-pressure drop characteristic curves of various valves (Z41H-16C gate valve, D371X-10C butterfly valve, etc.) were collected. The control system logic was mirrored and modeled according to the PLC program to form a virtual entity that can simulate the hydraulic, thermal and control response of the system with high precision. Through the MQTT communication client deployed on the production control layer, field sensor data was collected in real time and synchronized to the digital twin platform. The data collection covered key variables such as pressure, temperature, flow rate, valve opening degree and pump frequency. The sampling frequency was 1Hz. After preprocessing, the collected data was published to the corresponding topic of the digital twin model in JSON format via the MQTT protocol. The twin engine subscribed to and parsed the key variable data, driving the corresponding parameters in the digital twin model to update in real time, ensuring that the system state in the digital twin model is highly consistent with the actual system, realizing dynamic mapping and state synchronization between the physical system and the digital model. In a consistent digital twin model, a model predictive control algorithm is embedded for multi-strategy simulation optimization. With the system energy efficiency ratio and operational stability as the comprehensive objective functions, rolling optimization simulations are performed on the control variables of cooling water flow distribution, valve adjustment curves, cooling tower fan speed, and bypass valve opening in a virtual environment. The prediction time domain is set to 10 minutes, and the control time domain is 1 minute. The simulation results output a set of optimized control parameters, including the flow compensation coefficients of each branch, the target frequency of the fan inverter, and the valve action speed correction coefficient. After each control parameter passes safety verification, it is deployed online to the PLC of the actual control system via the OPCUA protocol to update the original control logic or parameter settings. In actual operation, performance feedback data is continuously collected and synchronized to the digital twin model to correct model deviations, forming a continuous closed-loop optimization system of actual operation - data synchronization - simulation optimization - parameter deployment, thereby achieving a gradual improvement in system energy efficiency and stability.

[0091] Example 3, as Figures 1 to 3As shown, based on Examples 1-2, the present invention provides a technical solution: It should be noted that the cooling water system adopts an open circulation process, using municipal tap water or softened water as supplementary water. After being cooled by the cooling tower, it is pressurized by the circulating water pump and delivered to each piece of equipment that needs cooling. After absorbing heat, the water temperature rises, and then it returns to the cooling tower for cooling down, forming a closed loop circulation. The specific process is as follows:

[0092] Makeup water → softening treatment unit → cooling tower water collection tank → circulating water pump → precision filter → main water supply pipeline → branch pipeline → various cooling equipment (first-stage compressor heat exchanger, second-stage compressor heat exchanger, motor cooling heat exchanger, main unit casing, inverter cabinet) → return water pipeline → cooling tower → water collection tank (circulation).

[0093] The cooling water circuit design for each piece of equipment is as follows:

[0094] Two shell-and-tube heat exchangers are installed after the compression stage, using a parallel water supply method. The main water supply line branches into two branches, which are connected to the shell-side inlet of each of the two heat exchangers respectively. The shell-side outlets of the heat exchangers are combined and connected to the main return water line. Each heat exchanger is equipped with valves, pressure gauges, and thermometers at its inlet and outlet for easy individual maintenance and flow regulation. Flexible joints are installed in the inlet and outlet pipelines to reduce the impact of equipment vibration on the pipelines.

[0095] The motor cooling air shell heat exchanger adopts a "series + bypass" water circuit design. The cooling water first enters the shell side of the motor cooling air heat exchanger. After absorbing the heat of the motor cooling air, part of the water flow can bypass to the cooling pipe of the main unit shell (improving energy utilization). The remaining water flow flows into the main return water pipe. Flow regulating valves are set at the inlet and outlet of the heat exchanger to accurately control the cooling water volume and ensure that the motor cooling air outlet temperature is stable below 40℃.

[0096] The main unit casing employs a combined "spray + jacket" cooling system. The casing is equipped with a cooling jacket and spray piping for targeted cooling of high-temperature areas. Cooling water bypasses the motor cooling heat exchanger, passes through the jacket and spray system, and then flows into the return water pipe. The spray piping is equipped with atomizing nozzles to improve heat dissipation efficiency. Differential pressure gauges are installed at the jacket inlet and outlet to monitor for pipe blockage.

[0097] The inverter cabinet adopts an "independent branch" water supply design, with a separate branch line leading from the main water supply line to the cooling coils inside the cabinet. After cooling, the water returns directly to the main return line. This branch line is equipped with a small variable frequency water pump (auxiliary) and a precision filter to ensure stable water pressure and clean water quality, preventing coil blockage. The branch line also features a temperature control valve that automatically adjusts the cooling water volume based on the cabinet temperature (set value 35℃) to achieve precise temperature control.

[0098] The cooling tower selected is a crossflow fiberglass cooling tower, model: GFNL-200, with a water treatment capacity of 200m³. 3 / h, cooling capacity 4000kW, fan power 15kW (frequency control); tower body material is fiberglass (FRP), corrosion resistant and lightweight; the packing uses PVC inclined wave packing, which has high heat dissipation efficiency; equipped with automatic water replenishment device and liquid level gauge to ensure stable liquid level in the collection tank.

[0099] The water softening system uses a fully automatic sodium ion exchanger, model: FLECK-9000, with a processing capacity of 5m³. 3 The water hardness is ≤0.03mmol / L, meeting the requirements for circulating water quality; it is equipped with a brine tank and regeneration system to achieve automatic regeneration and continuous water supply, reducing the risk of scaling.

[0100] In addition, regarding the pipeline layout design: it is necessary to combine the site layout to shorten the pipeline length as much as possible, reduce the number of elbows and valves, and reduce frictional resistance and local resistance; the main supply / return water pipeline adopts a ring layout to ensure uniform water supply to each branch pipeline;

[0101] Pipeline slope settings: water supply pipeline slope ≥ 0.003 (in the direction of water flow), return water pipeline slope ≥ 0.005 (in the direction of water flow) to facilitate air venting and drainage; air vent valve is installed at the highest point of the pipeline, and drain valve is installed at the lowest point;

[0102] Flexible joints (rubber expansion joints) are installed on pipelines near the equipment to reduce the transmission of equipment vibration to the pipelines; the pipeline supports are sliding supports or fixed supports, and the spacing complies with the requirements of GB 50235-2010 "Code for Construction of Industrial Metal Piping Engineering";

[0103] Each equipment branch pipeline is equipped with independent valves, pressure gauges, and thermometers for easy individual adjustment and monitoring; the inverter cabinet cooling branch uses insulated pipelines (rock wool insulation, 50mm thick) to reduce heat loss.

[0104] For pipe selection: the main supply / return water pipeline (DN≥200) uses Q235B welded steel pipe with welded connection; the branch pipeline (DN<200) uses galvanized steel pipe with threaded or flanged connection; the inverter cabinet cooling branch uses 304 stainless steel pipe with flanged connection; all pipeline inner walls are treated with anti-corrosion treatment (epoxy coal tar coating), and the outer walls are treated with anti-rust paint and topcoat.

[0105] For water pump control: The main circulating water pump adopts a "pressure linkage + manual switching" control mode. Based on the pressure sensor signal of the main water supply pipeline (set pressure 0.8MPa), the standby pump is automatically started and stopped. When the water supply pressure is lower than 0.6MPa, the standby pump is started. When the pressure is higher than 1.0MPa, one water pump is shut down. The auxiliary water pump (inverter cabinet) adopts frequency conversion control and automatically adjusts the speed according to the temperature signal inside the cabinet to control the cooling water volume.

[0106] For temperature control: The cooling water volume of the inverter cabinet and motor cooling heat exchanger is automatically adjusted by the temperature control valve to ensure that the temperature inside the inverter cabinet is stable at 35℃±2℃ and the motor cooling air outlet temperature is ≤40℃; when the cooling tower outlet water temperature is higher than 32℃, the cooling tower fan speed is automatically increased to enhance the cooling effect.

[0107] For water quality control: The water softening device automatically monitors the hardness of the influent. When the hardness of the effluent exceeds the standard, the regeneration program is automatically started. The circulating water system is equipped with an online water quality monitor (monitoring hardness, pH value, and turbidity). When the water quality exceeds the standard, an audible and visual alarm is issued, and the drain valve and water supply valve are automatically opened to replace the circulating water.

[0108] For monitoring point setup: pressure gauges, thermometers, and flow meters are installed at key locations such as the main supply / return water pipeline, the inlet and outlet of each equipment, the water pump outlet, and the inlet and outlet of the cooling tower; the system is equipped with a PLC control system and a touch screen to monitor and display various parameters (flow rate, pressure, temperature, water quality) in real time, record operating data (≥1 year), and support fault alarms and historical data queries.

[0109] 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 a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit, characterized in that, Includes the following steps: Step 1: Integrate adaptive fuzzy PID control with graph convolutional neural network to monitor and predict the heat load of each cooling equipment in the cryogenic BOG reliquefaction unit in real time, and generate a dynamic cooling demand map. Step 2: Based on the heat load prediction results, dynamically adjust the cooling water flow distribution strategy of each parallel branch, and execute the on-demand distribution and precise cooling strategy through intelligent valves; Step 3 involves introducing a pressure wave suppression frequency domain analysis model and a reinforcement learning adaptive damping algorithm to predict water hammer vibrations caused by flow regulation in real time, dynamically adjust valve actions and pipeline damping, and smooth pressure fluctuations. Specifically, this includes: A frequency domain analysis model for pressure wave suppression, incorporating pipeline geometric parameters, fluid properties, and valve characteristics, was established to simulate the pressure response spectrum under different flow step changes. Based on real-time flow regulation commands, the frequency and amplitude of water hammer caused by pressure wave suppression are predicted by a frequency domain analysis model for pressure wave suppression, and high-risk periods of pressure fluctuations are identified. By combining reinforcement learning adaptive damping algorithm, the action curve of smart valve is pre-adjusted before the pressure peak is predicted, and the impedance parameters of hydraulic damper are dynamically adjusted at key nodes in the pipeline. Step 3 also includes: Using pressure fluctuation amplitude, vibration acceleration and noise level as evaluation indicators, a damping control reward function is constructed, and a damping adjustment strategy is trained through a deep Q-learning network. Before and after each valve action, data from pipeline pressure and vibration sensors are collected, input into a trained strategy network, and the optimal damper adjustment command is output. The strategy network parameters are updated based on the real-time adjustment effect to achieve adaptive optimization of the damping system for varying working conditions and continuously suppress water hammer vibration and structural noise. Step 4 involves using a group of temperature sensors deployed at the return water inlets of each branch to calculate the return water temperature distribution in real time using a distributed temperature field reconstruction algorithm. This identifies areas of uneven temperature distribution and dynamically adjusts the flow compensation between branches based on a co-evolutionary strategy to achieve uniform return water temperature. Specifically, this includes: A high-density temperature sensor array was deployed at the return water inlet of each branch to collect spatiotemporal distribution data of return water temperature and construct a two-dimensional temperature field observation matrix. A distributed temperature field reconstruction algorithm based on Kalman filtering and radial basis function interpolation is adopted to reconstruct the return water temperature field with high precision, identify the uneven temperature distribution areas and their distribution characteristics, and the uneven temperature distribution areas are local high temperature areas and low temperature areas. By combining a co-evolution strategy, a temperature compensation mechanism is established between adjacent branches. By dynamically adjusting the flow distribution weight, the return water temperature field tends to be more uniformly distributed. Step 4 also includes: Each branch is treated as a population, and the return water temperature field uniformity index is used as the fitness function. The flow compensation coefficient of each branch is optimized by a genetic algorithm. Within each control cycle, fitness is calculated based on the temperature field reconstruction results. Superior individuals are selected for crossover and mutation operations to generate a new generation of flow allocation schemes. Superior individuals are branch combinations. The optimized flow compensation command is sent to the intelligent valves of each branch, which quickly converges and maintains the return water temperature field in a steady state. Step 5: Based on the return water temperature equilibrium results, dynamically adjust the cooling tower fan speed and system bypass ratio to maximize waste heat recovery and stabilize the cooling tower heat load. This specifically includes: The total residual heat of the system is calculated based on the balanced return water temperature field. The theoretical heat dissipation requirement is determined by combining the cooling tower performance curve with the current ambient wet-bulb temperature. The speed of the cooling tower fan is adjusted by a frequency converter to match the actual heat dissipation with the theoretical requirements, while the real-time relationship between fan energy consumption and heat dissipation efficiency is monitored. Based on the total system load and return water temperature distribution, the opening of the bypass pipeline valve from the motor cooling heat exchanger to the main unit casing is dynamically adjusted to achieve the cascade utilization of waste heat and the active balance of the system heat load. Step 6: Establish a digital twin model of the cooling system of the cryogenic BOG reliquefaction unit, evaluate the system's energy efficiency ratio and stability in real time, form a closed-loop control system, and continuously iterate and optimize the cooling strategy.

2. The method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit according to claim 1, characterized in that: Step 1 specifically includes: Temperature sensors and flow meters are installed at the inlet, outlet and key heat exchange surfaces of each equipment requiring cooling in the cryogenic BOG reliquefaction unit to collect real-time heat load-related time-series data and establish an equipment operation status monitoring network. A graph structure is constructed based on the collected heat load related time series data, where nodes represent equipment or measuring points and edges represent heat transfer relationships. The thermal coupling characteristics between equipment are extracted through graph convolutional neural networks, and the short-term heat load change trend is predicted, outputting the heat load prediction value. The prediction results are input into an adaptive fuzzy PID controller, which, combined with the real-time operating conditions of the equipment and environmental parameters, generates a dynamic cooling demand map.

3. The method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit according to claim 2, characterized in that: Step 1 further includes: Based on the heat load prediction value output by the graph convolutional neural network, combined with the equipment heat capacity parameters and heat exchanger efficiency model, the theoretical cooling water demand of each equipment requiring cooling under the current operating conditions is calculated. Based on an adaptive fuzzy rule base, prediction errors are corrected online, PID control parameters are optimized, and heat load fluctuations are quickly tracked and suppressed. The corrected cooling demand is output in the form of a spatial distribution map, namely a dynamic cooling demand map. At the same time, the target flow rate and temperature difference of each branch are marked in the dynamic cooling demand map and associated with the intelligent valve control command queue.

4. The method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit according to claim 1, characterized in that: Step 2 specifically includes: Receive the target flow signals of each branch from the dynamic cooling demand map, and calculate the target adjustment amount of each smart valve by combining the current pipeline pressure and valve opening status. The PLC sends control commands to the intelligent valves of each branch, and the valve opening is adjusted by a segmented easing strategy to achieve a smooth flow transition. During the flow regulation process, the inlet water temperature and temperature difference of each cooling device are monitored in real time. The valve position is finely adjusted through closed-loop feedback to ensure that the cooling effect of each cooling device meets the set temperature control range.

5. The method for parallel cooling of multiple devices in a cryogenic BOG reliquefaction unit according to claim 1, characterized in that: Step 6 specifically includes: Based on physical modeling and data-driven methods, a full-element digital twin model of the cooling system of a cryogenic BOG reliquefaction unit, including the pipeline network, heat exchange equipment, pumps, valves and control system, was constructed. Real-time synchronization of on-site operational data to the digital twin model via the MQTT protocol enables consistency and dynamic mapping between the virtual and actual systems. The model predictive control algorithm is run in a twin environment to simulate the system energy efficiency and stability indicators under different cooling strategies, and the optimized control parameters are deployed online to the actual control system to form a closed-loop optimization system.

Images