A method and system for intelligent control of gas cooler in a CO2 refrigeration system

Abstract

The application relates to the technical field of refrigeration system control, and discloses a CO2 refrigeration system air cooler intelligent regulation and control method and system. The method comprises the following steps: obtaining operation environment parameters and operation state parameters of an air cooler; constructing a thermodynamic state tensor and generating a segmented heat exchange adjacency graph and a pressure sensitivity matrix according to the operation environment parameters and the operation state parameters; calculating the mass flow ratio of a cooling liquid and air and the liquid film coverage, inputting a marginal benefit model to obtain the marginal cooling benefit of the cooling liquid and setting a gain upper limit; determining a wind side pressure drop penalty term in combination with nozzle parameters and the liquid film coverage; generating a candidate air supply rotation speed sequence and a candidate liquid supply flow sequence based on the pressure sensitivity matrix, the gain upper limit, the pressure drop penalty term and the thermodynamic state tensor; performing evaporation reachability checking on the predicted liquid film thickness and the predicted operation pressure corresponding to the sequence; and generating a final control sequence and updating model parameters according to the checking result. The application effectively improves the control stability of the CO2 refrigeration system air cooler.

Description

Technical Field

[0001] This application relates to the field of refrigeration system control technology, and in particular to a method and system for intelligent control of gas coolers in a CO2 refrigeration system. Background Technology

[0002] Transcritical carbon dioxide refrigeration systems, with their environmentally friendly characteristics and high-efficiency heat transfer capabilities, have been widely used in industrial and commercial cold chain logistics, data center thermal management, and other fields. As the core heat dissipation device on the high-pressure side of this system, the air cooler's operating status directly determines the overall energy efficiency and equipment safety. However, under transcritical conditions, the operating pressure of carbon dioxide often exceeds ten MPa, and even small changes in ambient temperature and humidity can cause drastic pressure fluctuations. This makes the air cooler's control far more difficult than that of conventional refrigerants. Currently, air coolers generally use a combination of spray cooling and forced airflow to improve heat exchange capacity, but existing control strategies mainly rely on preset spray flow rate and fan speed ratios, or only on feedback adjustment based on a single temperature measurement point, making it difficult to adapt to real-time changes in ambient humidity. Excessive spraying under dry conditions leads to water waste, while poor liquid film drainage under high humidity conditions causes a thick water layer to accumulate on the tube bundle surface. This not only hinders airflow, increases air-side pressure drop and fan power consumption, but also directly increases system operating pressure due to impaired heat exchange. Once the pressure limit of the equipment is approached, the compressor is forced to reduce its frequency or even shut down urgently, paralyzing the entire refrigeration system.

[0003] The root cause of these problems lies in the strong coupling between spray cooling and air-side heat transfer, while the boundary condition of ambient humidity is constantly changing. Traditional control methods cannot quantify the actual cooling benefit of increasing spray volume under the current humidity conditions, nor can they predict the extent to which liquid film accumulation will trigger a sudden change in air resistance, and it is even more difficult to balance the dynamic equilibrium between heat dissipation demand and pressure safety. When the ambient wet-bulb temperature rises rapidly during high-load periods, blindly increasing the spray flow rate based solely on the outlet temperature exceeding the limit will result in a large number of water droplets failing to evaporate and directly adhering to the heat exchange tube bundle under near-saturated air humidity conditions. This leads to a rapid increase in liquid film thickness and a doubling of air-side resistance. Even with the fan running at full speed, it is difficult to maintain sufficient airflow, ultimately causing the high-pressure side pressure to rapidly rise to the equipment protection threshold. In this situation, the control system lacks an effective assessment of the marginal cooling benefit of the coolant, lacks judgment on the liquid film drainage threshold, cannot intervene in advance before excessive spraying, and has not established a quantitative mapping from the segmented heat transfer characteristics of the air cooler to the operating pressure sensitivity, resulting in a disconnect between control actions and pressure response.

[0004] Therefore, providing a smart control method and system for CO2 refrigeration system air coolers, which can determine the maximum effective spray volume corresponding to the current humidity in real time under environmental temperature and humidity fluctuations and load changes, simultaneously predict the impact of liquid film accumulation on wind-side resistance, and combine the pressure sensitivity of each heat exchange section of the air cooler to generate control commands that take into account heat exchange, safety and energy consumption through rolling optimization and evaporation reachability verification, and avoid ineffective spraying and pressure exceeding limits, is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a method and system for intelligent control of gas coolers in CO2 refrigeration systems, which is used to improve the control stability of gas coolers in CO2 refrigeration systems.

[0006] In a first aspect, this application provides a method for intelligent control of a gas cooler in a CO2 refrigeration system, comprising: S1. Obtain the operating environment parameters, including wet-bulb temperature data and liquid film distribution data, as well as the operating status parameters, including operating pressure, liquid supply flow rate, air supply speed, and internal fluid flow direction, of the CO2 refrigeration system air cooler. S2. Construct a thermodynamic state tensor based on operating environment parameters and operating state parameters, generate a segmented heat transfer adjacency diagram by combining the internal fluid flow direction, calculate the influence weight of the heat transfer section on the operating pressure, and generate a pressure sensitivity matrix. S3. Calculate the mass flow ratio of coolant to air based on the supply flow rate and air speed, calculate the liquid film coverage based on the liquid film distribution data, input the mass flow ratio, wet-bulb temperature data and liquid film coverage into the pre-trained marginal benefit model to obtain the marginal cooling benefit of coolant, and set the upper limit of coolant gain. S4. Obtain the parameters of the liquid supply nozzle, determine the coolant particle size class based on the liquid supply flow rate, and determine the wind-side pressure drop penalty item based on the liquid film coverage, coolant particle size class and air supply speed. S5. Based on the pressure sensitivity matrix, coolant gain upper limit, wind-side pressure drop penalty term and thermodynamic state tensor, generate candidate air supply speed sequence and candidate coolant flow rate sequence. S6. Extract the predicted liquid film thickness corresponding to the candidate liquid supply flow rate sequence and the predicted operating pressure corresponding to the candidate air supply speed sequence, and perform evaporation reachability verification. S7. Generate the final control sequence based on the evaporation reachability verification results, and correct the coolant marginal cooling benefit and pressure sensitivity matrix.

[0007] Secondly, this application provides an intelligent control system for a CO2 refrigeration system air cooler, comprising: The multi-source sensing module is used to acquire operating environment parameters, including wet-bulb temperature data and liquid film distribution data, as well as operating status parameters, including operating pressure, liquid supply flow rate, air supply speed, and internal fluid flow direction, in the air cooler of the CO2 refrigeration system. The thermal modeling module is used to construct a thermal state tensor based on operating environment parameters and operating state parameters, generate a segmented heat transfer adjacency diagram by combining the internal fluid flow direction, calculate the influence weight of the heat transfer section on the operating pressure, and generate a pressure sensitivity matrix. The marginal evaluation module is used to calculate the mass flow ratio of coolant to air based on the supply flow rate and the air speed, calculate the liquid film coverage based on the liquid film distribution data, input the mass flow ratio, wet-bulb temperature data and liquid film coverage into the pre-trained marginal benefit model to obtain the marginal cooling benefit of the coolant, and set the upper limit of coolant gain. The resistance calculation module is used to obtain the parameters of the liquid supply nozzle, determine the coolant particle size class in combination with the liquid supply flow rate, and determine the wind-side pressure drop penalty item based on the liquid film coverage, coolant particle size class and air supply speed. The target optimization module is used to generate candidate air supply speed sequences and candidate coolant flow rate sequences based on the pressure sensitivity matrix, coolant gain upper limit, wind-side pressure drop penalty term and thermodynamic state tensor. The scheme verification module is used to extract the predicted liquid film thickness corresponding to the candidate liquid supply flow rate sequence and the predicted operating pressure corresponding to the candidate air supply speed sequence, and to perform evaporation reachability verification. The decision execution module is used to generate the final control sequence based on the results of the evaporation accessibility verification and to correct the marginal cooling benefit and pressure sensitivity matrix of the coolant.

[0008] Thirdly, this application provides a computer device comprising: a memory and at least one processor, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is running, the processor communicates with the memory via a bus, and when the machine-readable instructions are executed by the processor, the steps of the above-described intelligent control method for a CO2 refrigeration system air cooler are performed.

[0009] Compared with the prior art, the beneficial effects of this application are at least as follows: (1) The marginal cooling benefit of the coolant is evaluated by mass flow ratio, wet bulb temperature difference and liquid film coverage and an upper limit of gain is set. At the same time, the evaporation accessibility is verified by combining the liquid film drainage threshold. This solves the problem that the traditional method cannot determine whether the spray is excessive or ineffective, and avoids water waste and liquid film accumulation under high humidity conditions.

[0010] (2) By introducing liquid film coverage, coolant particle size grade and air supply speed to jointly calculate the wind side pressure drop penalty term and embedding it into the objective function of rolling time domain optimization, the problem of traditional control ignoring the impact of spray on wind side energy consumption is solved, and the surge in fan power consumption caused by excessive spraying is suppressed.

[0011] (3) By introducing a pressure barrier function in rolling optimization, combined with pressure threshold verification and safety degradation control, the problem of lack of active intervention when the pressure approaches the safety boundary in traditional methods is solved, the risk of operating pressure exceeding the limit is reduced, and the safety of equipment operation under extreme conditions is guaranteed. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of an intelligent control method for a CO2 refrigeration system air cooler according to this application; Figure 2 This is a graph showing the decline in marginal cooling benefit of the coolant in this application. Figure 3 This is a comparison chart of the operating pressure trajectory of this application; Figure 4 This is a schematic diagram of the structure of an intelligent control system for a CO2 refrigeration system air cooler according to this application; Figure 5 This is a schematic block diagram of the structure of an intelligent control device for a CO2 refrigeration system air cooler according to this application. Detailed Implementation

[0014] This application provides an intelligent control method and system for a CO2 refrigeration system air cooler. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent control method for a CO2 refrigeration system air cooler in this application includes: Step S1: Obtain the operating environment parameters, including wet-bulb temperature data and liquid film distribution data, and the operating status parameters, including operating pressure, liquid supply flow rate, air supply speed, and internal fluid flow direction, of the CO2 refrigeration system air cooler.

[0016] In one specific embodiment, the process of performing step S1 may specifically include the following steps: An array of environmental sensors is deployed on the air inlet side of the air cooler to collect wet-bulb temperature data of the air cooler, and a liquid film detection device is deployed on the windward side of the air cooler to collect liquid film distribution data on the surface of the air cooler. The wet-bulb temperature data and the liquid film distribution data together constitute the operating environment parameters. A pressure transmitter is installed at the inlet manifold of the air cooler to collect the operating pressure of the air cooler. An electromagnetic flowmeter is installed on the upstream pipeline of the nozzle of the liquid supply device to collect the liquid supply flow rate of the liquid supply device. The real-time feedback signal of the frequency converter of the air supply device is read to obtain the air supply speed of the air supply device. The internal fluid flow direction of the air cooler is read from the equipment configuration file preset by the control system. The operating pressure, liquid supply flow rate, air supply speed and internal fluid flow direction together constitute the operating status parameters.

[0017] Specifically, an environmental sensor array is deployed on the air inlet side of the air cooler. This array consists of a dry-bulb temperature sensor, a wet-bulb temperature sensor, and an atmospheric pressure sensor. Each sensor is connected to the analog input channel of the controller via a shielded cable. The sampling period is set to 1 second. The acquired analog signals are converted into digital signals by an analog-to-digital converter and stored in the controller register. The digital signals are processed by a first-order low-pass filter algorithm with a filtering time constant set to 0.5 seconds to eliminate high-frequency noise caused by airflow pulsation. The wet-bulb temperature data is directly measured by the wet-bulb temperature sensor. The temperature-sensing element of this sensor is wrapped in damp gauze and placed in the air inlet airflow. The evaporation of moisture carries away heat, causing the temperature of the temperature-sensing element to be lower than the dry-bulb temperature, outputting a millivolt-level thermoelectric potential signal. This signal is amplified by a temperature transmitter into a 4 to 20 mA standard current signal before being input to the controller. In the event of sensor malfunction or maintenance, the controller invokes the psychohumidity map conversion model. This model takes dry-bulb temperature and relative humidity data as input, calculates the saturated water vapor partial pressure using the Antoine equation, and then combines this with the percentage of relative humidity to determine the current air humidity content. Finally, it obtains the wet-bulb temperature data through enthalpy-humidity map lookup or polynomial regression. For example, in a hot and humid region during summer, if the dry-bulb temperature is 36 degrees Celsius and the relative humidity is 65%, the corresponding wet-bulb temperature is approximately 29 degrees Celsius. In subsequent processing, the wet-bulb temperature data is subtracted from the CO2 outlet temperature to form the wet-bulb temperature difference, which serves as the input value for the first dimension of the thermodynamic state tensor. Simultaneously, the wet-bulb temperature data is subtracted from the saturated humidity content to form the humidity margin, which serves as the input value for the second dimension of the thermodynamic state tensor. This establishes a quantitative correlation between the environmental thermodynamic state and the air cooler's operating state.

[0018] A liquid film detection device is installed on the windward side of the air cooler. This device employs a capacitive liquid film sensor array or an infrared thermal imager. The capacitive sensor array consists of paired electrodes arranged on the fin surface. When a liquid film covers the electrode area, the change in dielectric constant causes a change in capacitance. The controller detects the capacitance value of each electrode through a capacitance-to-digital conversion circuit. The excitation frequency is set to 10 kHz, and the detection resolution is set to 0.1 picofarads. Electrodes with capacitance values ​​exceeding a threshold are marked as liquid film covered. This threshold is set based on the reference capacitance value of the electrode to air in a dry state plus three times the standard deviation, used to distinguish between the presence of a liquid film and the state of an air gap. The infrared thermal imager detects the temperature distribution on the fin surface, identifying low-temperature areas with temperatures below the dry-bulb temperature and continuous distribution as liquid film covered areas. The raw signal output by the liquid film detection device is processed by the controller into liquid film distribution data, which is stored in the form of a two-dimensional matrix. The rows and columns of the matrix correspond to the spatial grid division of the windward side, and each matrix element records whether there is an effective liquid film in the corresponding grid cell. Statistical calculations are performed on the liquid film distribution data. The number of grid cells marked as effective liquid film is divided by the total number of grid cells to obtain the liquid film coverage. This coverage is then input into the marginal revenue model in subsequent steps as an effective area correction coefficient to correct for evaporation area loss caused by uneven liquid film distribution. For example, if the windward side of the air cooler is divided into 20×30 grid cells (600 cells), and 480 cells are covered, the liquid film coverage is 0.8. Simultaneously, the average liquid film thickness is calculated based on the estimated thickness of the effective liquid film region from the liquid film distribution data. This average liquid film thickness, together with the nominal fin clearance, determines the effective flow area of ​​the fin clearance, which is then used to calculate the wind-side pressure drop penalty term, thereby quantifying the impact of liquid film accumulation on airflow resistance.

[0019] A pressure transmitter is installed at the inlet manifold of the air cooler. The high-pressure side interface of the transmitter is connected to the CO2 pipeline. The internal diaphragm senses the CO2 working fluid pressure and deforms. The deformation is converted into an electrical signal linearly related to the pressure by a capacitive sensing element. The range is set from 0 to 16 MPa. After analog-to-digital conversion, the signal is used to generate operating pressure data. An electromagnetic flowmeter is installed upstream of the nozzle of the liquid supply device. An excitation coil and a pair of detection electrodes are installed in the measuring section of the flowmeter. The excitation current is set to 50 mA. When the coolant flows through the measuring section, it cuts the magnetic lines of force, generating an induced electromotive force. The voltage signal detected by the electrodes is proportional to the volumetric flow rate of the coolant. Based on this voltage signal and the coolant density parameter, the coolant mass flow rate is calculated. This mass flow rate is used as the supply flow rate data in the calculation of the mass flow rate ratio. The real-time feedback signal from the frequency converter of the air supply unit is read via RS485 communication bus using the Modbus RTU protocol. The signal reading period is set to 100 milliseconds. This signal is the current frequency value output by the frequency converter to the motor, which is converted into air supply speed data according to the correspondence between the number of motor pole pairs and the frequency value. The internal fluid flow direction data is stored in the equipment configuration file preset by the control system in the form of a topology description file. This file records the inlet node number, outlet node number, intermediate branch node number, and confluence node number of each tube in the air cooler. During the system initialization phase, this file is read and a topology matrix is ​​constructed. The element in the i-th row and j-th column of the topology matrix represents the connection relationship of the i-th tube at the j-th node, thus completely describing the flow path of CO2 working fluid inside the air cooler. Operating pressure, liquid supply flow rate, air supply speed, and internal fluid flow direction together constitute operating status parameters, while wet-bulb temperature data and liquid film distribution data together constitute operating environment parameters. The operating environment parameters and operating status parameters are encapsulated into a structure object, which uses a timestamp as an index field to ensure that each parameter is strictly aligned in time, forming the data foundation for the subsequent construction of the thermodynamic state tensor and the segmented heat transfer adjacency graph.

[0020] Step S2: Construct a thermodynamic state tensor based on the operating environment parameters and operating state parameters, generate a segmented heat transfer adjacency diagram by combining the internal fluid flow direction, calculate the influence weight of the heat transfer section on the operating pressure, and generate a pressure sensitivity matrix.

[0021] In one specific embodiment, the process of performing step S2 may specifically include the following steps: The wet-bulb temperature difference, moisture content margin, heat exchange section ratio, and pressure deviation are calculated based on the operating environment parameters and operating status parameters. The thermodynamic state tensor is constructed by combining the wet-bulb temperature difference, moisture content margin, heat exchange section ratio, and pressure deviation. The air cooler is divided into multiple heat exchange sections by combining the thermodynamic state tensor and the internal fluid flow direction. A segmented heat exchange adjacency diagram is generated based on the heat transfer relationship between the multiple heat exchange sections. The influence weight of each heat exchange section on the operating pressure is calculated based on the segmented heat exchange adjacency diagram, and a pressure sensitivity matrix is ​​generated.

[0022] Specifically, wet-bulb temperature data and CO2 outlet temperature data are read from the register. The wet-bulb temperature difference is obtained by subtracting the wet-bulb temperature data from the CO2 outlet temperature data. This wet-bulb temperature difference characterizes the degree of convergence between the CO2 working fluid at the air cooler outlet and the ambient wet-bulb temperature. The smaller the value, the closer the evaporative cooling effect is to the theoretical limit. By querying the enthalpy-humidity chart under the current dry-bulb and wet-bulb temperature data, the difference between the current moisture content and the saturated moisture content is calculated to obtain the moisture content margin. This moisture content margin reflects the ability of air to further carry water vapor. Based on the CO2 inlet temperature data, CO2 outlet temperature data, and operating pressure data, combined with the transcritical CO2 working fluid property table, the supercritical section, quasi-critical section, and... are deduced inside the air cooler. The proportion of the tube length occupied by the subcritical liquid phase section is used to obtain the heat exchange section proportion. The difference between the current operating pressure data and the pre-calibrated optimal operating pressure data under this condition is used to obtain the pressure deviation. The optimal operating pressure data is obtained through offline experiments based on the energy efficiency extreme value conditions of CO2 transcritical cycle. The wet-bulb temperature difference is assigned to the first dimension of the tensor, the moisture content margin is assigned to the second dimension of the tensor, the triplet of the heat exchange section proportion is assigned to the third dimension of the tensor, and the pressure deviation is assigned to the fourth dimension of the tensor. These are then spliced ​​to generate the thermodynamic state tensor. This thermodynamic state tensor serves as the initial state input for subsequent rolling time-domain optimization, enabling the optimizer to simultaneously perceive the environmental thermodynamic boundary, the working fluid phase distribution, and the pressure safety margin.

[0023] Based on this, internal fluid flow data is read from the equipment configuration file. This data describes the flow path of the CO2 working fluid in the serpentine coil or microchannel flat tube, including the topological relationships of the inlet node, outlet node, branch point, and confluence point. The air cooler is divided into multiple heat exchange sections along the flow path. The number of sections is determined based on the tube length and the balance of the controller's computing power. For example, for a three-loop parallel air cooler, each loop is evenly divided into 5 sections along the flow direction, for a total of 15 heat exchange sections. The length of each section is controlled between 0.5 meters and 1 meter to ensure that the change in the physical properties of the working fluid within the section does not exceed 5%, satisfying the lumped parameter assumption. Each heat exchange section corresponds to a finite volume unit, and its internal temperature, pressure, and phase parameters are obtained by mapping the data of the corresponding dimension in the thermodynamic state tensor. The heat exchange section is used as a node, and the heat transfer direction and mass transfer direction are used as edges. A segmented heat transfer adjacency graph is constructed. This graph is a directed graph structure, where the edges between nodes represent the heat and mass transfer relationships between adjacent segments. The weight of the edges is determined by the heat transfer coefficients between adjacent segments. The heat transfer coefficients are calculated using the thermal resistance network method based on the fin geometry parameters, the air-side convective heat transfer coefficient, and the CO2-side convective heat transfer coefficient. The air-side convective heat transfer coefficient is calculated using the Churchill-Bernstein correlation formula based on the supply air speed data and the fin geometry parameters. The CO2-side convective heat transfer coefficient is calculated using the Gnielinski formula based on the CO2 property data and the mass flow rate within the segment. This formula is more applicable to the drastic changes in property in the quasi-critical region under transcritical conditions. This segmented heat transfer adjacency graph abstracts the spatial distribution characteristics of the air cooler into a graph topology, enabling the controller to track the segmented attenuation process of heat transfer from the CO2 working fluid to the air.

[0024] A one-dimensional distributed parameter model of the air cooler is established based on a segmented heat transfer adjacency graph. This model treats each heat transfer segment as a lumped parameter node, and each segment is coupled to energy and mass through edges in the adjacency graph. A small perturbation is applied to the model, specifically: the heat transfer coefficient of each heat transfer segment is sequentially increased by an increment of 1% of the baseline value. This increment is based on the principle that in engineering sensitivity analysis, a 1% perturbation is sufficient to elicit a measurable pressure response while maintaining the system within the linear response range, avoiding nonlinear distortion caused by large perturbations. The parameters of the remaining segments are kept unchanged. The change in operating pressure at the air cooler inlet manifold under this perturbation is recorded. The change in operating pressure of the α-th heat transfer segment caused by the perturbation of the heat transfer coefficient of the β-th heat transfer segment is divided by the perturbation amount of the heat transfer coefficient to obtain the influence weight. In this matrix, α and β represent the index numbers of the heat exchange sections. All influence weights are arranged in rows and columns to generate a pressure sensitivity matrix. This matrix is ​​a square matrix with the number of rows and columns equal to the total number of heat exchange sections. The element in the α-th row and β-th column represents the influence coefficient of the heat transfer disturbance of the β-th heat exchange section on the operating pressure of the α-th section. For example, when the heat transfer coefficient of a certain section in the middle of the supercritical section increases by 10% due to the increase of liquid accumulation on the fins, the corresponding pressure sensitivity element reflects the quantitative relationship that the disturbance causes a decrease in operating pressure of about 0.2 MPa. This pressure sensitivity matrix serves as the core of the linearized prediction model in the subsequent rolling time-domain optimization, linearizing the nonlinear relationship between the control variables and the pressure response at the local operating point, enabling the optimizer to complete the deduction from the air supply speed and liquid supply flow rate to the operating pressure trajectory in milliseconds.

[0025] Step S3: Calculate the mass flow rate ratio of coolant to air based on the supply flow rate and air speed, calculate the liquid film coverage based on the liquid film distribution data, input the mass flow rate ratio, wet-bulb temperature data and liquid film coverage into the pre-trained marginal revenue model to obtain the marginal cooling revenue of coolant, and set the upper limit of coolant gain.

[0026] In one specific embodiment, the process of performing step S3 may specifically include the following steps: The supply flow rate is converted into coolant mass flow rate. The air volume is calculated based on the air supply speed and the preset fan characteristic curve and converted into air mass flow rate. The ratio of coolant mass flow rate to air mass flow rate is calculated to obtain the mass flow rate ratio. The proportion of the area covered by the effective liquid film on the surface of the air cooler is statistically analyzed based on the liquid film distribution data to obtain the liquid film coverage. Using the mass flow rate ratio as the saturation factor, the wet-bulb temperature difference corresponding to the wet-bulb temperature data as the driving potential difference, and the liquid film coverage as the effective area correction coefficient, the model is input into a pre-trained marginal revenue model, and the output is the amount of CO2 outlet temperature decrease corresponding to each unit increase in coolant flow rate, thus obtaining the marginal cooling revenue of the coolant. Plot the decay curve of the marginal cooling benefit of the coolant as the coolant flow rate changes, calculate the first derivative of the decay curve, and determine the benefit saturation point when the absolute value of the derivative drops to 10% of the initial value. Set the coolant flow rate corresponding to the benefit saturation point as the upper limit of the coolant gain.

[0027] Specifically, the volumetric flow rate of the supplied liquid output by the electromagnetic flowmeter is read. This volumetric flow rate is the volumetric rate at which the coolant passes through the upstream pipeline of the nozzle, with the unit being cubic meters per hour. The coolant density parameter is then retrieved. This coolant density parameter is obtained by looking up a table based on the coolant temperature or by linear interpolation after actual measurement by a temperature sensor. For example, for normal temperature clean water conditions, 998 kg per cubic meter is used. A multiplication operation is performed to multiply the volumetric flow rate of the supplied liquid by the coolant density parameter to obtain the coolant mass flow rate. This coolant mass flow rate represents the mass of coolant passing through the nozzle per unit time. The air supply speed, fed back from the frequency converter of the air supply device, is read via the communication bus. This air supply speed is the rotational speed corresponding to the current rotational frequency of the fan motor, measured in revolutions per minute (rpm). The fan characteristic curve database in memory is retrieved. This database records the relationship between the fan outlet static pressure and the air supply volume at different air supply speeds in the form of a data table. Using the air supply speed as the query index, a piecewise linear interpolation algorithm is used to match the air supply volume in the database. This air supply volume represents the volume of air passing through the air-cooled unit's windward side per unit time, measured in cubic meters per hour (m³ / h). The air density parameter is retrieved. This air density parameter is calculated using the ideal gas law based on atmospheric pressure and dry-bulb temperature; for example, 1.2 kg / m³ is used for standard operating conditions. The air supply volume is multiplied by the air density parameter to obtain the air mass flow rate, which represents the mass of air participating in heat exchange per unit time. The coolant mass flow rate is divided by the air mass flow rate to obtain the mass flow rate ratio. This mass flow rate ratio is a dimensionless parameter, referred to as the liquid-gas ratio in spray evaporative cooling engineering, and is used to determine whether evaporation is sufficient. The liquid film distribution data output by the liquid film detection device is read. This liquid film distribution data is a two-dimensional matrix. The rows and columns of the matrix correspond to the spatial grid division of the windward side. Each element records whether there is an effective liquid film in the corresponding grid cell. Statistical calculations are performed on the liquid film distribution data. The number of grid cells marked as having an effective liquid film is divided by the total number of grid cells to obtain the liquid film coverage. This liquid film coverage characterizes the ratio of the fin surface area with an effective liquid film to the total fin surface area on the surface of the air cooler. It is used to correct the evaporation area loss caused by uneven liquid film distribution.

[0028] The mass flow rate ratio, wet-bulb temperature difference, and liquid film coverage corresponding to the wet-bulb temperature data are combined into an input vector. This input vector serves as the input to a pre-trained marginal reward model. This marginal reward model employs a feedforward neural network structure, comprising an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer has 3 nodes, receiving normalized mass flow rate ratio, wet-bulb temperature difference, and liquid film coverage, respectively. The normalization coefficients are set based on training set statistics. The first hidden layer has 16 nodes, determined through trial and error based on the input and output dimensions. For example, 16 nodes are sufficient to fit the three dimensions of mass flow rate ratio, wet-bulb temperature difference, and liquid film coverage. The first hidden layer has a weight matrix and a bias vector. The output is processed by a rectified linear unit activation function, which sets negative outputs to zero and retains positive outputs. The second hidden layer has 8 nodes, which are used to compress the feature dimension. This layer also contains a weight matrix and a bias vector. The output is processed by a rectified linear unit activation function. The output layer has 1 node and contains a weight vector and a bias scalar. The activation function is a linear function. The output value is the marginal cooling benefit of the coolant. This marginal cooling benefit represents the decrease in carbon dioxide outlet temperature corresponding to each unit increase in coolant flow rate, with the unit being degrees Celsius per cubic meter per hour. The training process of this marginal revenue model is as follows: A training sample set is constructed from historical operation logs. Each sample contains an input vector and a label value. The label value is calculated using measured data within adjacent control cycles. For example, the changes in liquid supply volumetric flow rate and carbon dioxide outlet temperature are recorded. The label value is obtained by dividing the change in carbon dioxide outlet temperature by the change in liquid supply volumetric flow rate. The loss function uses mean squared error, which is the average of the squared difference between the predicted value and the label value over the batch of samples. The optimizer uses adaptive moment estimation, with a learning rate set to 0.001. This value is determined based on the empirical range of stable convergence of gradients in small-to-medium-scale neural networks. The first-order moment decay coefficient is set to 0.9, and the second-order moment decay coefficient is set to 0.9. 99. This set of parameters is the default configuration for the adaptive moment estimation optimizer, suitable for exponential decay estimation of the first and second moments of the gradient. The batch size is set to 32, which strikes a balance between memory usage and gradient estimation stability. The number of training rounds is set to 100, which is determined through trial and error based on the training set size and the convergence curve of the validation set loss. After training, the model parameters are stored in the controller memory. During online operation, parameter correction is performed every 10 minutes. A forgetting factor of 0.95 is used to weight and fuse historical parameters with new observation data. This forgetting factor is set according to the requirements of engineering control to balance historical statistical stability and recent operating condition response sensitivity, so that the model can adapt to characteristic drift caused by nozzle aging or water quality changes.

[0029] Within the coolant flow range from zero to the maximum rated flow, the marginal cooling benefit of the coolant at each flow point is calculated in steps of 0.05 cubic meters per hour. This step size is set based on the principle of ensuring resolution while controlling computational load within the typical rated flow range of 1.2 cubic meters per hour, forming a discrete sequence. Each element contains a flow point and its corresponding marginal cooling benefit. Numerical differentiation is performed on this discrete sequence, and the slope between two adjacent points is calculated as the first derivative. This first derivative characterizes the rate of change of the marginal cooling benefit of the coolant with the increase of the coolant flow. The absolute value of the initial derivative is extracted, i.e., the derivative when the flow approaches zero. The absolute value is calculated, and then the values ​​are compared point by point along the direction of increasing flow. When the absolute value of the derivative at a certain point drops to 10% of the absolute value of the initial derivative (this percentage is determined based on engineering experience), the input-output ratio of continuing to increase the coolant flow rate is lower than acceptable when the rate of marginal revenue decreases to 10% of the initial rate. The flow rate corresponding to this point is recorded as the revenue saturation point. This revenue saturation point represents the position where continuing to increase the coolant flow rate no longer brings effective cooling benefits. The coolant flow rate corresponding to the revenue saturation point is assigned to the upper limit of coolant gain. This upper limit of coolant gain serves as the upper bound of the hard constraint on the supply flow rate in subsequent rolling time-domain optimization, avoiding unnecessary exploration in directions that waste computing power.

[0030] refer to Figure 2 This figure shows the marginal cooling benefit decay curve of the coolant, illustrating the trend of coolant marginal cooling benefit changing with increasing supply flow rate and its first derivative. The solid blue line represents the marginal benefit curve, and the dashed red line marks the benefit saturation point, corresponding to the point where the derivative drops to 10% of the initial value. This figure visually demonstrates that the proposed marginal benefit model can accurately identify the upper limit of coolant gain, preventing a decrease in the input-output ratio after the supply flow rate exceeds the benefit saturation point.

[0031] Step S4: Obtain the parameters of the liquid supply nozzle, determine the coolant particle size class based on the liquid supply flow rate, and determine the wind-side pressure drop penalty based on the liquid film coverage, coolant particle size class and air supply speed.

[0032] In one specific embodiment, the process of performing step S4 may specifically include the following steps: The parameters of the liquid supply nozzle of the liquid supply device are read from the preset equipment configuration file. The actual pressure difference before and after the nozzle is calculated in combination with the liquid supply flow rate. The Sotter average particle size of the coolant is determined. The Sotter average particle size is divided into three levels: coarse droplets, fine mist and overload, to obtain the coolant particle size level. The equivalent occupied width of the air cooler fin gap is calculated based on the liquid film coverage and the preset average liquid film thickness. The effective flow width of the fin gap is obtained by combining it with the nominal fin gap, and the effective flow area of ​​the fin gap is calculated. The fin friction resistance coefficient of air flowing through the air cooler is calculated based on the effective flow area, the liquid film blockage resistance coefficient is calculated based on the liquid film coverage, and the corresponding droplet momentum exchange resistance coefficient increment is matched based on the coolant particle size grade. The fin friction resistance coefficient, liquid film blockage resistance coefficient and droplet momentum exchange resistance coefficient increment are summed to obtain the comprehensive resistance coefficient. The actual wind-side pressure drop is calculated based on the comprehensive drag coefficient and the air supply speed. The excess of the actual wind-side pressure drop relative to the design pressure drop is converted into a wind-side pressure drop penalty term. When the actual wind-side pressure drop is lower than the design pressure drop, the wind-side pressure drop penalty term is set to zero.

[0033] Specifically, the parameters of the liquid supply nozzle are read from a preset equipment configuration file. These parameters include at least the nozzle orifice diameter, spray angle, arrangement density, and atomization method. This file is provided by the manufacturer at the time of equipment shipment and stored in non-volatile memory in a structured data format. The liquid supply flow rate is measured in real time by an electromagnetic flowmeter in the pipeline upstream of the nozzle. This flow rate characterizes the volume of coolant passing through the nozzle per unit time. The actual pressure difference across the nozzle is calculated as the difference between the outlet pressure of the liquid supply device and the ambient pressure at the nozzle outlet. The outlet pressure of the liquid supply device is collected by a pressure sensor upstream of the nozzle, and the ambient pressure at the nozzle outlet is the atmospheric pressure on the air inlet side of the air cooler. The Soter average particle size is determined by the nonlinear relationship between the actual pressure difference across the nozzle, the nozzle orifice diameter, and the liquid supply flow rate. The specific formula is: Soter average particle size equals a proportionality constant multiplied by 0.5 times the nozzle orifice diameter, multiplied by 0.3 times the ratio of the actual pressure difference across the nozzle to the liquid supply flow rate. The proportionality constant ranges from 1.5 to 2.5 depending on the nozzle type. This Soter average particle size is an indicator of the average diameter of the atomized droplet group, reflecting the degree of droplet breakage. The Sotter average particle size is classified into three levels: coarse droplets, fine mist, and overload. Specifically, when the supply flow rate is less than 60% of the nozzle's rated lower limit, the nozzle outlet pressure difference is insufficient to completely break up the liquid column, resulting in larger droplets with a Sotter average particle size greater than 200 micrometers, classified as coarse droplets. When the supply flow rate is within the rated range, the droplets are refined into a mist, with a Sotter average particle size between 50 and 120 micrometers, classified as fine mist. When the supply flow rate exceeds 110% of the rated upper limit, liquid column penetration occurs, with a Sotter average particle size greater than 200 micrometers and uneven distribution, classified as overload. This coolant particle size classification is based on nozzle atomization characteristic experimental calibration. In the coarse droplet level, droplet momentum loss is large but distribution is uneven; in the fine mist level, droplets flow naturally with the airflow; and in the overload level, liquid column penetration leads to increased liquid film accumulation.

[0034] The liquid film coverage is obtained by statistically analyzing the liquid film distribution data output by the liquid film detection device. This coverage is defined as the ratio of the fin surface area with an effective liquid film to the total fin surface area. The preset average liquid film thickness is set based on the steady-state solution of the liquid film mass balance relationship under current operating conditions, or directly measured by the liquid film detection device. This preset average liquid film thickness characterizes the average thickness of the coolant layer within the liquid film coverage area. The equivalent occupied width of the fin gap is determined jointly by the liquid film coverage and the preset average liquid film thickness. Specifically, it is achieved by multiplying the liquid film coverage by the preset average liquid film thickness and then dividing by the liquid film coverage to obtain the equivalent reduction width of the liquid film in the fin gap. The nominal fin gap is the design spacing of the air cooler fins in a dry state, determined by the equipment manufacturing drawings. The effective flow width of the fin gap is obtained by subtracting the equivalent occupied width from the nominal fin gap. This effective flow width characterizes the actual net clearance that air can penetrate between the fins. The effective flow area between the fins is obtained by multiplying the effective flow width by the product of the fin height and the number of fins. This effective flow area represents the actual flow cross-sectional area that air can penetrate and is used for subsequent drag coefficient calculations.

[0035] The fin friction drag coefficient is determined by the frictional characteristics of airflow through the dry fin. Specifically, it is achieved by using a fin friction drag correction formula based on the ratio of effective flow area to nominal flow area. This formula uses the dry fin friction drag coefficient as a benchmark and multiplies it by the negative square of the effective flow area ratio to obtain the fin friction drag coefficient under the current liquid film state. The dry fin friction drag coefficient is calculated by empirical correlation between fin geometric parameters and air Reynolds number. The air Reynolds number is calculated from the face wind speed converted from the air supply speed, the fin characteristic length, and the air kinematic viscosity. The liquid film blockage drag coefficient is directly determined by the liquid film coverage. Specifically, the liquid film blockage drag coefficient is equal to the liquid film coverage multiplied by the blockage drag benchmark coefficient. This benchmark coefficient is calibrated from experimental resistance data when the liquid film is fully covered, and its value ranges from 0.3 to 0.5. This value is determined based on the statistical results of wind tunnel resistance experiments on typical hydrophilic coated fins under fully wetted conditions. The increment of the droplet momentum exchange resistance coefficient is obtained by matching the coolant particle size levels. The increment of the droplet momentum exchange resistance coefficient corresponding to the coarse droplet level is 15% of the dry fin friction resistance coefficient, the increment corresponding to the fine mist level is 8%, and the increment corresponding to the overload level is more than 30%. This matching relationship is obtained by fitting the gas-liquid two-phase flow resistance measurement data in wind tunnel experiments. Under the coarse droplet level, the large droplet momentum loss leads to significant additional resistance. Under the fine mist level, the additional resistance of droplets moving with the airflow is small. Under the overload level, liquid column penetration triggers violent momentum exchange. The comprehensive resistance coefficient is obtained by summing the fin friction resistance coefficient, the liquid film blockage resistance coefficient, and the increment of the droplet momentum exchange resistance coefficient. This comprehensive resistance coefficient characterizes the total flow resistance encountered by air flowing through the air cooler.

[0036] The actual wind-side pressure drop is determined by the combined drag coefficient, the square of the oncoming wind speed converted from the supply air speed, and the air density. Specifically, the actual wind-side pressure drop equals the combined drag coefficient multiplied by the square of the oncoming wind speed, then multiplied by the air density and divided by 2. The oncoming wind speed is calculated from the supply air speed, the fan characteristic curve, and the air-cooler's frontal area. The design pressure drop is the allowable design value of the wind-side pressure drop for the air cooler under dry-state rated operating conditions, given in the equipment technical specifications. The wind-side pressure drop penalty term is derived from the difference between the actual wind-side pressure drop and the design pressure drop. Specifically, when the actual wind-side pressure drop is lower than the design pressure drop, the wind-side pressure drop penalty term is zero; when the actual wind-side pressure drop exceeds the design pressure drop, the wind-side pressure drop penalty term equals the product of the excess ratio and the design pressure drop, multiplied by a penalty coefficient. This penalty coefficient is set to 1.0, based on the principle that the excess pressure drop should have the same order of magnitude as the power consumption of the supply air device in the objective function, preventing the optimizer from excessively pursuing heat exchange efficiency while ignoring the fan's additional power consumption. The wind-side pressure drop penalty term is included as an additional term in the objective function in the rolling time-domain optimization, thereby suppressing the excessive increase in air supply speed and liquid supply flow rate.

[0037] Step S5: Based on the pressure sensitivity matrix, coolant gain upper limit, wind-side pressure drop penalty term, and thermodynamic state tensor, generate candidate air supply speed sequence and candidate coolant flow rate sequence.

[0038] In one specific embodiment, the process of performing step S5 may specifically include the following steps: Using the thermodynamic state tensor as the current operating state input, the pressure sensitivity matrix as the pressure dynamic response model of the air cooler, the upper limit of the coolant gain is set as the maximum constraint value of the supply flow rate, and the wind-side pressure drop penalty term is set as the energy consumption weight coefficient of the objective function to construct a multi-constraint optimization model. A rolling time-domain predictive optimization algorithm with a pressure barrier function is used to solve the multi-constraint optimization model. Under the premise of satisfying the safety boundary of the air cooler operating pressure, the control scheme with the lowest comprehensive control cost is obtained by iterative calculation. Extract the speed adjustment value and flow rate adjustment value corresponding to the time step from the control scheme to generate candidate air supply speed sequence and candidate liquid supply flow rate sequence.

[0039] Specifically, the thermodynamic state tensor serves as the operating state input. This tensor is composed of wet-bulb temperature difference, moisture margin, heat exchange section ratio, and pressure deviation, and is used to initialize the state variables of the multi-constraint optimization model. The pressure sensitivity matrix serves as the dynamic pressure response model for the air cooler. Each element in this matrix represents the influence coefficient of heat transfer disturbances in each heat exchange section on the operating pressure. During the optimization process, this matrix is ​​used as the linearization prediction kernel, mapping the changes in air supply speed and liquid supply flow rate to the future operating pressure increment. The system energy efficiency ratio model takes the air cooler outlet temperature, operating pressure, and evaporation temperature as inputs and outputs the ratio of cooling capacity to total input power. The pressure fluctuation model takes the air supply volume, spray volume, and environmental disturbances as inputs and outputs the standard deviation of pressure fluctuation. The power consumption model of the air supply device maps the fan speed to electrical power consumption based on the cubic law of fan speed. The power consumption model of the liquid supply device includes the hydraulic power consumption of the pump and the electrical power consumption corresponding to the pressure difference before and after the nozzle. After converting each physical quantity into dimensionless ratios or economic costs, an objective function is constructed. Specifically, this is achieved as follows: the reciprocal of the system energy efficiency ratio is normalized to the rated operating value and multiplied by a weight of 0.5; pressure fluctuation is normalized to the allowable fluctuation range and multiplied by a weight of 0.2; the power consumption of the air supply device is normalized to the rated power and multiplied by a weight of 0.1; the power consumption of the liquid supply device is normalized to the rated power and multiplied by a weight of 0.1; and the wind-side pressure drop penalty is normalized to the design pressure drop and multiplied by a weight of 0.1. The sum of these five terms yields the objective function value. The weight of 0.5 is set because the overall energy efficiency of the transcritical carbon dioxide system is the core of optimization; the weight of 0.2 is set because the high-pressure side pressure stability is directly related to equipment safety; and the remaining weights of 0.1 are set because the power consumption of air supply and liquid supply, as well as wind-side resistance, are used as coupled constraints in the comprehensive trade-off. The upper limit of coolant gain is used as a constraint value for the liquid supply flow rate, applied as a constraint to the upper bound of the liquid supply flow rate decision variable. The pressure barrier function is constructed using limit values, with an upper limit of 12 MPa and a lower limit of 7.5 MPa. The function adopts a logarithmic reciprocal form, and the function value approaches zero when the operating pressure is far from the limit value, and the function value diverges rapidly when the operating pressure approaches the limit value. The function is embedded in the objective function as a constraint term to maintain the differentiability of the objective function to support gradient solving.

[0040] A rolling time-domain predictive optimization algorithm with a pressure barrier function is employed to solve the multi-constraint optimization model. This algorithm starts at the computation time and sets a prediction time domain forward to 60 seconds. This length is determined based on the thermal inertia time constant of the air cooler and the pressure response delay characteristics, covering the dynamic transition process of pressure. Within this time domain, the control variables are discretized into control step sizes, each set to 5 seconds. This step size is determined based on the matching principle between the inverter's response speed and the control cycle, enabling the actuator to track command changes. Using the thermodynamic state tensor as the initial state, the pressure sensitivity matrix is ​​used to extrapolate the state trajectory at future times. The cumulative value of the objective function is calculated within each step. This cumulative value is the weighted sum of the objective function values ​​at each step within the prediction time domain, with step sizes closer to the computation time having a larger weight. The weight decay coefficient is set to 0.9, based on engineering experience that decisions made at nearby times have a greater impact on the state. The pressure barrier function is constructed with an upper pressure limit of 12 MPa and a lower pressure limit of 7.5 MPa as boundaries. The upper limit corresponds to the equipment's pressure safety red line, and the lower limit corresponds to the boundary between transcritical and subcritical states. The two together constitute the operating pressure safety corridor. The function adopts a logarithmic inverse form, mapping the margin between the current predicted operating pressure value and the limit value as a penalty term. The penalty coefficient is set to 1000, which is determined based on the typical order of magnitude of the reciprocal of the system energy efficiency ratio in the objective function. This causes the barrier term to approach zero when the pressure is far from the limit and to diverge rapidly when the pressure approaches the limit. For example, when the predicted operating pressure value is 11.8 MPa, the corresponding upper limit margin is only 0.2 MPa, and the barrier term jumps to about 4094, which is comparable to the energy efficiency ratio term, forcing the solver to back off. The total value of the barrier term is embedded in the objective function as an additional term. It is directly added to the reciprocal of the system energy efficiency ratio, pressure fluctuation, power consumption of the air supply device, power consumption of the liquid supply device, and the wind-side pressure drop penalty term. It maintains a finite value within the pressure safety corridor and increases sharply at the boundary to form a continuously differentiable soft-constraint potential energy wall. This differs from the step abrupt change of traditional hard constraints, ensuring that the quasi-Newton method can calculate a stable Hessian matrix approximation. The barrier terms at each step in the prediction time domain are weighted and accumulated with a weight decay coefficient of 0.9, increasing in weight as they approach the calculation time. The penalty coefficient is adjusted online based on the actual operating pressure. When the actual operating pressure is below 10 MPa for 10 consecutive control cycles, it remains at 1000. When it exceeds 10 MPa, it automatically increases to 2000. This warning interval is set at 2 MPa below the pressure upper limit, making the barrier wall steeper to enhance the ability to suppress exceeding limits. After falling back below 10 MPa and remaining below it for 5 cycles, it returns to 1000, avoiding the accumulation of conservative optimization due to long-term high penalties.The solver employs a sequential quadratic programming algorithm to search for the optimal solution within the range of zero coolant flow rate to the upper limit of coolant gain and minimum air supply speed to rated speed. During the iteration process, the gradient of the objective function with respect to the control variables is calculated, and the search direction is updated using a quasi-Newton method. The maximum number of iterations is set to 50, which is determined based on the balance between the real-time constraints and convergence accuracy of the solver in the industrial control computer. The convergence threshold is set to the difference between the objective function of two adjacent iterations being less than 0.001, which is determined based on the order of magnitude of the objective function and the error range. The pressure barrier function is automatically penalized during iteration. When the predicted operating pressure is in the range of 9 MPa to 10 MPa, the barrier term is close to zero, and the objective function value is dominated by the inverse of the system energy efficiency ratio. The solver continues to search in the direction of reducing energy consumption. When the predicted operating pressure approaches 11.8 MPa, the barrier term jumps to an order of magnitude comparable to the inverse of the system energy efficiency ratio, and the total value of the objective function increases significantly. The solver automatically backtracks to try solutions in a region with lower pressure based on gradient information. This backtracking process is automatically completed by the quasi-Newton method without the need for manual intervention in the constraints, thereby satisfying the safe boundary of the air cooler's operating pressure.

[0041] The rotational speed and flow rate adjustment values ​​corresponding to the time steps are extracted from the obtained control scheme to generate candidate air supply speed sequences and candidate liquid supply flow rate sequences. The candidate air supply speed sequence is a time series whose length equals the number of steps in the prediction time domain. Each element in the sequence corresponds to a target rotational speed at a time step, measured in revolutions per minute (rpm). Similarly, the candidate liquid supply flow rate sequence is also a time series, with each element corresponding to a target flow rate at a time step, measured in cubic meters per hour (m³ / h). For example, for a prediction time domain of 12 steps, the candidate air supply speed sequence increases from 1200 rpm to 1350 rpm, while the corresponding candidate liquid supply flow rate sequence smoothly transitions from 0.6 m³ / h to 0.75 m³ / h. This increase and transition process is automatically generated by the optimizer based on the trade-off between heat dissipation demand and pressure safety under high afternoon load conditions. This output format allows the system to verify the operating trajectory over a future period before issuing control commands, rather than making single-point decisions, thereby ensuring the synchronization of the air supply and liquid supply device's actions.

[0042] refer to Figure 3 This figure is a comparison of operating pressure trajectories, showing the change trajectory of the air cooler's operating pressure in the prediction time domain before and after rolling time-domain optimization with a pressure barrier function. The blue curve in the figure represents the unoptimized naturally decaying pressure trajectory, while the red curve represents the target pressure trajectory solved by the optimizer. The upper and lower black dashed lines mark the upper and lower limits of the pressure threshold, respectively, and the green semi-transparent area represents the pressure safety boundary. This figure visually demonstrates that the optimization algorithm can generate a pressure control sequence that smoothly approaches the target value while meeting pressure safety constraints, thus preventing the operating pressure from touching the safety boundary.

[0043] Step S6: Extract the predicted liquid film thickness corresponding to the candidate liquid supply flow rate sequence and the predicted operating pressure corresponding to the candidate air supply speed sequence, and perform evaporation reachability verification.

[0044] In one specific embodiment, the process of performing step S6 may specifically include the following steps: Determine the prediction time domain and time step corresponding to the rolling time domain prediction optimization, calculate the predicted liquid film thickness for each time step based on the candidate liquid supply flow rate sequence and the candidate air supply speed sequence, and calculate the predicted operating pressure for each time step based on the pressure sensitivity matrix and the thermodynamic state tensor. The predicted liquid film thickness at each time step is compared with the preset drainage threshold, and the predicted operating pressure at each time step is compared with the preset pressure threshold. When the predicted liquid film thickness is less than or equal to the preset drainage threshold and the predicted operating pressure is less than or equal to the preset pressure threshold, the corresponding time step is marked as verified. Otherwise, it is marked as verified as unsuccessful. The marking results are summarized to generate the result of the evaporation reachability verification.

[0045] Specifically, the prediction time domain for the rolling time-domain prediction optimization is set to 60 seconds. This duration is determined based on the thermal inertia time constant and pressure response delay characteristics of the air cooler, covering the complete transition process of liquid film accumulation and pressure fluctuation. The time step is set to 5 seconds, determined based on the response speed and control cycle matching principle of the frequency converters of the air supply and liquid supply devices, enabling the actuator to physically track command changes. Thus, the prediction time domain is divided into 12 time steps, each corresponding to a discrete decision point.

[0046] The predicted liquid film thickness for each time step is calculated based on the candidate liquid supply flow rate sequence. This calculation is achieved through the liquid film mass balance relationship. The liquid film mass balance relationship is expressed as follows: the rate of change of liquid film thickness is proportional to the difference between the spray mass flow rate and the evaporation mass flow rate and the drainage mass flow rate. The spray mass flow rate is obtained by multiplying the flow rate value at the corresponding time step in the candidate liquid supply flow rate sequence by the coolant density. The evaporation mass flow rate is jointly determined by the wet-bulb temperature difference, the moisture content margin, and the air-cooled cooler's frontal area. Specifically, the evaporation mass flow rate is equal to the convective mass transfer coefficient multiplied by the saturated water vapor partial pressure difference driven by the wet-bulb temperature difference, and then multiplied by the air-cooled cooler's frontal area. The convective mass transfer coefficient is determined by the frontal wind speed converted from the air supply speed and the air kinematic viscosity through the Chilton-Colburn analogy. This analogy relates the dimensionless numbers of heat transfer and mass transfer, allowing the mass transfer coefficient to be derived from the known convective heat transfer coefficient. The drainage mass flow rate is determined by the gravity of the liquid film and the geometric tilt angle of the fins. Specifically, the drainage mass flow rate is equal to the cube of the liquid film thickness multiplied by the product of the fin width and the sine of the fin tilt angle, multiplied by the square of the coolant density and the acceleration due to gravity, and divided by the product of the coolant dynamic viscosity and the fin length. This formula is based on the assumption of gravity-driven flow of the liquid film on an inclined wall. At each time step, the spray mass flow rate is subtracted from the evaporation mass flow rate and the drainage mass flow rate to obtain the net cumulative mass. This is then divided by the area of ​​the liquid film coverage region and the coolant density to obtain the liquid film thickness increment. This increment is added to the liquid film thickness of the previous time step to obtain the predicted liquid film thickness for the current time step. The initial liquid film thickness is given by the measured value from the liquid film detection device. For example, for a candidate liquid supply flow rate sequence with a peak value of 1 cubic meter per hour, combined with environmental conditions of a wet-bulb temperature difference of 5 degrees Celsius and a moisture content margin of 7 grams per kilogram of dry air, the predicted liquid film thickness under steady-state conditions is approximately 0.22 millimeters.

[0047] The predicted operating pressure for each time step is calculated based on the pressure sensitivity matrix and the thermodynamic state tensor, using the pressure deviation in the thermodynamic state tensor as the initial value. Each element in the pressure sensitivity matrix represents the influence coefficient of heat transfer disturbance in each heat exchange section on the operating pressure. Within each time step, the change in air speed in the candidate air supply speed sequence corresponding to the time step is multiplied by the influence coefficient of air supply speed on operating pressure in the pressure sensitivity matrix to obtain the pressure increment for that step. The influence coefficient of air supply speed on operating pressure is obtained by a linear combination of the air supply speed-related column vector and the influence weight of the heat exchange section in the pressure sensitivity matrix. This pressure increment is added to the predicted operating pressure of the previous time step to obtain the predicted operating pressure for the current time step. This process is repeated chronologically to form the predicted operating pressure trajectory. For example, if the initial operating pressure is 9.2 MPa, and the candidate air supply speed is gradually increased from 1200 rpm to 1400 rpm, the predicted operating pressure trajectory gradually decreases from 9.2 MPa to 8.8 MPa.

[0048] The predicted liquid film thickness at each time step is compared with a preset drainage threshold. The preset drainage threshold is set to 0.25 mm, obtained by measuring the water contact angle of a typical fin spacing and hydrophilic coating. When the liquid film thickness exceeds this value, the fin surface can no longer maintain a stable liquid film, resulting in significant dripping and water flow. The predicted operating pressure at each time step is compared with a preset pressure threshold. The preset pressure threshold includes an upper and lower limit. The upper limit is set to 11.5 MPa, a relatively hard constraint of 12 MPa with a 0.5 MPa safety margin. The lower limit is set to 8 MPa, corresponding to the lower boundary of the transition between transcritical and subcritical states. Within each time step, if the predicted liquid film thickness is less than or equal to the preset drainage threshold and the predicted operating pressure is greater than or equal to the lower limit of the preset pressure threshold but less than or equal to the upper limit of the preset pressure threshold, the time step is marked as a successful verification. If the predicted liquid film thickness is greater than the preset drainage threshold, or the predicted operating pressure is less than the lower limit of the preset pressure threshold, or the predicted operating pressure is greater than the upper limit of the preset pressure threshold, the time step is marked as a failed verification. This marking logic jointly judges the feasibility of liquid film drainage and the pressure safety boundary, avoiding the one-sidedness of satisfying one condition while failing another. The marking results of all time steps are summarized to generate the evaporation reachability verification result. This result is stored in the form of a Boolean sequence, the sequence length of which is equal to the number of steps in the prediction time domain, and each element corresponds to the pass or fail state of a time step.

[0049] Step S7: Generate the final control sequence based on the evaporation accessibility verification results, and correct the coolant marginal cooling benefit and pressure sensitivity matrix.

[0050] In one specific embodiment, the process of performing step S7 may specifically include the following steps: Based on the results of the evaporation reachability verification, for time steps marked as verified, the corresponding adjustment values ​​in the candidate air supply speed sequence and candidate liquid supply flow sequence are retained. For time steps marked as verified as not verified, the adjustment values ​​that were verified and valid in the previous cycle are used. The speed adjustment values ​​and flow adjustment values ​​of all time steps are combined in chronological order to generate the final control sequence. According to the final control sequence, the speed adjustment command with the corresponding time step is issued to the air supply device, and the flow rate adjustment command with the corresponding time step is issued to the liquid supply device; Obtain the actual operation feedback data after executing the final control sequence, and correct the coolant marginal cooling benefit and pressure sensitivity matrix based on the actual operation feedback data.

[0051] Specifically, the evaporation reachability verification results are stored in random access memory as a Boolean sequence. The length of this sequence is consistent with the number of time steps in the prediction time domain, and each element corresponds to a pass or fail status for a time step. Based on this result, for time steps marked as passed, the speed adjustment value and flow adjustment value at the corresponding index position are directly extracted from the candidate air supply speed sequence and candidate liquid supply flow sequence. This extraction operation is implemented through array index addressing, with the index number corresponding one-to-one with the time step number, ensuring that the data mapping is not misaligned. For time steps marked as failed, the verified and valid adjustment values ​​saved in the previous control cycle are retrieved from non-volatile memory. These adjustment values ​​are the speed adjustment value and flow adjustment value that passed the evaporation reachability verification after execution in the corresponding time step of the previous cycle. The storage period is consistent with the prediction time domain and is set to 60 seconds. This period is determined based on the thermal inertia time constant of the air cooler to ensure that the rollback data has physical similarity to the current operating condition and avoids control mismatch caused by using outdated data. The speed adjustment values ​​processed at each time step are concatenated in chronological order to form the final air supply speed sequence, and the flow adjustment values ​​processed at each time step are concatenated in chronological order to form the final liquid supply flow sequence. The final air supply speed sequence and the final liquid supply flow sequence are combined to form the final control sequence. This combination process is implemented through a structure array. Each array element contains three fields: timestamp, speed adjustment value, and flow adjustment value. The timestamp is incremented at 5-second intervals based on the start of the control cycle to ensure that the instruction timing is strictly aligned with the prediction time domain, forming a control instruction set with a complete time dimension.

[0052] According to the final control sequence, a speed adjustment command with a corresponding time step is sent to the air supply device via the communication bus, and a flow adjustment command with a corresponding time step is sent to the liquid supply device. The speed adjustment command is in revolutions per minute and is sent to the air supply device inverter via the serial communication port using the Modbus RTU protocol. The communication baud rate is set to 9600 bits per second, with 8 data bits, 1 stop bit, and even parity. These parameters are determined based on the electromagnetic compatibility environment and communication stability standards of the industrial site. After receiving the command, the inverter converts the speed adjustment value into a motor frequency command. The conversion relationship is determined by the number of pole pairs of the motor and the rated frequency. For example, a 4-pole motor corresponds to 1500 revolutions per minute at a rated frequency of 50 Hz, so 1200 revolutions per minute corresponds to an output frequency of 40 Hz. The converted frequency command is used by the inverter's internal vector control algorithm to adjust the duty cycle of the inverter's switching transistors, driving the fan motor to reach the target speed. The flow rate adjustment command, in cubic meters per hour, is sent to the frequency converter of the liquid supply device via the same communication bus. The frequency converter adjusts the pump motor frequency to change the liquid supply flow rate. The conversion relationship between frequency and flow rate is determined by the pump's characteristic curve. At the rated frequency, the pump outputs the rated flow rate. Below the rated frequency, the flow rate decreases approximately linearly. For example, a rated flow rate of 1.2 cubic meters per hour corresponds to 50 Hz, so 0.6 cubic meters per hour corresponds to a 25 Hz output. The duty cycle signal is sent to the solenoid valve or pulse jet device in pulse width modulation form through the digital output port. The duty cycle is obtained by multiplying the ratio of the flow rate adjustment value to the nozzle's rated flow rate by 100%. It is used to maintain spray quality under low flow rate requirements while avoiding prolonged valve slack. The pulse period is set to 6 seconds, which is determined based on the balance between nozzle atomization stability and valve mechanical life. The duty cycle and pulse period together determine the actual opening time of the nozzle per unit time.

[0053] The system acquires actual operational feedback data after executing the final control sequence. This data includes actual pressure changes, equipment outlet temperature changes, and actual equipment power consumption. Actual pressure changes are periodically collected by a pressure transmitter at the air cooler inlet manifold. The sampling frequency is set to twice per second, determined based on the main frequency components of pressure fluctuations, enabling the capture of pressure dynamics within 10 Hz. The actual pressure change is obtained by subtracting the sampled values ​​at the start and end of the control cycle; this difference reflects the net impact of the control action on the high-pressure side. Equipment outlet temperature changes are collected by a temperature sensor at the air cooler outlet. This sensor uses a platinum resistance thermometer or a thermocouple, and its sampling period is synchronized with the pressure transmitter. The equipment outlet temperature change is obtained by subtracting the temperature sampled values ​​before and after the control cycle; this difference characterizes the net contribution of the control action to the heat exchange effect. Actual equipment power consumption is reported by the electrical parameter acquisition modules of the air supply unit's frequency converter and the liquid supply unit's frequency converter. This module detects the three-phase voltage and current on the input side of the frequency converter, calculates the active power through instantaneous power, and obtains the actual equipment power consumption change by subtracting the power values ​​before and after the control cycle. The aforementioned feedback data is transmitted back to the random access memory via the communication bus. The timestamp is used as an index to establish a correspondence with the instruction values ​​in the final control sequence, forming the data basis for the control effect evaluation. This enables subsequent model corrections to trace the causal chain between specific instruction actions and physical responses.

[0054] The marginal cooling benefit of the coolant is adjusted based on the actual pressure change and the actual equipment power consumption. The actual changes in equipment outlet temperature and power consumption caused by changes in coolant flow rate during the current cycle are recorded and compared with the corresponding values ​​previously predicted by the pre-trained marginal benefit model to obtain the model prediction bias. This bias is then weighted by a forgetting factor and injected into the update of the marginal benefit model's benefit coefficients. The forgetting factor is set to 0.97, a value determined based on the requirement in engineering control to balance historical statistical stability with sensitivity to recent operating conditions. This means the model is most sensitive to the most recent operating condition while retaining statistical stability over a longer window, avoiding model oscillation caused by completely discarding historical experience. Specifically, if the model predicts that increasing the spray volume by 0.1 cubic meters per hour will result in a 0.8-degree Celsius drop in outlet temperature, but only a 0.5-degree Celsius drop is actually observed, the corresponding benefit coefficient for this operating condition is reduced. The reduction is determined by multiplying the bias by the learning rate of 0.01. This learning rate is determined through trial and error based on the model's convergence speed and stability, ranging from 0.005 to 0.02, with 0.01 being a compromise value. In similar environments in the future, the upper limit of coolant gain will be correspondingly tightened to avoid continued reliance on degraded model predictions. The update cycle is set to be executed every 10 minutes. This cycle is determined based on the timescale of air cooler operating condition drift to avoid frequent oscillations caused by short-term noise. Each update only corrects the local benefit coefficient corresponding to the current wet-bulb temperature difference and mass flow rate ratio range, while the coefficients for other ranges remain unchanged.

[0055] The pressure sensitivity matrix is ​​corrected based on the change in equipment outlet temperature. A recursive identification method is used to record the change in air supply speed and the corresponding actual pressure change. The difference between the actual pressure change and the pressure change predicted by the pressure sensitivity matrix is ​​used to obtain the matrix element deviation. If the air supply speed increases from 1200 rpm to 1300 rpm, the model predicts a pressure decrease of 0.1 MPa, but the actual decrease is only 0.06 MPa. In this case, the influence weights in the corresponding rows and columns are reduced. The reduction is determined by multiplying the deviation by a correction step size of 0.01. This step size is determined based on the balance between matrix convergence stability and tracking speed to avoid excessively large single corrections that could lead to matrix oscillations or singularities. Conversely, if the actual decrease exceeds the prediction, the corresponding weights are increased. The pressure sensitivity matrix is ​​updated every 10 minutes. This frequency is lower than the correction frequency of the marginal benefit model. It is determined based on the long-term stability and short-term noise characteristics of the pressure response. Each update only corrects some elements related to the current heat exchange section ratio and air supply speed range. Inactive elements retain their original values ​​to reduce the amount of calculation and ensure the sparsity of the matrix. This allows the corrected matrix to reflect the heat transfer characteristic drift caused by equipment aging, fin dust accumulation, or load step changes.

[0056] It is understood that the executing entity of this application can be an intelligent control system for a CO2 refrigeration system air cooler, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0057] The above describes a method for intelligent control of a CO2 refrigeration system air cooler in an embodiment of this application. The following describes an intelligent control system for a CO2 refrigeration system air cooler in an embodiment of this application. Please refer to [link / reference]. Figure 4 One embodiment of the intelligent control system for a CO2 refrigeration system air cooler in this application includes: The multi-source sensing module is used to acquire operating environment parameters, including wet-bulb temperature data and liquid film distribution data, as well as operating status parameters, including operating pressure, liquid supply flow rate, air supply speed, and internal fluid flow direction, in the air cooler of the CO2 refrigeration system. The thermal modeling module is used to construct a thermal state tensor based on operating environment parameters and operating state parameters, generate a segmented heat transfer adjacency diagram by combining the internal fluid flow direction, calculate the influence weight of the heat transfer section on the operating pressure, and generate a pressure sensitivity matrix. The marginal evaluation module is used to calculate the mass flow ratio of coolant to air based on the supply flow rate and the air speed, calculate the liquid film coverage based on the liquid film distribution data, input the mass flow ratio, wet-bulb temperature data and liquid film coverage into the pre-trained marginal benefit model to obtain the marginal cooling benefit of the coolant, and set the upper limit of coolant gain. The resistance calculation module is used to obtain the parameters of the liquid supply nozzle, determine the coolant particle size class in combination with the liquid supply flow rate, and determine the wind-side pressure drop penalty item based on the liquid film coverage, coolant particle size class and air supply speed. The target optimization module is used to generate candidate air supply speed sequences and candidate coolant flow rate sequences based on the pressure sensitivity matrix, coolant gain upper limit, wind-side pressure drop penalty term and thermodynamic state tensor. The scheme verification module is used to extract the predicted liquid film thickness corresponding to the candidate liquid supply flow rate sequence and the predicted operating pressure corresponding to the candidate air supply speed sequence, and to perform evaporation reachability verification. The decision execution module is used to generate the final control sequence based on the results of the evaporation accessibility verification and to correct the marginal cooling benefit and pressure sensitivity matrix of the coolant.

[0058] Through the collaborative efforts of the aforementioned components, the multi-source sensing module collects and transmits multi-dimensional parameters such as wet-bulb temperature, liquid film distribution, operating pressure, liquid supply flow rate, and air supply speed to the thermal modeling module and the edge evaluation module. This ensures that subsequent decisions are always based on complete field data, avoiding the information gaps caused by traditional solutions relying on only a single temperature measurement point. The thermal modeling module divides the air cooler into multiple heat exchange sections along the fluid flow direction and establishes a segmented heat exchange adjacency diagram. Combined with the quantitative output of the pressure sensitivity matrix, the target optimization module can predict the direction and magnitude of the impact of heat exchange disturbances in different sections on the overall operating pressure, thus possessing the ability to perceive pressure response in advance before control commands are generated. The marginal evaluation module and the resistance calculation module quantify the spraying behavior from two dimensions: cooling benefit and wind-side energy consumption. One module provides the actual cooling effect and upper limit of gain under the current humidity conditions, while the other provides the penalty for wind-side pressure drop caused by liquid film accumulation and atomized particle size. Both pieces of information are simultaneously fed into the target optimization module for rolling time-domain optimization, ensuring that the final output air supply speed sequence and liquid supply flow sequence achieve a balance between heat exchange demand, pressure safety, and fan energy consumption. Before the candidate sequence is issued, the scheme verification module performs evaporation accessibility verification using liquid film drainage thresholds and pressure thresholds, intercepting physically infeasible spraying schemes in advance. This avoids liquid film accumulation and airflow blockage caused by excessive spraying under high humidity conditions, and also prevents the erroneous decision to continue increasing spraying when the pressure approaches the safety limit. After executing control commands, the decision execution module continuously collects actual pressure changes, outlet temperature changes, and equipment power consumption. This feedback data is used to adjust the benefit coefficients in the marginal benefit model and the influence weights in the pressure sensitivity matrix online. This allows the entire control logic to gradually adjust its parameters in response to actual disturbances such as equipment aging, nozzle scaling, sudden changes in ambient humidity, and load jumps, ensuring consistent decision-making quality over long-term operation. The modules form a data closed loop through intermediate variables such as the thermodynamic state tensor, pressure sensitivity matrix, and wind-side pressure drop penalty term. Spray control is no longer an open-loop execution of preset ratios, but a closed-loop optimization process integrating state perception, benefit assessment, energy consumption quantification, pressure prediction, feasibility verification, and online learning. Overall, this improves the robustness of CO2 refrigeration system air coolers under complex operating conditions, operational safety, and water and electricity utilization efficiency.

[0059] above Figure 4 The intelligent control system for a CO2 refrigeration system air cooler in this application embodiment is described in detail from the perspective of modular functional entities. The intelligent control device for a CO2 refrigeration system air cooler in this application embodiment is described in detail from the perspective of hardware processing.

[0060] Figure 5This is a schematic diagram of the structure of an intelligent control device for a CO2 refrigeration system air cooler provided in an embodiment of this application. The intelligent control device 300 for a CO2 refrigeration system air cooler can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the intelligent control device 300 for a CO2 refrigeration system air cooler. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute a series of instruction operations in the storage media 330 on the intelligent control device 300 for a CO2 refrigeration system air cooler to implement the steps of the aforementioned intelligent control method for a CO2 refrigeration system air cooler.

[0061] A CO2 refrigeration system air cooler intelligent control device 300 may further include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 5 The illustrated structure of an intelligent control device for a CO2 refrigeration system air cooler does not constitute a limitation on the intelligent control device for a CO2 refrigeration system air cooler provided in this application. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0062] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent control of a gas cooler in a CO2 refrigeration system, characterized in that, include: S1. Obtain the operating environment parameters, including wet-bulb temperature data and liquid film distribution data, as well as the operating status parameters, including operating pressure, liquid supply flow rate, air supply speed, and internal fluid flow direction, of the CO2 refrigeration system air cooler. S2. Construct a thermodynamic state tensor based on the operating environment parameters and the operating state parameters, generate a segmented heat transfer adjacency diagram by combining the internal fluid flow direction, calculate the influence weight of the heat transfer section on the operating pressure, and generate a pressure sensitivity matrix. S3. Calculate the mass flow rate ratio of coolant to air based on the supply flow rate and the air supply speed, calculate the liquid film coverage based on the liquid film distribution data, input the mass flow rate ratio, the wet-bulb temperature data and the liquid film coverage into the pre-trained marginal benefit model to obtain the marginal cooling benefit of coolant, and set the upper limit of coolant gain. S4. Obtain the parameters of the liquid supply nozzle, determine the coolant particle size class based on the liquid supply flow rate, and determine the wind-side pressure drop penalty item based on the liquid film coverage, the coolant particle size class and the air supply speed. S5. Based on the pressure sensitivity matrix, the upper limit of coolant gain, the wind-side pressure drop penalty term, and the thermodynamic state tensor, generate candidate air supply speed sequences and candidate coolant flow rate sequences. S6. Extract the predicted liquid film thickness corresponding to the candidate liquid supply flow rate sequence and the predicted operating pressure corresponding to the candidate air supply speed sequence, and perform evaporation reachability verification. S7. Generate the final control sequence based on the evaporation reachability verification results, and correct the coolant marginal cooling benefit and pressure sensitivity matrix.

2. The method according to claim 1, characterized in that, S1 includes: An array of environmental sensors is deployed on the air inlet side of the air cooler to collect wet-bulb temperature data of the air cooler, and a liquid film detection device is deployed on the windward side of the air cooler to collect liquid film distribution data on the surface of the air cooler. The wet-bulb temperature data and the liquid film distribution data together constitute the operating environment parameters. A pressure transmitter is installed at the inlet manifold of the air cooler to collect the operating pressure of the air cooler. An electromagnetic flow meter is installed on the upstream pipeline of the nozzle of the liquid supply device to collect the liquid supply flow rate of the liquid supply device. The real-time feedback signal of the frequency converter of the air supply device is read to obtain the air supply speed of the air supply device. The internal fluid flow direction of the air cooler is read from the equipment configuration file preset by the control system. The operating pressure, liquid supply flow rate, air supply speed and internal fluid flow direction together constitute the operating status parameters.

3. The method according to claim 1, characterized in that, S2 includes: The wet-bulb temperature difference, moisture content margin, heat exchange section ratio, and pressure deviation are calculated based on the operating environment parameters and operating status parameters. The thermodynamic state tensor is constructed by combining the wet-bulb temperature difference, moisture content margin, heat exchange section ratio, and pressure deviation. The air cooler is divided into multiple heat exchange sections by combining the thermodynamic state tensor and the internal fluid flow direction. A segmented heat exchange adjacency diagram is generated based on the heat transfer relationship between the multiple heat exchange sections. The influence weight of each heat exchange section on the operating pressure is calculated based on the segmented heat exchange adjacency diagram, and a pressure sensitivity matrix is ​​generated.

4. The method according to claim 1, characterized in that, S3 includes: The supply flow rate is converted into coolant mass flow rate. The air volume is calculated based on the air supply speed and the preset fan characteristic curve and converted into air mass flow rate. The ratio of coolant mass flow rate to air mass flow rate is calculated to obtain the mass flow rate ratio. The proportion of the area covered by the effective liquid film on the surface of the air cooler is statistically analyzed based on the liquid film distribution data to obtain the liquid film coverage. Using the mass flow rate ratio as the saturation factor, the wet-bulb temperature difference corresponding to the wet-bulb temperature data as the driving potential difference, and the liquid film coverage as the effective area correction coefficient, the model is input into a pre-trained marginal revenue model, and the output is the amount of CO2 outlet temperature decrease corresponding to each unit increase in coolant flow rate, thus obtaining the marginal cooling revenue of the coolant. Plot the decay curve of the marginal cooling benefit of the coolant as the coolant flow rate changes, calculate the first derivative of the decay curve, and determine the benefit saturation point when the absolute value of the derivative drops to 10% of the initial value. Set the coolant flow rate corresponding to the benefit saturation point as the upper limit of the coolant gain.

5. The method according to claim 1, characterized in that, S4 includes: The parameters of the liquid supply nozzle of the liquid supply device are read from the preset equipment configuration file. The actual pressure difference before and after the nozzle is calculated in combination with the liquid supply flow rate. The Sotter average particle size of the coolant is determined. The Sotter average particle size is divided into three levels: coarse droplets, fine mist and overload, to obtain the coolant particle size level. The equivalent occupied width of the air cooler fin gap is calculated based on the liquid film coverage and the preset average liquid film thickness. The effective flow width of the fin gap is obtained by combining it with the nominal fin gap, and the effective flow area of ​​the fin gap is calculated. The fin friction resistance coefficient of air flowing through the air cooler is calculated based on the effective flow area, the liquid film blockage resistance coefficient is calculated based on the liquid film coverage, and the corresponding droplet momentum exchange resistance coefficient increment is matched based on the coolant particle size grade. The fin friction resistance coefficient, liquid film blockage resistance coefficient and droplet momentum exchange resistance coefficient increment are summed to obtain the comprehensive resistance coefficient. The actual wind-side pressure drop is calculated based on the comprehensive drag coefficient and the air supply speed. The excess of the actual wind-side pressure drop relative to the design pressure drop is converted into a wind-side pressure drop penalty term. When the actual wind-side pressure drop is lower than the design pressure drop, the wind-side pressure drop penalty term is set to zero.

6. The method according to claim 1, characterized in that, S5 includes: Using the thermodynamic state tensor as the current operating state input, the pressure sensitivity matrix as the pressure dynamic response model of the air cooler, the upper limit of the coolant gain is set as the maximum constraint value of the supply flow rate, and the wind-side pressure drop penalty term is set as the energy consumption weight coefficient of the objective function to construct a multi-constraint optimization model. A rolling time-domain predictive optimization algorithm with a pressure barrier function is used to solve the multi-constraint optimization model. Under the premise of satisfying the safety boundary of the air cooler operating pressure, the control scheme with the lowest comprehensive control cost is obtained by iterative calculation. Extract the speed adjustment value and flow rate adjustment value corresponding to the time step from the control scheme to generate candidate air supply speed sequence and candidate liquid supply flow rate sequence.

7. The method according to claim 1, characterized in that, S6 includes: Determine the prediction time domain and time step corresponding to the rolling time domain prediction optimization, calculate the predicted liquid film thickness for each time step based on the candidate liquid supply flow rate sequence and the candidate air supply speed sequence, and calculate the predicted operating pressure for each time step based on the pressure sensitivity matrix and the thermodynamic state tensor. The predicted liquid film thickness at each time step is compared with the preset drainage threshold, and the predicted operating pressure at each time step is compared with the preset pressure threshold. When the predicted liquid film thickness is less than or equal to the preset drainage threshold and the predicted operating pressure is less than or equal to the preset pressure threshold, the corresponding time step is marked as verified. Otherwise, it is marked as verified as unsuccessful. The marking results are summarized to generate the result of the evaporation reachability verification.

8. The method according to claim 1, characterized in that, The S7 includes: Based on the results of the evaporation reachability verification, for time steps marked as verified, the corresponding adjustment values ​​in the candidate air supply speed sequence and candidate liquid supply flow sequence are retained. For time steps marked as verified as not verified, the adjustment values ​​that were verified and valid in the previous cycle are used. The speed adjustment values ​​and flow adjustment values ​​of all time steps are combined in chronological order to generate the final control sequence. According to the final control sequence, the speed adjustment command corresponding to the time step is issued to the air supply device, and the flow rate adjustment command corresponding to the time step is issued to the liquid supply device. Obtain the actual operation feedback data after executing the final control sequence, and correct the coolant marginal cooling benefit and pressure sensitivity matrix based on the actual operation feedback data.

9. A smart control system for a CO2 refrigeration system air cooler, used to implement the method as described in any one of claims 1-8, characterized in that, include: The multi-source sensing module is used to acquire operating environment parameters, including wet-bulb temperature data and liquid film distribution data, as well as operating status parameters, including operating pressure, liquid supply flow rate, air supply speed, and internal fluid flow direction, in the air cooler of the CO2 refrigeration system. The thermal modeling module is used to construct a thermal state tensor based on the operating environment parameters and the operating state parameters, generate a segmented heat transfer adjacency diagram by combining the internal fluid flow direction, calculate the influence weight of the heat transfer segment on the operating pressure, and generate a pressure sensitivity matrix. The marginal evaluation module is used to calculate the mass flow ratio of coolant to air based on the supply flow rate and the air speed, calculate the liquid film coverage based on the liquid film distribution data, input the mass flow ratio, the wet-bulb temperature data and the liquid film coverage into the pre-trained marginal benefit model to obtain the marginal cooling benefit of the coolant, and set the upper limit of coolant gain. The resistance calculation module is used to obtain the parameters of the liquid supply nozzle, determine the coolant particle size class in combination with the liquid supply flow rate, and determine the wind-side pressure drop penalty item based on the liquid film coverage, the coolant particle size class and the air supply speed. The target optimization module is used to generate candidate air supply speed sequences and candidate liquid supply flow sequences based on the pressure sensitivity matrix, the upper limit of coolant gain, the wind-side pressure drop penalty term, and the thermodynamic state tensor. The scheme verification module is used to extract the predicted liquid film thickness corresponding to the candidate liquid supply flow rate sequence and the predicted operating pressure corresponding to the candidate air supply speed sequence, and to perform evaporation reachability verification. The decision execution module is used to generate the final control sequence based on the results of the evaporation accessibility verification and to correct the marginal cooling benefit and pressure sensitivity matrix of the coolant.

10. A smart control device for a CO2 refrigeration system air cooler, characterized in that, The device includes: a memory and at least one processor, wherein the memory stores instructions; The processor calls the instructions in the memory to cause the intelligent control device for the air cooler of a CO2 refrigeration system to execute the intelligent control method for the air cooler of a CO2 refrigeration system as described in any one of claims 1-8.

Images