Search

CN-122015564-A - Control method of closed cooling tower for dynamic dew point temperature tracking based on AI algorithm model

CN122015564ACN 122015564 ACN122015564 ACN 122015564ACN-122015564-A

Abstract

The application discloses a control method of a closed cooling tower for tracking dynamic dew point temperature based on an AI algorithm model, which is used for controlling the closed cooling tower and comprises the following steps of 1, collecting data in real time, 2, preprocessing the data, 3, carrying out an AI algorithm model of a supervised learning pre-training and reinforcement learning online optimization hybrid architecture by a central controller in a decision stage of AI algorithm model reasoning, outputting an optimal cooling mode based on preprocessed data, 4, executing and switching the cooling mode, controlling each module to execute a corresponding cooling mode in a dry cooling mode, an evaporation cooling mode and a collaborative cooling mode according to model instructions by the central controller, 5, continuously monitoring the running state to form closed loop optimization after the running state is monitored and adjusted by the closed loop feedback, and 6, processing a special scene.

Inventors

  • WANG MINGWEI
  • HAN GANGGANG
  • LIU SHOUFU
  • SHEN JIYONG
  • WANG CHUNWEI
  • SUN QUANQUAN
  • JIAO ZONGCHANG

Assignees

  • 山东凯翔传热科技有限公司

Dates

Publication Date
20260512
Application Date
20251231

Claims (10)

  1. 1. The control method of the closed cooling tower for dynamic dew point temperature tracking based on the AI algorithm model is characterized in that the control method is used for controlling the closed cooling tower, the closed cooling tower comprises a central controller, a main heat exchanger (2), an atomization system and a heat pump system, the main heat exchanger (2) is connected with a user side load heat source (3) through a closed working medium loop, the main heat exchanger (2) is also connected with a plate heat exchanger (4), and the plate heat exchanger (4) is connected with the heat pump system; the heat pump system comprises an evaporator (6), a compressor (8), a condenser (5) and a throttling and depressurization component (9), wherein the evaporator (6) and the condenser (5) are connected to two sides of the compressor (8), an output pipeline of the condenser (5) is connected with the plate heat exchanger (4), the throttling and depressurization component (9) is arranged on an output pipeline of the condenser (5), and the evaporator (6) is connected with an induced draft fan (7); The atomization system comprises an atomization pump (11), the atomization pump (11) is connected with an atomization nozzle (12), and the atomization nozzle (12) is arranged above the main heat exchanger (2); an air inlet (10) is arranged below the main heat exchanger (2); The central controller is connected with a sensor group, an atomizing pump (11), a draught fan (7), a compressor (8) and valves of an air inlet (10); the cooling mode of the closed cooling tower comprises a dry cooling mode, an evaporation cooling mode and a cooperative cooling mode; The control method comprises the following steps: step 1, acquiring data in real time, namely acquiring environment-type raw data, equipment-type raw data and energy efficiency-type raw data through a sensor group, wherein the environment-type raw data are real-time dew point temperature T1, environment temperature T2 and relative humidity RH, the equipment-type raw data are working medium inlet temperature T_in, working medium outlet temperature T_out, working medium flow Q, spray water temperature and fan rotating speed, and the energy efficiency-type raw data are unit cooling capacity energy consumption and mode switching frequency; Step 2, data preprocessing, namely performing filtering smoothing processing, normalization processing, 3 sigma principle detection outlier processing and feature extraction processing on the data, eliminating interference and extracting effective features; Step 3, in the decision stage of AI algorithm model reasoning, the central controller carries an AI algorithm model of an on-line optimization mixed architecture of supervised learning pre-training and reinforcement learning, and an optimal cooling mode is output based on the pre-processing data; Step 4, executing a cooling mode and switching modes, wherein the central controller controls each module to execute a corresponding cooling mode of a dry cooling mode, an evaporation cooling mode and a cooperative cooling mode according to a model instruction, wherein the triggering condition of the dry cooling mode is T1< T_out, the triggering condition of the evaporation cooling mode is T1 approximately T_out +/-2 ℃, and the triggering condition of the cooperative cooling mode is T1> T_out; step 5, closed loop feedback, namely monitoring and adjusting the running state, and continuously monitoring the running state by the system after the cooling mode is executed to form closed loop optimization; And 6, special scene processing, namely triggering a refrigeration source according to a dew point threshold value and starting the refrigeration source in a grading manner by using a working medium temperature compensation logic, starting the heat pump system to recover latent heat, gradually increasing the refrigeration load if the working medium temperature does not reach the standard, and maintaining a dry cooling mode when T1 is far lower than an extremely low dew point of T_out.
  2. 2. The method for controlling a closed cooling tower for dynamic dew point temperature tracking based on an AI algorithm model as set forth in claim 1, wherein the feature extraction in step 2 comprises the steps of: step 2.1, calculating derived features, and deriving key associated features from the original data; the working medium temperature deviation is calculated by the calculation formula as the current working medium outlet temperature-target cooling temperature; the calculation formula is real-time dew point temperature T1/ambient temperature T2; the working medium inlet temperature difference and the working medium outlet temperature difference are calculated according to a formula of working medium inlet temperature T_in-working medium outlet temperature T_out; the refrigeration source load matching degree is calculated by the calculation formula as the current refrigeration source load/rated refrigeration source load; the difference value between the spray water temperature and the dew point is calculated as the spray water temperature-real-time dew point temperature T1; The unit energy consumption cooling efficiency is calculated by the calculation formula of working medium temperature deviation/unit cooling energy consumption; Step 2.2, feature integration, namely merging the environment type original data, the equipment type original data and the energy efficiency type original data with the 6 derivative features calculated in the step 2.1 to form an initial feature pool; and 2.3, screening the features to form a model input feature set, eliminating redundant low-correlation features based on feature importance evaluation of the pre-trained random forest classifier, and finally reserving 12 core input features.
  3. 3. The method for controlling a closed cooling tower for dynamic dew point temperature tracking based on an AI algorithm model as set forth in claim 2, wherein the specific process of step 2.3 is as follows: Step 2.3.1, determining an evaluation index, and taking the Gini coefficient reduction amount originally supported by the random forest classifier as a core evaluation index; step 2.3.2, training an AI algorithm model and calculating importance; Inputting 17 initial features after feature integration and a corresponding optimal cooling mode label into a random forest classifier to complete full feature training; In the model training process, each decision tree records the Gini coefficient change of each feature when participating in splitting, and finally, the average value is obtained through the results of 200 decision trees to obtain the Gini importance score of each feature, and the Gini importance score is normalized to the [0,1] interval; For the features of 15 before the Gini score, further calculating the arrangement importance, namely randomly disturbing each feature for 10 times, re-calculating the model classification accuracy after each disturbing, and counting the average accuracy reduction value to be used as a secondary verification score; Step 2.3.3, feature importance sorting and threshold setting; combining the two index scores, calculating the comprehensive importance scores of the features according to the Gini score multiplied by 0.7+ arrangement importance score multiplied by 0.3, and sorting in descending order; Setting a screening threshold value, namely, a hard threshold value, namely, a comprehensive importance score is more than or equal to 0.05, namely, the contribution degree of the features to model decision is not lower than 5%, and the features lower than the threshold value are judged to be low-value redundant features; Step 2.3.4, screening, verifying and finally determining; screening candidate features according to a threshold value, and constructing a simplified feature set by about 12-14 items; Retraining a random forest classifier by using the candidate feature set, and verifying classification accuracy: If the accuracy is more than or equal to 95% of the original full-feature model, namely the performance of the model is not obviously reduced after screening, the candidate set is reserved; 2.3.5, extracting the characteristics and outputting a result; the finally output 12 core feature vectors cover three dimensions of environmental states, equipment operation and energy efficiency expression, and specifically comprise environmental classes, namely real-time dew point temperature, environmental temperature, relative humidity and dew point to environmental temperature ratio, equipment classes, namely working medium inlet and outlet temperature difference, working medium flow, spray water temperature and dew point difference, fan rotating speed and refrigeration source load matching degree, and energy efficiency classes, namely unit cooling capacity energy consumption, mode switching frequency and unit energy consumption cooling efficiency.
  4. 4. The method for controlling a closed cooling tower for dynamic dew point temperature tracking based on an AI algorithm model as set forth in claim 3, wherein said step 3 comprises the steps of: step 3.1, preliminary judgment of a pre-training model, namely inputting 12 feature vectors into a pre-trained random forest classifier, wherein the specific process is as follows: Step 3.1.1, pre-preparation, wherein the feature vector and the model state are ready; The input feature vector standardization comprises the steps of preprocessing 12 core features, completing normalization mapping and dimension alignment, wherein the dimension alignment is arranged according to a fixed sequence, namely, real-time dew point temperature, environment relative humidity, dew point to environment temperature ratio, working medium inlet and outlet temperature difference, working medium flow, spray water temperature and dew point difference, fan rotating speed, refrigeration source load matching degree, unit cooling energy consumption, mode switching frequency, unit energy consumption cooling efficiency and working medium temperature deviation, forming a feature vector of 1X 12 dimensions, and inputting the feature vector into a pre-trained random forest classifier; Step 3.1.2, classification reasoning of feature vectors by a single decision tree The 200 decision trees independently judge the input feature vectors, and the reasoning logic of each tree is consistent; The splitting rule is that each layer of nodes is determined through Gini coefficient minimization in the pre-training according to the optimal splitting threshold value of the characteristic value, if the working medium temperature deviation normalized value is more than 0.3 and does not reach the standard, the characteristic vector is distributed to the next level of child nodes until the leaf nodes are reached; The leaf node output category is that each leaf node corresponds to a unique cooling mode in dry cooling/evaporation cooling/cooperative cooling, and the leaf node category which the feature vector finally falls into is the preliminary judgment result of the decision tree; Step 3.1.3,200, integrated voting of decision trees; Each decision tree independently outputs 1 cooling mode judgment result to form a set of 200 independent judgment results, counts the number of tickets obtained from various cooling modes in the set, and selects the mode with the largest number of tickets obtained as the integrated judgment result of the random forest according to a few rules of obeying majority; step 3.1.4, mode suggestion output and confidence level labeling; The integrated judging result is used as an initial mode suggestion and is transmitted to a subsequent DQN reinforcement learning model; Calculating the mode confidence degree, namely, the confidence degree= (the highest ticket obtaining quantity/the total decision tree quantity) ×100% for the optimization reference of the subsequent DQN model, wherein if the confidence degree is more than or equal to 80%, the DQN model preferentially reserves the suggestion; Step 3.1.5, strengthening the optimization of the learning model; The DQN reinforcement learning model aims at optimizing long-term energy consumption, optimizes preliminary suggestions, takes a state space as a current sensor feature vector, takes an action space as three cooling modes, and takes a reward function as R=alpha (target temperature standard reaching rate) -beta (unit energy consumption) -gamma (mode switching frequency), wherein alpha, beta and gamma are adjustable weight coefficients, and finally outputs a determined cooling mode instruction.
  5. 5. The method for controlling a closed cooling tower for dynamic dew point temperature tracking based on an AI algorithm model as set forth in claim 3, wherein said step 3 further comprises the steps of: step 3.2, an AI algorithm self-learning mechanism, which comprises the following specific processes; Step 3.2.1, zhou Du threshold dynamic optimization, adjusting a mode switching threshold of dry cooling/evaporative cooling/cooperative cooling based on recent operation data aiming at a random forest classifier, and adapting to season or environmental humidity changes; data screening and pretreatment: triggering once a week, and extracting operation data of the last 30 days; Screening effective data, namely eliminating abnormal data caused by sensor faults and extreme interference, and keeping records of cooling effect reaching the standard rate of more than or equal to 80% and no abrupt change of energy consumption data; Grouping according to dew point intervals with low dew point <10 ℃, medium dew point 10-25 ℃ and high dew point >25 ℃, and counting the optimal combination of mode- & gt energy consumption- & gt standard rate in each group of data; Traversing possible threshold ranges by adopting a grid search method for each group of dew point intervals, and calculating comprehensive energy efficiency scores under different thresholds, wherein the score=0.6 standard reaching rate+0.4 (1-relative energy consumption); Retraining a random forest classifier by using the screened 30 days data, and fine-tuning the splitting threshold value of the decision tree; Step 3.2.2, performing online fine adjustment on the daily degree reinforcement learning, and adjusting the network weight of the DQN model through daily degree operation data feedback to enable the mode decision to be more fit with the target with the lowest long-term energy consumption; the method comprises the steps of storing experience data, triggering once every 24 hours of operation, adding state-action-rewarding data of the same day into an experience playback pool, fixing the capacity to 10000, and removing earliest data by adopting a first-in first-out strategy; the DQN model fine tuning training comprises the steps of randomly sampling 1000 pieces of data from an experience playback pool to serve as a fine tuning training set, setting training parameters, wherein the learning rate=0.001, training rounds=50, and the batch size=32; The balance of exploration and utilization is adopted, and an epsilon-greedy strategy is adopted, wherein the epsilon value is reduced along with the running time, the initial value is 0.3, the epsilon value is reduced by 0.05 every 30 days of running, and the epsilon value is reduced to 0.05 at the lowest value; and 3.2.3, scene migration self-adaption, namely quickly adapting to a new scene through migration learning when equipment is deployed to a new climate, so that the retraining cost is reduced.
  6. 6. The method for controlling a closed cooling tower for dynamic dew point temperature tracking based on an AI algorithm model as set forth in claim 3, wherein said step 3 further comprises the steps of: And 3.3, constructing a four-layer protection mechanism of a data layer, a model layer, a decision layer and an exception handling layer by anti-interference supplement of an AI algorithm model, wherein the specific process is as follows: step 3.3.1, redundant collection and filtering noise reduction of a data layer are carried out, and the data of an input model is ensured to be real and stable: The multi-sensor redundancy backup and fusion comprises the steps of aiming at a core input real-time dew point temperature T1, deploying 2 independent dew point sensors, collecting data at the same time, taking an average value of two sensor data as input under normal conditions, automatically judging that the single sensor data exceeds a reasonable range or does not respond, switching to the effective data of the other sensor if the single sensor data exceeds a reasonable range or does not respond, and triggering a subsequent abnormal processing mechanism if both the two sensors fail; The hierarchical filtering treatment comprises basic filtering, filtering all sensor data by adopting a sliding window of 5 seconds, wherein a calculation formula is that a current filtering value= (the previous 4 seconds of data+the current data)/5, smoothing high-frequency noise; setting an interference threshold value which is more than or equal to 10V/m as an exceeding standard, if the interference is not exceeding the standard, maintaining the original filtering strength, automatically expanding a sliding window to 10 seconds by a plurality of interference exceeding standards, simultaneously improving the prediction weight of Kalman filtering, and further suppressing noise; Step 3.3.2, model layer robustness training and anti-interference adaptation are carried out, tolerance of the model to interference is improved, correct decision can be output by the model when slight deviation exists in input data, and misguidance of the interfered data is avoided; Step 3.3.3, decision layer smoothing mechanism and switching limitation; setting a mode switching cooling period, namely setting a fixed cooling period, and not allowing continuous switching of modes within 5 minutes, continuously monitoring data by processing logic in the cooling period, executing switching if a new mode triggering condition is still met after the cooling period is finished, and maintaining a current mode if the condition is disappeared; If the mode proposal is collaborative cooling, three core characteristic conditions of dew point > target temperature, working medium temperature deviation >0, spray water temperature and dew point difference <2 ℃ are simultaneously satisfied, if only single characteristic is satisfied and other characteristics are contradictory, the mode proposal is judged to be misproposal caused by interference, and the mode proposal is temporarily not output, and the current mode is continued to be used; Step 3.3.4, performing exception handling layer fallback mechanism and manual intervention; Setting a fault judgment standard, if more than or equal to 3 core sensors fail or data are abnormal at the same time, namely the average value is +/-3 times of standard deviation, automatically triggering a safety mode, switching to an evaporation cooling mode, and simultaneously closing a refrigeration source grading starting logic to maintain fixed spraying quantity and fan rotating speed; And (3) decision deviation checking and manual auditing, namely calculating the deviation between the model output and the actual working condition in real time, wherein the deviation value= (average energy consumption in 10 minutes after mode switching/average energy consumption in 10 minutes before mode switching) - (cooling standard reaching rate in 10 minutes after mode switching/cooling standard reaching rate in 10 minutes before mode switching), setting a deviation threshold value, and if the deviation value is more than or equal to 20%, judging that the decision deviation exists when the energy consumption is suddenly increased after mode switching but the cooling effect is not improved, and triggering the manual auditing flow.
  7. 7. The method for controlling the closed cooling tower for dynamic dew point temperature tracking based on the AI algorithm model as set forth in claim 1, wherein the dry cooling mode in the step 4 is switched to the evaporative cooling mode, and the method comprises the following specific steps: the switching triggering condition is that the dew point temperature T1 is continuously met within 20 seconds and is within a range of the working medium outlet temperature T_out plus or minus 2 ℃ and a 20 second hysteresis threshold is required to be maintained, and the deviation value is corrected in extreme environments, wherein the ultra-low temperature environment deviation is less than 10 ℃ and is plus or minus 1 ℃, and the high temperature environment deviation is 35-40 ℃ and is plus or minus 3 ℃; the central controller continuously receives data such as T1, T_out and the like acquired by the sensor group, and outputs a switching instruction after confirming that the switching conditions all reach standards; the valve of the air inlet (10) is fully opened in a dry-cold mode, and the valve is kept in a fully opened state in the switching process, so that sufficient air is provided for the subsequent evaporation process; The atomizing pump (11) is started from a closed state, a water source is pumped under pressure and is conveyed to the atomizing nozzle (12), the nozzle is opened along with the starting, and tiny liquid drops are uniformly sprayed to the outer surface of the finned tube of the main heat exchanger (2); the induced draft fan (7) is gradually increased from a low rotating speed in a dry-cold mode, and finally a negative pressure environment of 0.8-0.9atm is formed in the tower, so that external air is guided to flow through the outer side of the finned tube and fully contact with atomized liquid drops; The plate heat exchanger (4) and the heat pump system are kept closed in the whole switching process and do not participate in heat exchange; the high-temperature working medium output by the user side load heat source (3) still flows through the main heat exchanger (2) along the original path, the heat exchange mode is switched from sensible heat exchange with dry air to latent heat exchange with atomized liquid drops, and the temperature of the working medium gradually approaches to the dew point; The sensor group continuously monitors data such as dew point, working medium temperature flow, negative pressure in the tower and the like, and feeds back the data to the central controller, dynamically adjusts the output power of the atomizing pump and the rotating speed of the induced draft fan, and ensures stable operation of the evaporative cooling mode.
  8. 8. The method for controlling the closed cooling tower for dynamic dew point temperature tracking based on the AI algorithm model as set forth in claim 1, wherein the evaporating cold mode in the step 4 is switched to the dry cold mode, and the method comprises the following specific steps: The switching condition is that 15 continuous seconds meet the condition that T1< T_out-2 ℃, the hysteresis threshold is maintained for 25 seconds, and the trigger threshold of a dry-cold mode is relaxed to T1< T_out+1 ℃ under the ultralow temperature environment of-10 ℃; after confirming that the switching condition reaches the standard through the sensor data, the central controller sends an instruction for stopping evaporating cold and starting dry cold; The atomizing pump (11) is stopped firstly, the atomizing nozzle (12) is closed, and the spraying of liquid drops to the main heat exchanger (2) is stopped, so that the waste of water resources is avoided; The induced draft fan (7) gradually reduces from medium-high rotation speed to low rotation speed in a dry-cold mode, only maintains basic airflow, meets the airflow requirement of sensible heat exchange of dry air and working medium, and reduces energy consumption; the valve of the air inlet (10) is kept in a fully-opened state, and external dry air is continuously introduced; the plate heat exchanger (4) and the heat pump system are still kept closed and do not participate in the heat exchange process; The working medium flows through the main heat exchanger (2) in a heat exchange mode, the sensible heat exchange is returned from the latent heat exchange, and the temperature reduction is realized by absorbing the heat of the working medium through dry air; The sensor group monitors the temperature of the dry air and the temperature difference data of the inlet and outlet of the working medium, and the cooling effect of the working medium reaches the standard and is prevented from being frequently started and stopped by the components.
  9. 9. The method for controlling the closed cooling tower for dynamic dew point temperature tracking based on the AI algorithm model as set forth in claim 1, wherein the evaporating cold mode switching cooperative cold mode in the step 4 comprises the following specific steps: The switching condition is that the continuous 10 seconds meet the condition that T1 is more than T_out+2 ℃, the hysteresis threshold is maintained for 15 seconds, and the high-temperature load is responded quickly, under the high-temperature environment of 35-40 ℃, the cooperative cold trigger threshold is advanced to the condition that T1 is more than T_out+1 ℃, and the cooperative cold operation is stable under the ultra-high-temperature environment of more than or equal to 40 ℃; After confirming the switching condition, the central controller outputs a cooperative cold mode instruction, and starts the heat pump system and adjusts the state of the related components on the basis of maintaining the operation of the evaporative cold core component; the atomizing pump (11) and the atomizing nozzle (12) continue to operate, so as to keep spraying liquid drops outside the fin tubes of the main heat exchanger (2) and maintain the atomizing evaporation process; The induced draft fan (7) maintains a medium-high rotation speed, maintains negative pressure of 0.8-0.9atm in the tower, and simultaneously guides the hot and humid air after heat exchange of the main heat exchanger to flow to the evaporator (6); The heat pump system is started from a closed state, namely a throttling and depressurization component (9) depressurizes and cools high-pressure refrigerant to form low-temperature liquid refrigerant to flow into the plate heat exchanger (4), a compressor (8) is started to compress gaseous refrigerant transmitted from the evaporator (6) into high-temperature high-pressure gas; The working medium output by the user side load heat source (3) firstly flows through the main heat exchanger (2) to exchange latent heat with atomized liquid drops, and enters the plate heat exchanger (4) after primary cooling, and then performs secondary heat exchange with low-temperature refrigerant to realize two-stage cooling; The valve of the air inlet (10) is kept fully open, so as to provide sufficient air for evaporation and heat pump latent heat recovery; The sensor group monitors dew point, working medium temperature flow and refrigeration source load data in real time, and the central controller dynamically adjusts atomization amount, fan rotating speed and compressor load, so that the working medium is ensured to reach target temperature, and meanwhile full-load starting of the refrigeration source is avoided.
  10. 10. The method for controlling the closed cooling tower for dynamic dew point temperature tracking based on the AI algorithm model as set forth in claim 1, wherein the switching of the evaporative cooling mode in the step 4 in cooperation with the cooling mode comprises the following specific steps: The switching condition is that the continuous 30 seconds meet the temperature of T1-out+2 ℃, the hysteresis threshold is maintained for 30 seconds, and the stable switching from the high-energy consumption mode to the low-energy consumption mode is ensured; after confirming that the switching condition reaches the standard through sensor data, the central controller sends an instruction for closing the heat pump system and maintaining the operation of the evaporative cooling core; The heat pump system is stopped step by step, the compressor (8) stops working and does not compress the refrigerant, the throttling and depressurization component (9) stops reducing the pressure and the temperature of the refrigerant, the cold energy transmission between the condenser (5) and the plate heat exchanger (4) is stopped, and the plate heat exchanger (4) stops participating in the heat exchange of working media; The atomizing pump (11) and the atomizing nozzle (12) continue to operate, and liquid drops are sprayed outside the fin tubes of the main heat exchanger (2) to maintain atomization evaporation latent heat exchange; The induced draft fan (7) maintains a medium-high rotation speed, maintains negative pressure in the tower, directly discharges the wet and hot air flowing through the main heat exchanger out of the tower, and does not lead the wet and hot air to the evaporator (6); The valve of the air inlet (10) is still kept fully open, so that sufficient air is provided for the evaporation process; The heat exchange path of the working medium is cooled in two stages from the main heat exchanger and the plate heat exchanger, and returns to perform latent heat exchange with atomized liquid drops only through the main heat exchanger; the sensor group continuously monitors working medium outlet temperature, negative pressure in the tower and energy consumption data, the central controller adjusts the output power of the atomizing pump and the rotating speed of the induced draft fan, the cooling effect is ensured to reach the standard, and meanwhile, the operation fluctuation caused by the shutdown of the heat pump is reduced.

Description

Control method of closed cooling tower for dynamic dew point temperature tracking based on AI algorithm model Technical Field The invention relates to the technical field of closed cooling towers, in particular to energy-saving closed cooling equipment based on dynamic adjustment of ambient dew point temperature, which is suitable for industrial cooling, process air conditioning and other scenes and can realize annual natural cold source utilization and intelligent energy-saving operation. Background Although the existing dew point temperature related cooling tower technology (such as negative pressure evaporation type, wide temperature zone precooling and the like) can realize the cooling effect close to or lower than the dew point temperature, the technology has key defects in various aspects in practical application, and the key defects are as follows: 1. The mode switching lacks intelligence and dynamic adaptation capability is insufficient The prior art is generally not provided with a high-precision real-time environment dew point sensor, the dew point is indirectly estimated by a multi-dependent temperature/humidity sensor, the error is large, the current environment dew point data cannot be accurately acquired, the system is difficult to perceive an extremely high/low dew point state, and a targeted adjustment strategy cannot be triggered. The cooling strategy of the closed tower is mostly operated based on fixed parameters or a single temperature threshold, lacks a dynamic optimization algorithm based on dew point, and has significantly reduced efficiency in extremely high/low dew point environments. 2. The closed loop has low coupling degree with dew point utilization, and poor heat dissipation matching property And the path branch is missing, namely the closed working medium only dissipates heat through a fixed air-cooled heat exchanger (or a spray heat exchanger), and the latent heat recovery branch switching logic based on the dew point is not available. In the high dew point environment, the working medium still leaves the air cooling branch, the latent heat of the hot and humid air cannot be absorbed preferentially through the heat pump plate type heat exchanger, so that the latent heat resource is wasted, and in the low dew point environment, the working medium still forcedly leaves the heat pump cooperative branch, so that the unnecessary energy consumption is increased. The heat exchange area distribution is fixed, namely the heat exchange area ratio of the closed working medium to the hot and humid air and the heat pump evaporator is preset to be a fixed value, and cannot be dynamically adjusted according to the dew point. The increase of the latent heat exchange area at the high dew point is not realized, which leads to insufficient latent heat absorption, and the increase of the sensible heat area at the low dew point is difficult to reduce the ineffective duty ratio of the latent heat exchange. And the working medium flow is regulated and rigidified, namely the rotation speed of the closed working medium pump is fixed, and a dew point linkage-free flow dynamic regulation mechanism is adopted. Sensible heat radiation is fast when the dew point is low, the flow can be reduced, the energy consumption is reduced, but the existing system still maintains high-flow operation, and energy waste is caused. 3. Waste heat is not effectively recovered, and the energy utilization rate is low The existing closed tower takes sensible heat radiation as a core, the technical principle is that latent heat recovery is not taken as a key link of energy utilization, only sensible heat taken away by spray water evaporation is focused, latent heat released by condensation of vapor in hot and humid air (the phase change latent heat of the vapor is about 2260kJ/kg and is several times of the sensible heat) is ignored, the hot and humid air is regarded as waste gas which completes heat radiation, not an energy source, and thermodynamic cycle design of latent heat conversion is lacked. Meanwhile, no latent heat conversion technology such as heat pump circulation is introduced, so that the latent heat of the hot and humid air cannot be converted into available high-grade heat, and a large amount of latent heat is directly discharged. 4. High dew point environment has weak adaptability, unbalanced cooling efficiency and energy consumption The evaporative cooling is essentially limited in that the traditional closed tower core relies on spray water evaporation and air sensible heat exchange, and the evaporation is driven by the humidity difference between air and spray water. At high dew point, the air is nearly saturated (humidity difference is extremely small), evaporation amount is suddenly reduced, sensible heat exchange becomes a main heat dissipation mode, but sensible heat efficiency (20-50W/(m 2. Mu. K)) is far lower than evaporative cooling (100-300W/(m 2. Mu. K)), and target temperature requirements are di