Automatic control method and system for energy-saving optimization of the cold end of an air-cooled island

Abstract

This invention relates to the field of heat exchange control technology, and particularly to an automatic control method and system for energy-saving optimization of the cold end of an air-cooled island. The method includes: dynamically partitioning the cold end of the air-cooled island based on aerodynamic characteristics and local heat flux characteristics; obtaining the average sub-region feature vector and sub-region feature parameters of each control sub-region; inputting the obtained sub-region feature parameters into a pre-constructed regional state type prediction model, and outputting the regional state of the corresponding control sub-region. This invention dynamically partitions the cold end of the air-cooled island based on overall aerodynamic characteristics and local heat flux characteristics, realizing multi-objective sub-region division. This allows each sub-region to form a control unit with the same aerodynamic behavior and heat load characteristics during operation, thereby identifying abnormal states such as local backflow, high-temperature coverage, flow deflection, and abnormal back pressure in real time. This avoids the energy waste and temperature rebound risk caused by traditional average regulation, and improves the stability of the local state of the cold end and maintains the aerodynamic consistency of the entire field.

Description

Technical Field

[0001] This invention relates to the field of heat exchange control technology, specifically to an automatic control method and system for energy-saving optimization of the cold end of an air-cooled island. Background Technology

[0002] The cold end of the air-cooled island is the final heat exchange area of ​​the air-cooled system, a crucial link in the final discharge of heat to the ambient air. Its efficiency directly affects the heat exchange effect, system energy consumption, and equipment operational safety. Current technologies often employ a global average adjustment strategy for cold end control, maintaining the overall outlet temperature by uniformly adjusting the fan unit, flow guide components, and bypass devices. However, this control method has significant shortcomings:

[0003] On the one hand, traditional control methods cannot identify local flow anomalies and changes in heat transfer status in real time. During cold-end operation, there may be phenomena such as backflow, high-temperature coverage, flow direction deflection, and air duct blockage. However, existing systems lack precise monitoring of three-dimensional flow velocity, flow direction, turbulence intensity, and local heat flux characteristics, making it impossible to detect these anomalies in a timely manner, resulting in insufficient heat transfer or overcooling in some areas.

[0004] On the other hand, existing control strategies aim at overall average performance, typically increasing fan power to meet the needs of low-heat-exchange areas. This leads to increased energy consumption per unit of heat exchange power, and exacerbates energy waste due to localized overcooling or overheating. Furthermore, the lack of dynamic optimization and differentiated adjustment capabilities based on real-time operating conditions makes it difficult for the system to achieve a dynamic balance between cold-end heat exchange efficiency and energy consumption under various operating conditions. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an automatic control method and system for energy-saving optimization at the cold end of an air-cooled island.

[0006] This invention employs the following technical solution: an automatic control method for energy-saving optimization at the cold end of an air-cooled island, comprising:

[0007] Based on aerodynamic characteristics and local heat flux characteristics, the cold end of the air-cooled island is dynamically partitioned, so that the cold end of the air-cooled island forms a multi-objective control sub-region with the same aerodynamic behavior and heat load characteristics during operation.

[0008] Obtain the average sub-region feature vector and sub-region feature parameters for each control sub-region;

[0009] The acquired sub-region feature parameters are input into the pre-built regional state type prediction model, which outputs the regional state type of the corresponding control sub-region and sets the label to obtain the regional state type label.

[0010] When the regional state of the control sub-region is any region other than the dynamic equilibrium region, the regional state type label, sub-region feature parameters and sub-region feature vector are input into the pre-constructed control parameter prediction model, and the corresponding set of recommended control parameters is output.

[0011] Based on the recommended set, the speed, guide angle, and opening of the fan unit, guide vane, and bypass valve corresponding to the control sub-zone are adjusted to achieve dynamic adjustment of air volume distribution, flow direction guidance, and energy return in the control sub-zone.

[0012] As a further description of the above technical solution: the method for performing the dynamic partitioning to obtain the control sub-region includes:

[0013] A three-dimensional spatial model is established based on the cold end structure of the air-cooled island, and the model is divided into equal-interval sections to form an initial three-dimensional mesh.

[0014] A three-dimensional wind speed sensor, wind direction sensor, heat flow meter and surface temperature sensor array are deployed at the cold end to collect the aerodynamic characteristics and local heat flux characteristics of the cold end at preset time intervals, and the data are preprocessed.

[0015] Calculate aerodynamic and local heat flux characteristics on each grid cell and generate a grid feature vector for each grid cell;

[0016] For two adjacent grid cells, aerodynamic feature similarity and heat flux feature similarity are extracted, and composite similarity is calculated based on aerodynamic feature similarity and heat flux feature similarity.

[0017] Traverse all grid cells, set a preset composite similarity threshold, compare and analyze the obtained composite similarity with the preset composite similarity threshold, merge the grid cells to form control sub-regions.

[0018] As a further description of the above technical solution: the aerodynamic characteristics include local flow velocity, flow direction deviation angle, and turbulence intensity;

[0019] The local heat flux characteristics include surface temperature gradient and local surface heat flux.

[0020] As a further description of the above technical solution: the method for obtaining composite similarity includes:

[0021] For each adjacent pair of grid cells and grid cells Calculate aerodynamic feature similarity;

[0022] For each adjacent pair of grid cells and grid cells Calculate the similarity of heat flux characteristics based on local heat flux characteristics;

[0023] The composite similarity is obtained by weighted summation of the acquired aerodynamic feature similarity and heat flux feature similarity.

[0024] As a further description of the above technical solution: the method for merging grid cells to form control sub-regions includes:

[0025] Traverse all grid cells, preset a composite similarity threshold, and merge adjacent grid cells with a composite similarity greater than or equal to the composite similarity threshold to form a control sub-region;

[0026] A preset threshold for the number of grid cells in a control sub-region is set. The number of grid cells in the control sub-region is compared with the preset threshold. When the number of grid cells in the control sub-region is less than or equal to the preset threshold, the control sub-region is marked as a sub-region to be merged.

[0027] When a sub-region to be merged or a single isolated grid cell appears, the sub-region to be merged or the single isolated grid cell is merged with the control sub-region with the fewest grid cells in the adjacent control sub-region.

[0028] As a further description of the above technical solution: the sub-region feature vector is obtained by obtaining the grid feature vector of all grid cells in the control sub-region and then dividing it by the number of grid cells;

[0029] The sub-region characteristic parameters include average outlet temperature, temperature rise rate, inlet wind speed, and pressure difference.

[0030] As a further description of the above technical solution: the regional state types of the control sub-region include: insufficient heat exchange zone, excessive energy consumption zone, abnormal back pressure zone, excessive cooling zone, and dynamic balance zone.

[0031] As a further description of the above technical solution: the training method of the region state type prediction model includes:

[0032] H sets of state type training data are collected in advance, where H is a positive integer greater than 0. The state type training data includes sub-region feature parameters and corresponding regional state type labels of control sub-regions. The corresponding regional state type labels are set for the heat exchange insufficiency zone, energy consumption insufficiency zone, back pressure abnormal zone, cooling overload zone and dynamic balance zone respectively.

[0033] Gradient boosting regression tree model is used as the regional state type prediction model. The initial hyperparameters are set as follows: the collected state type training data are divided into training set, validation set and test set according to a preset ratio.

[0034] The model is trained using the training set, employing a binary classification cross-entropy loss function. Leaf node weights are optimized using gradient descent, and model parameters are updated based on the negative gradient of the training set loss. Hyperparameters are tuned using Bayesian optimization. The validation set F1 score is calculated. Training is stopped when the F1 score improves by less than 0.01 for three consecutive times, and the model parameters with the highest validation set F1 score are saved. The trained model is evaluated using the test set. If the test set precision is ≥90%, the model performance evaluation meets the standards, and the model is deployed and applied.

[0035] As a further description of the above technical solution: the recommended set of control parameters is as follows:

[0036] ,in, Let n be the fan speed corresponding to the set label n. This refers to the deflector angle corresponding to the set label n. This represents the bypass valve opening corresponding to the set label n, where n is the label of the recommended set of control sub-region regulation parameters.

[0037] An automatic control system for optimizing energy saving at the cold end of an air-cooled island, used to implement the aforementioned automatic control method for optimizing energy saving at the cold end of the air-cooled island, the system comprising:

[0038] The partitioning module dynamically partitions the cold end of the air-cooled island based on aerodynamic characteristics and local heat flux characteristics, so that the cold end of the air-cooled island forms a multi-objective control sub-region with the same aerodynamic behavior and heat load characteristics during operation.

[0039] The parameter acquisition module obtains the average sub-region feature vector and sub-region feature parameters for each control sub-region.

[0040] The state prediction module inputs the acquired sub-region feature parameters into the pre-built regional state type prediction model, outputs the corresponding regional state type of the control sub-region, and sets the label to obtain the regional state type label.

[0041] The control parameter generation module, when the regional state of the control sub-region is any region other than the dynamic equilibrium region, inputs the regional state type label, sub-region feature parameters and sub-region feature vector into the pre-constructed control parameter prediction model, and outputs the corresponding set of recommended control parameters.

[0042] The execution control module adjusts the speed, guide angle, and opening of the fan unit, guide vane, and bypass valve corresponding to the control sub-zone according to the recommended set, so as to realize the dynamic adjustment of air volume distribution, flow direction guidance, and energy return in the control sub-zone.

[0043] Beneficial effects:

[0044] The automatic control method and system for energy-saving optimization of the cold end of an air-cooled island provided by this invention achieves multi-objective sub-region division of the cold end of the air-cooled island through dynamic zoning based on the overall aerodynamic characteristics and local heat flux characteristics. This enables each sub-region to form a control unit with the same aerodynamic behavior and heat load characteristics during operation, thereby enabling real-time identification of abnormal states such as local backflow, high temperature coverage, flow deflection, and abnormal back pressure. By accurately acquiring and quantifying the characteristic parameters and characteristic vectors of the sub-regions, combined with a pre-constructed regional state type prediction model, the system can quickly determine the operating state type of each sub-region, including insufficient heat exchange zone, excessive energy consumption zone, abnormal back pressure zone, over-cooling zone, and dynamic equilibrium zone. This avoids the energy waste and temperature rebound risk caused by traditional average regulation, and achieves improved stability of the local state of the cold end and maintenance of overall aerodynamic consistency. Attached Figure Description

[0045] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0046] Figure 1 A flowchart of an automatic control method for energy-saving optimization of the cold end of an air-cooled island provided in Embodiment 1 of the present invention;

[0047] Figure 2 A flowchart of a method for performing the dynamic partitioning to obtain a control sub-region provided in Embodiment 1 of the present invention;

[0048] Figure 3 This is a flowchart of a method for merging grid cells to form a control sub-region, as provided in Embodiment 1 of the present invention.

[0049] Figure 4 This is a module connection diagram of the automatic control system for energy-saving optimization of the cold end of the air-cooled island provided in Embodiment 2 of the present invention. Detailed Implementation

[0050] To make the technical means, creative features, objectives, and effects of this invention readily understandable, the invention is further described below with reference to specific illustrations. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0051] Example 1

[0052] Please see Figures 1-3 This invention provides a technical solution: an automatic control method for energy-saving optimization of the cold end of an air-cooled island, comprising:

[0053] It should be noted that the cold end of the air-cooled island refers to the end heat exchange area of ​​the air-cooled island, which is the part of the entire power generation process where heat is finally discharged into the ambient air. Its working process is as follows: cooling air enters the heat exchange fin area from the outside, heat is dissipated into the atmosphere through the fins and tube bundles, and the cooled air is discharged by the fan.

[0054] Based on aerodynamic characteristics and local heat flux characteristics, the cold end of the air-cooled island is dynamically partitioned, so that the cold end of the air-cooled island forms a multi-objective control sub-region with the same aerodynamic behavior and heat load characteristics during operation.

[0055] The method for performing the dynamic partitioning to obtain the control sub-region includes:

[0056] A three-dimensional spatial model is established based on the cold end structure of the air-cooled island (air duct length, heat sink arrangement, and guide plate layout), and then divided at equal intervals to form an initial three-dimensional mesh.

[0057] A three-dimensional wind speed sensor, wind direction sensor, heat flow meter, and surface temperature sensor array are deployed at the cold end. The aerodynamic characteristics and local heat flux characteristics of the cold end are collected at preset time intervals. The data is then preprocessed, including low-pass filtering, median pulse rejection, and outlier correction of the collected raw data.

[0058] Aerodynamic and local heat flux characteristics are calculated on each grid cell, and a grid feature vector is generated for each grid cell to provide quantitative indicators for partitioning.

[0059] The aerodynamic characteristics include local flow velocity, flow direction deviation angle, and turbulence intensity;

[0060] The methods for obtaining the local flow velocity, flow direction deviation angle, and turbulence intensity include:

[0061] A three-dimensional wind speed vector with a timestamp is acquired by a three-dimensional wind speed sensor, and is represented as: ;

[0062] For each mesh cell, the instantaneous three-dimensional velocity components are obtained. , , Calculate the local flow velocity;

[0063] The formula for calculating the local flow velocity is: In the formula, For local flow velocity;

[0064] The formula for calculating the flow direction deviation angle is: In the formula, The flow direction deviation angle, It is an inverse cosine function. As the unit vector in the mainstream direction, The dot product of the three-dimensional wind speed vector and the reference flow direction. The magnitude of the velocity vector, i.e., the local flow velocity. The magnitude of the reference flow vector.

[0065] It should be noted that the reference flow direction is defined with reference to the mainstream direction of the cold end, such as [1,0,0].

[0066] The turbulence intensity is calculated using the following formula:

[0067] ;

[0068] In the formula, For turbulence intensity, For local flow velocity, The average speed within a preset time window, optionally with a time window length of 5-10 seconds;

[0069] The local heat flux characteristics include surface temperature gradient and local surface heat flux;

[0070] The method for obtaining the surface temperature gradient includes: obtaining the temperature of the grid cells through a surface temperature sensor array, calculating the spatial rate of change of temperature relative to the x-direction, the spatial rate of change of temperature in the y-direction, and the spatial rate of change of temperature in the z-direction, and obtaining the surface temperature gradient;

[0071] The surface temperature gradient is expressed as: ;

[0072] In the formula, For the surface temperature gradient, Let be the rate of change of temperature relative to the x-direction in space. This represents the rate of change of temperature relative to the Y direction in space. Let be the rate of change of temperature in space relative to the z-direction.

[0073] The local surface heat flux is acquired in real time by a heat flux meter and denoted as . .

[0074] The grid feature vector is represented as follows:

[0075] For two adjacent grid cells, aerodynamic feature similarity and heat flux feature similarity are extracted, and composite similarity is calculated based on aerodynamic feature similarity and heat flux feature similarity.

[0076] The method for obtaining composite similarity includes:

[0077] For each adjacent pair of grid cells and grid cells Calculate aerodynamic feature similarity;

[0078] The formula for calculating the aerodynamic feature similarity is: ;

[0079] In the formula, For grid cells and grid cells Aerodynamic feature similarity For grid cells aerodynamic characteristic vectors For grid cells aerodynamic characteristic vectors The sensitivity parameter is set manually based on experience, and the aerodynamic characteristic vector is represented as follows. ;

[0080] For each adjacent pair of grid cells and grid cells Calculate the similarity of heat flux characteristics based on local heat flux characteristics;

[0081] The formula for calculating the similarity of the heat flux features is as follows: ;

[0082] In the formula, For grid cells and grid cells Heat flux characteristic similarity For grid cells The heat flux feature vector. For grid cells The heat flux feature vector. This is a sensitivity parameter, set manually based on experience. .

[0083] The composite similarity is obtained by weighted summation of the acquired aerodynamic feature similarity and heat flux feature similarity.

[0084] Optionally, the formula for calculating the composite similarity is:

[0085] ;

[0086] In the formula, For composite similarity, and The weighting coefficients are used for calculation. It should be noted that the weighting coefficients in the formula are set by those skilled in the art based on actual conditions or obtained through simulation of a large amount of data.

[0087] Traverse all grid cells, preset composite similarity threshold, compare and analyze the obtained composite similarity with the preset composite similarity threshold, merge grid cells to form control sub-regions;

[0088] The method for merging grid cells to form control sub-regions includes:

[0089] Traverse all grid cells, preset a composite similarity threshold, and merge adjacent grid cells with a composite similarity greater than or equal to the composite similarity threshold to form a control sub-region;

[0090] A preset threshold for the number of grid cells in a control sub-region is set. The number of grid cells in the control sub-region is compared with the preset threshold. When the number of grid cells in the control sub-region is less than or equal to the preset threshold, the control sub-region is marked as a sub-region to be merged.

[0091] When a sub-region to be merged or a single isolated grid cell appears, the sub-region to be merged or the single isolated grid cell is merged with the control sub-region with the fewest grid cells in the adjacent control sub-region.

[0092] In this embodiment, the cold end is dynamically partitioned based on the overall aerodynamic characteristics and local heat flux characteristics. This allows the cold end to form multi-objective control sub-regions with the same aerodynamic behavior and heat load characteristics during operation. This enables real-time identification of local backflow, high-temperature coverage, flow deflection, and back pressure anomalies. It also ensures that the fan units, flow guide components, and bypass valves corresponding to each sub-region can perform differentiated adjustments according to the actual heat exchange requirements within the sub-region. This results in improved local stability and maintained overall aerodynamic consistency, further reducing energy waste and temperature rebound risks caused by average regulation.

[0093] Obtain the average sub-region feature vector and sub-region feature parameters for each control sub-region.

[0094] The sub-region feature vector is obtained by obtaining the grid feature vectors of all grid cells within the control sub-region and then dividing them by the number of grid cells;

[0095] The sub-region characteristic parameters include average outlet temperature, temperature rise rate, inlet wind speed, and pressure difference.

[0096] It should be noted that the average outlet temperature is obtained by detecting the temperature of each outlet in the control sub-area through a temperature sensor, summing the temperatures of each outlet, and then dividing by the number of outlets to obtain the average temperature.

[0097] The method for obtaining the temperature rise rate includes obtaining the average temperature of the last sampled temperature and the average temperature of the current sampled temperature at a preset time interval, and obtaining the temperature difference between the current average temperature and the average temperature of the last sampled temperature by dividing by time to obtain the temperature rise rate.

[0098] The inlet wind speed can be directly collected by controlling the corresponding fan unit in the sub-zone.

[0099] The pressure difference is collected by a pressure difference sensor to measure the pressure difference before and after the fins, reflecting the flow resistance and blockage degree of the unit.

[0100] The acquired sub-region feature parameters are input into the pre-built regional state type prediction model, which outputs the regional state type of the corresponding control sub-region and sets the label to obtain the regional state type label.

[0101] The regional state types of the control sub-region include: insufficient heat exchange zone, excessive energy consumption zone, abnormal back pressure zone, excessive cooling zone, and dynamic balance zone.

[0102] It should be noted that in the insufficient heat exchange zone, the instantaneous heat exchange load is significantly lower than the expected heat exchange power;

[0103] In areas with excessively high energy consumption, the corresponding heat exchange intensity is close to or exceeds the design limit, but the corresponding fan power consumption is high, and the energy consumption per unit heat exchange power increases.

[0104] Flow deflection zone, which corresponds to a local flow direction deviating from the mainstream direction by an angle exceeding a set threshold, such as 15°-25°, forming a non-uniform flow field;

[0105] An abnormal back pressure zone corresponds to a local pressure higher than the average of adjacent areas, indicating backflow or local air duct blockage.

[0106] The overcooling zone corresponds to a local temperature that is significantly lower than the average cold end outlet temperature, which poses a risk of overcooling or condensation at the cold end.

[0107] The dynamic equilibrium zone corresponds to the difference between the instantaneous heat exchange load and the expected heat exchange power being within the allowable deviation range (e.g., ±5%).

[0108] Training methods for regional state type prediction models include:

[0109] H sets of state type training data are collected in advance, where H is a positive integer greater than 0. The state type training data includes sub-region feature parameters and corresponding regional state type labels of control sub-regions. It should be noted that the initial regional state of the control sub-region is calibrated by the staff. The insufficient heat exchange zone, excessive energy consumption zone, abnormal back pressure zone, excessive cooling zone, and dynamic equilibrium zone are respectively assigned corresponding labels, namely 01, 02, 03, 04, and 05. That is, output label 01 corresponds to the insufficient heat exchange zone, output label 02 corresponds to the excessive energy consumption zone, output label 03 corresponds to the abnormal back pressure zone, output label 04 corresponds to the excessive cooling zone, and output label 05 corresponds to the dynamic equilibrium zone.

[0110] The gradient boosting regression tree model is used as the regional state type prediction model. The initial hyperparameters are set as follows: the number of decision trees is 100-200, the maximum depth of a single tree is 5-8 (to control overfitting and avoid learning noisy features), the minimum number of samples for node splitting is 10, and the maximum number of features considered during splitting is 3.

[0111] The collected state type training data is divided into training set, validation set and test set according to a preset ratio; optionally, the ratio is 6:3:1.

[0112] The model is trained using a training set, with a binary classification cross-entropy loss function as the loss function. The weights of the leaf nodes are optimized using gradient descent, and the model parameters are updated based on the negative gradient of the training set loss. The hyperparameters are tuned using a Bayesian optimization method. The optimization range includes: 20-50 trees, a learning rate of 0.01-0.1, and a regularization coefficient of 0.05-0.2.

[0113] Calculate the validation set F1 score for every 20 trees in each iteration; when the F1 score improvement is less than 0.01 for 3 consecutive iterations (60 trees), stop training to avoid overfitting the model to specific scenarios in the training set; save the model parameters with the highest validation set F1 score.

[0114] The trained model is evaluated using a test set. If the test set precision is ≥90% and the model performance evaluation meets the standard, it can be deployed and applied.

[0115] In this embodiment, the anomaly type of the control sub-region is obtained in real time by using the regional state type label output by the prediction model. This enables adaptive matching of the optimal control strategy for the actual operating state of different sub-regions, thereby achieving differentiated adjustment of fan speed, guide angle and bypass valve opening.

[0116] When the region state of the control sub-region is a dynamic equilibrium region, the current operating state is maintained;

[0117] When the regional state of the control sub-region is any region other than the dynamic equilibrium region, the regional state type label, sub-region feature parameters and sub-region feature vector are input into the pre-constructed control parameter prediction model, and the corresponding set of recommended control parameters is output.

[0118] The recommended set of control parameters is as follows:

[0119] ,in, Let n be the fan speed corresponding to the set label n. This refers to the deflector angle corresponding to the set label n. This represents the bypass valve opening corresponding to the set label n, where n is the label of the recommended set of control sub-region regulation parameters;

[0120] The training method for the control sub-region regulation parameter prediction model includes:

[0121] Historical control parameter data of the control sub-region is collected. The historical control parameter data is collected under the condition that the control effect meets the target. K sets of control parameter training data are collected, where K is a positive integer greater than 0. The control parameter training data includes the regional state type label of the control sub-region, the sub-region feature parameters and the sub-region feature vector, as well as the corresponding control parameter recommendation set label.

[0122] The region state type label, sub-region feature parameters, and sub-region feature vectors are converted into a corresponding set of feature vectors. Each set of feature vectors is used as the input to the control sub-region regulation parameter prediction model. The control sub-region regulation parameter prediction model outputs a set of recommended regulation parameter labels corresponding to each set of region state type label, sub-region feature parameters, and sub-region feature vectors. The prediction target is the set of recommended regulation parameter labels actually corresponding to each set of region state type label, sub-region feature parameters, and sub-region feature vectors. The training objective is to minimize the loss function value of the control sub-region regulation parameter prediction model. Training stops when the loss function value of the control sub-region regulation parameter prediction model is less than or equal to the preset target loss value. The control sub-region regulation parameter prediction model is a deep learning model.

[0123] The system acquires the fan unit, guide vane, and bypass valve corresponding to the control sub-zone. Based on the control parameters in the recommended set of control parameters for the control sub-zone, it adjusts the fan speed, guide vane agitation, and bypass valve opening to achieve dynamic adjustment of air volume distribution, flow direction guidance, and energy return in the control sub-zone.

[0124] Based on the regional characteristics of different control sub-regions, differentiated adjustment strategies are implemented for the corresponding fan units, flow guide components, and bypass valves to achieve dynamic balance of aerodynamic state and heat exchange efficiency and energy consumption optimization in each cold end region.

[0125] In this embodiment, the present invention achieves multi-objective sub-region division of the cold end of the air-cooled island through dynamic partitioning based on the overall aerodynamic characteristics and local heat flux characteristics. This enables each sub-region to form a control unit with the same aerodynamic behavior and heat load characteristics during operation, thereby enabling real-time identification of abnormal states such as local backflow, high temperature coverage, flow deflection, and abnormal back pressure. By accurately acquiring and quantifying the characteristic parameters and characteristic vectors of the sub-regions, combined with a pre-constructed regional state type prediction model, the system can quickly determine the operating state type of each sub-region, including insufficient heat exchange zone, excessive energy consumption zone, abnormal back pressure zone, over-cooling zone, and dynamic equilibrium zone. This avoids the energy waste and temperature rebound risk caused by traditional average regulation, and achieves improved stability of the local state of the cold end and maintenance of overall aerodynamic consistency.

[0126] Furthermore, this invention uses a control sub-zone regulation parameter prediction model to map the abnormal states and characteristic data of each sub-zone into differentiated adjustment schemes for fan speed, guide angle, and bypass valve opening, enabling each sub-zone to perform adaptive control according to actual heat exchange requirements. This method achieves dynamic adjustment of cold-end airflow distribution, flow direction guidance, and energy return, ensuring a dynamic optimal match between heat exchange efficiency and energy consumption in different sub-zones. Through historical data training and deep learning optimization, the system can adaptively cope with multiple operating conditions and load changes, achieving coordinated and balanced operation of the overall aerodynamics and thermal field at the cold end. This provides intelligent and refined control methods for energy-saving operation of air-cooled islands, significantly improving energy utilization efficiency and reducing operational risks.

[0127] Example 2

[0128] Please see Figure 4 This invention provides a technical solution: an automatic control system for energy-saving optimization at the cold end of an air-cooled island, which is used to implement the aforementioned automatic control method for energy-saving optimization at the cold end of an air-cooled island. The system includes:

[0129] The partitioning module dynamically partitions the cold end of the air-cooled island based on aerodynamic characteristics and local heat flux characteristics, so that the cold end of the air-cooled island forms a multi-objective control sub-region with the same aerodynamic behavior and heat load characteristics during operation.

[0130] The parameter acquisition module obtains the average sub-region feature vector and sub-region feature parameters for each control sub-region.

[0131] The state prediction module inputs the acquired sub-region feature parameters into the pre-built regional state type prediction model, outputs the corresponding regional state type of the control sub-region, and sets the label to obtain the regional state type label.

[0132] The control parameter generation module, when the regional state of the control sub-region is any region other than the dynamic equilibrium region, inputs the regional state type label, sub-region feature parameters and sub-region feature vector into the pre-constructed control parameter prediction model, and outputs the corresponding set of recommended control parameters.

[0133] The execution control module adjusts the speed, guide angle, and opening of the fan unit, guide vane, and bypass valve corresponding to the control sub-zone according to the recommended set, so as to realize the dynamic adjustment of air volume distribution, flow direction guidance, and energy return in the control sub-zone.

[0134] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An automatic control method for optimizing energy saving at the cold end of an air-cooled island, characterized in that, include: Based on aerodynamic characteristics and local heat flux characteristics, the cold end of the air-cooled island is dynamically partitioned, so that the cold end of the air-cooled island forms a multi-objective control sub-region with the same aerodynamic behavior and heat load characteristics during operation. The average sub-region feature vector and sub-region feature parameters of each control sub-region are obtained. The average sub-region feature vector is obtained by obtaining the grid feature vector of all grid cells in the control sub-region and then dividing it by the number of grid cells. The sub-region feature parameters include average outlet temperature, temperature rise rate, inlet wind speed and pressure difference. The acquired sub-region feature parameters are input into the pre-built regional state type prediction model, which outputs the regional state type of the corresponding control sub-region and sets the label to obtain the regional state type label. When the regional state of the control sub-region is any region other than the dynamic equilibrium region, the regional state type label, sub-region feature parameters and sub-region feature vector are input into the pre-constructed control parameter prediction model, and the corresponding set of recommended control parameters is output. Based on the recommended set, the speed, guide angle, and opening of the fan unit, guide vane, and bypass valve corresponding to the control sub-zone are adjusted to achieve dynamic adjustment of air volume distribution, flow direction guidance, and energy return in the control sub-zone.

2. The automatic control method for energy-saving optimization of the cold end of an air-cooled island according to claim 1, characterized in that, The method for performing the dynamic partitioning to obtain the control sub-region includes: A three-dimensional spatial model is established based on the cold end structure of the air-cooled island, and the model is divided into equal-interval sections to form an initial three-dimensional mesh. A three-dimensional wind speed sensor, wind direction sensor, heat flow meter and surface temperature sensor array are deployed at the cold end to collect the aerodynamic characteristics and local heat flux characteristics of the cold end at preset time intervals, and the data are preprocessed. Calculate aerodynamic and local heat flux characteristics on each grid cell and generate a grid feature vector for each grid cell; For two adjacent grid cells, aerodynamic feature similarity and heat flux feature similarity are extracted, and composite similarity is calculated based on aerodynamic feature similarity and heat flux feature similarity. Traverse all grid cells, set a preset composite similarity threshold, compare and analyze the obtained composite similarity with the preset composite similarity threshold, merge the grid cells to form control sub-regions.

3. The automatic control method for energy-saving optimization of the cold end of an air-cooled island according to claim 2, characterized in that, The aerodynamic characteristics include local flow velocity, flow direction deviation angle, and turbulence intensity; The local heat flux characteristics include surface temperature gradient and local surface heat flux.

4. The automatic control method for energy-saving optimization of the cold end of an air-cooled island according to claim 2 or 3, characterized in that, The method for obtaining composite similarity includes: For each adjacent pair of grid cells and grid cells Calculate aerodynamic feature similarity; For each adjacent pair of grid cells and grid cells Calculate the similarity of heat flux characteristics based on local heat flux characteristics; The composite similarity is obtained by weighted summation of the acquired aerodynamic feature similarity and heat flux feature similarity.

5. The automatic control method for energy-saving optimization of the cold end of an air-cooled island according to claim 2, characterized in that, The method for merging grid cells to form control sub-regions includes: Traverse all grid cells, preset a composite similarity threshold, and merge adjacent grid cells with a composite similarity greater than or equal to the composite similarity threshold to form a control sub-region; A preset threshold for the number of grid cells in a control sub-region is set. The number of grid cells in the control sub-region is compared with the preset threshold. When the number of grid cells in the control sub-region is less than or equal to the preset threshold, the control sub-region is marked as a sub-region to be merged. When a sub-region to be merged or a single isolated grid cell appears, the sub-region to be merged or the single isolated grid cell is merged with the control sub-region with the fewest grid cells in the adjacent control sub-region.

6. The automatic control method for energy-saving optimization of the cold end of an air-cooled island according to claim 1, characterized in that, The regional state types of the control sub-region include: insufficient heat exchange zone, excessive energy consumption zone, abnormal back pressure zone, excessive cooling zone, and dynamic balance zone.

7. The automatic control method for energy-saving optimization of the cold end of an air-cooled island according to claim 6, characterized in that, The training method for the regional state type prediction model includes: H sets of state type training data are collected in advance, where H is a positive integer greater than 0. The state type training data includes sub-region feature parameters and corresponding regional state type labels of control sub-regions. The corresponding regional state type labels are set for the heat exchange insufficiency zone, energy consumption insufficiency zone, back pressure abnormal zone, cooling overload zone and dynamic balance zone respectively. Gradient boosting regression tree model is used as the regional state type prediction model. The initial hyperparameters are set as follows: the collected state type training data are divided into training set, validation set and test set according to a preset ratio. The model is trained using the training set, employing a binary classification cross-entropy loss function. Leaf node weights are optimized using gradient descent, and model parameters are updated based on the negative gradient of the training set loss. Hyperparameters are tuned using Bayesian optimization. The validation set F1 score is calculated. Training is stopped when the F1 score improves by less than 0.01 for three consecutive times, and the model parameters with the highest validation set F1 score are saved. The trained model is evaluated using the test set. If the test set precision is ≥90%, the model performance evaluation meets the standards, and the model is deployed and applied.

8. The automatic control method for energy-saving optimization of the cold end of an air-cooled island according to claim 1, characterized in that, The recommended set of control parameters is as follows: in, Let n be the fan speed corresponding to the set label n. This refers to the deflector angle corresponding to the set label n. This represents the bypass valve opening degree corresponding to the set label n, where n is the label of the recommended set of control sub-region regulation parameters.

9. An automatic control system for energy-saving optimization at the cold end of an air-cooled island, used to implement the automatic control method for energy-saving optimization at the cold end of an air-cooled island as described in any one of claims 1-8, characterized in that, The system includes: The partitioning module dynamically partitions the cold end of the air-cooled island based on aerodynamic characteristics and local heat flux characteristics, so that the cold end of the air-cooled island forms a multi-objective control sub-region with the same aerodynamic behavior and heat load characteristics during operation. The parameter acquisition module acquires the average sub-region feature vector and sub-region feature parameters for each control sub-region. The average sub-region feature vector is obtained by acquiring the grid feature vectors of all grid cells in the control sub-region and then dividing them by the number of grid cells. The sub-region feature parameters include the average outlet temperature, temperature rise rate, inlet wind speed, and pressure difference. The state prediction module inputs the acquired sub-region feature parameters into the pre-built regional state type prediction model, outputs the corresponding regional state type of the control sub-region, and sets the label to obtain the regional state type label. The control parameter generation module, when the regional state of the control sub-region is any region other than the dynamic equilibrium region, inputs the regional state type label, sub-region feature parameters and sub-region feature vector into the pre-constructed control parameter prediction model, and outputs the corresponding set of recommended control parameters. The execution control module adjusts the speed, guide angle, and opening of the fan unit, guide vane, and bypass valve corresponding to the control sub-zone according to the recommended set, so as to realize the dynamic adjustment of air volume distribution, flow direction guidance, and energy return in the control sub-zone.

Images