An indirect air cooling system operation regulation method considering both freezing prevention and economy

Abstract

The application discloses an indirect air cooling system operation regulation method considering both freezing prevention and economy, which comprises the following steps: firstly, preprocessing and DBSCAN clustering analysis are performed on historical operation data, and a corresponding relationship among load, environmental conditions, circulating water flow and back pressure is determined by combining a steam turbine variable working condition calculation model, a condenser heat balance model and an air cooling tower theoretical calculation model; secondly, the best back pressure value in the whole working condition domain is solved according to a best back pressure calculation model; further, the integrated database is normalized, and a best back pressure prediction model and a louver opening degree combination prediction model are constructed based on an integrated learning algorithm of Bayesian optimization; and finally, the best operation back pressure of the indirect air cooling unit is predicted according to real-time data, and the best louver opening degree combination considering both freezing prevention and economy is further predicted. The method can realize the collaborative optimization of the economy and safety of the cold end system, significantly improve the operation efficiency and freezing prevention capacity of the indirect air cooling system, and has important engineering application value.

Description

Technical Field

[0001] This invention relates to the field of energy-saving control technology for thermal power units, and in particular to a method for regulating the operation of an indirect air-cooled system that balances antifreeze and economy. Background Technology

[0002] Against the backdrop of my country's new power system construction, the functional positioning of thermal power plants has shifted, and the traditional operating mode of cold-end systems under design conditions is no longer suitable for the flexible adjustment requirements of the new power system. As the terminal link of the thermal cycle of thermal power units, the operating quality of the cold-end system directly affects the unit's economic indicators. However, existing indirect air-cooled systems generally face a conflict between the dual objectives of "economy and safety".

[0003] Current research largely focuses on single-objective optimization: either establishing economic backpressure optimization models based on genetic algorithms or developing anti-freezing protection strategies based on temperature field monitoring. Meanwhile, existing collaborative optimization methods for indirect air-cooled systems rely solely on historical operating conditions for economic calculations, lacking mechanistic calibration of the optimal backpressure. Therefore, providing a multi-scale optimization model for indirect air-cooled systems that balances anti-freezing safety and economical operation has become a key technical bottleneck in improving the flexible adjustment capabilities of thermal power units. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the present invention aims to provide an operation and control method for an indirect air-cooled system that takes into account both antifreeze and economy. This method can achieve synergistic optimization of the economy and safety of the cold end system, significantly improve the operating efficiency and antifreeze capability of the indirect air-cooled system, and has important engineering application value.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for controlling the operation of an indirect air-cooled system that balances antifreeze and economy includes the following steps: S1: Historical operational data preprocessing and cluster analysis S1.1: Clean the historical operating data from winter and handle missing and outlier values ​​in the database; S1.2: The density-based clustering algorithm DBSCAN is used to perform cluster analysis on similar working conditions to obtain an initial database; The operating conditions obtained from cluster analysis should include the following parameters: load, circulating water flow rate, air-cooled tower circulating water inlet and outlet temperatures, condenser circulating water inlet and outlet temperatures, ambient temperature, ambient wind speed, and louver opening combination. S2: Construction of Operational Characteristic Analysis Model for Indirect Air-Cooled Systems The indirect air-cooled system operation characteristic analysis model includes a turbine variable operating condition calculation model, a condenser heat balance model, and an air-cooled tower theoretical calculation model. S3: Determination of Optimal Back Pressure for Units under Multiple Boundary Conditions S3.1: Obtain the exhaust steam flow rate and exhaust steam enthalpy under the actual operating conditions of the unit based on the turbine variable operating condition calculation model; S3.2: Based on the obtained exhaust steam flow rate and exhaust steam enthalpy, combined with the back pressure of the condenser heat balance model computer group under different loads, circulating water flow rates and condenser circulating water inlet temperatures; S3.3: Calculate the condenser circulating water inlet temperature under different ambient wind speeds, ambient temperatures, and circulating water flow rates based on the air-cooled tower theoretical calculation model, i.e., the air-cooled tower circulating water outlet temperature; then replace the back pressure obtained in S3.2 under different loads, circulating water flow rates, and condenser circulating water inlet temperatures with parameters to obtain the back pressure under different loads, ambient temperatures, ambient wind speeds, and circulating water flow rates. S3.4: Construct the optimal back pressure calculation model, and perform global optimization of the back pressure under different unit loads, ambient temperatures, ambient wind speeds, and circulating water flow rates to obtain the optimal back pressure; S4: Determination of Fine-grained Control Strategies for Energy Saving and Freeze Prevention in Indirect Air-cooled Systems S4.1 integrates the optimal back pressure into the initial database to obtain an "economic database" containing load, circulating water flow, ambient temperature, ambient wind speed, and optimal back pressure, and an "energy-saving and anti-freeze database" containing load, circulating water flow, ambient temperature, ambient wind speed, optimal back pressure, and louver opening combination. S4.2: Normalize the data and use an ensemble learning algorithm based on Bayesian optimization to construct the optimal back pressure prediction model and the louver opening combination prediction model. S4.3: Predict the optimal back pressure based on real-time data of the unit and the environment, and obtain a combination of louver opening degrees that balances antifreeze and economy based on real-time data and predicted optimal back pressure.

[0006] Preferably, the density-based clustering DBSCAN algorithm in S1.2 is as follows: For dataset D, given the parameter neighborhood radius e The minimum number of neighborhood samples to form the core point MinPTs Its clustering partitions satisfy: like and x j for e -reachable x i ,but and ;like x i for e -connected to x j and x j for e -connected to x l ,but x i and x l for e - Connected; Noise point identification: If sample points x m If a point is a non-core point and cannot be reached by any core point, then: In the formula: D is the sample dataset of winter historical operation data after preprocessing; x i , x j , x l , x m Let be any sample point in dataset D; N ε ( x i ) is the sample point x i of e - The set of sample points within the neighborhood; C k For the first k One cluster; e - Reachability refers to the neighborhood reachability relationship of sample points in the DBSCAN algorithm; e - Connected refers to the neighborhood connectivity of sample points in the DBSCAN algorithm; Noise The set of noise points is ultimately divided into various working condition clusters and their corresponding boundary conditions.

[0007] Preferably, the turbine variable operating condition calculation model in S2 is specifically as follows: In the formula: D j For steam turbine j The actual steam flow rate of the stage group D j,0 For steam turbine j The reference steam flow rate of the stage; P j For steam turbinej The actual steam pressure before the stage. P j+1 For steam turbine j The actual steam pressure after the stage, P j,0 For steam turbine j The baseline steam pressure before the stage is P j+1,0 For steam turbine j The reference steam pressure after the stage; T j For steam turbine j The actual steam temperature before the stage T j,0 For steam turbine j The baseline operating temperature of the steam before the stage; or The relative internal efficiency of the turbine stage; h 1 represents the steam inlet enthalpy of the turbine stage. h 2 represents the steam outlet enthalpy of the turbine stage. h 2s This represents the isentropic enthalpy value of the steam outlet of the turbine stage; h 2,j For steam turbine j Steam outlet enthalpy of the stage group, h 1,j For steam turbine j The steam inlet enthalpy of the stage group. or j For steam turbine j The relative internal efficiency of the class group H j For steam turbine j The isentropic enthalpy decrease of the class group; D st This refers to the exhaust steam flow rate of the steam turbine. D ms This refers to the main steam flow rate of the steam turbine. D r,j For steam turbine j Stage extraction steam flow rate; h c For the exhaust enthalpy of the steam turbine, h n This represents the enthalpy of the steam inlet of the last stage of the steam turbine. or n This refers to the relative internal efficiency of the last stage of the steam turbine. h n,s This is the isentropic enthalpy value of the steam outlet of the last stage of the steam turbine.

[0008] Preferably, the condenser heat balance model in S2 is as follows: In the formula: Q To exchange heat for the condenser D w The circulating water flow rate entering the condenser. h c ′ For the enthalpy of condensation, c p,w The specific heat capacity of the circulating water. t w1 This refers to the condenser circulating water inlet temperature. t w2 This refers to the condenser circulating water outlet temperature. t s This represents the saturation temperature corresponding to the condenser pressure. t The cooling water temperature rise is δ. t Condenser terminal temperature difference; P c This is back pressure.

[0009] Preferably, the theoretical calculation model for the air-cooled tower in S2 is as follows: In the formula: R a It is the air equivalent number. r a,i air density, v For ambient wind speed, A 0 represents the inlet cross-sectional area of ​​the air-cooled tower. n The number of cooling triangles, c p,a The specific heat capacity of air; NTU The number of heat transfer units in a single cooling column; K The overall heat transfer coefficient of the air-cooled tower is... A y The frontal area of ​​a single cooling column; e t For the heat transfer efficiency of the air-cooled tower, C The heat capacity flow rate ratio; t w,o The outlet temperature of the air-cooled tower circulating water. t w,i The inlet temperature of the circulating water in the air-cooled tower. t a,i The ambient temperature.

[0010] The preferred Bayesian optimization-based ensemble learning algorithm in S4.2 is as follows: An ensemble prediction model is constructed using gradient boosting decision trees as base learners. In the formula: Back pressure prediction value Or predicted value of venetian blind opening combination ; x As the feature vector, in the optimal back pressure prediction model x =[ N , t a,i , v , D w In the venetian blind opening combination prediction model x =[ N , t a,i , v , D w , P c,opt ];, K The total number of base learners, For the first k Gradient boosting of decision tree output values; Define the hyperparameter optimization objective function: In the formula: Θ is the set of hyperparameters of the ensemble learning model, including the learning rate, regularization coefficient, split threshold, tree depth, and minimum number of samples in the leaf node; L (Θ) represents the model loss value corresponding to the hyperparameter Θ; N s The total number of samples in the training set; y i For the first i The true value of each sample For the model under hyperparameter Θ, the first i Predicted values ​​for each sample; l Ω is the penalty coefficient; f Θ The structure penalty term of the model is used to suppress overfitting. Bayesian optimization uses a Gaussian process to establish a surrogate model for the objective function: In the formula: GP It is a Gaussian process; m (Θ) is the mean function of the Gaussian process, which characterizes the overall trend of the objective function; k(Θ,Θ′) is the kernel function of the Gaussian process, which characterizes the correlation between the hyperparameter Θ and the objective function value corresponding to Θ′; The next evaluation point is determined by maximizing the expected improvement function: In the formula: m (Θ) is the mean of the surrogate model's predictions for the hyperparameter Θ; y best This represents the optimal objective function value corresponding to the hyperparameters evaluated in the current iteration. x To explore the factors, the exploration and utilization of the equilibrium model; Φ is the cumulative distribution function of the standard normal distribution; s (Θ) represents the standard deviation of the surrogate model's prediction of hyperparameter Θ; is the probability density function of the standard normal distribution; Latin hypercube sampling is used to generate the initial hyperparameter set. Calculate the corresponding objective function value; for each iteration, maximize... IE The (Θ) function determines the next evaluation point. Train the model and calculate the objective function value, then update the observation set; when the maximum number of iterations is reached... T max or continuous S The improvement in the next iteration is less than the threshold. g When the time is right, training terminates and the optimal set of hyperparameters Θ is output. opt .

[0011] Compared with the prior art, the present invention has the following advantages: 1. Achieving synergistic optimization of economy and antifreeze safety, resolving the conflict between dual objectives: This invention constructs an "energy-saving and antifreeze database" that integrates the optimal back pressure and louver antifreeze opening combination, thus incorporating antifreeze safety constraints into the entire process of economic optimization; based on real-time data, it sequentially predicts the optimal back pressure and the optimal louver opening combination, ensuring that the condenser back pressure is within the economic range while avoiding the risk of radiator freezing in winter through precise control of louver opening; 2. Achieving optimal back pressure calibration from a mechanistic perspective, with optimization results more closely matching the actual operating characteristics of the unit: This invention constructs an operating characteristic analysis model that couples the turbine variable operating condition calculation model, the condenser thermal balance model, and the air-cooled tower theoretical calculation model. It clarifies the quantitative relationship of key parameters such as exhaust steam enthalpy, exhaust steam flow rate, back pressure, and environmental parameters, breaking through the limitations of traditional research that relies solely on historical operating condition statistical analysis. It achieves accurate calibration of the optimal back pressure across the entire operating range from the thermodynamic cycle mechanism level, making the optimal back pressure highly matched with the actual operating characteristics of the unit. 3. Achieving precise control across the entire operating range based on multi-boundary condition optimization: In the process of optimizing the optimal back pressure, this invention incorporates unit load, ambient temperature, ambient wind speed, and circulating water flow into the boundary conditions, achieving optimal back pressure optimization across the entire operating range under multi-parameter coupling; combined with a real-time data-driven prediction model, it can quickly respond to different operating conditions and environmental conditions of the unit, and output the optimal combination of back pressure and louver opening. Attached Figure Description

[0012] Figure 1 This is a flowchart of the method described in this invention.

[0013] Figure 2 This is a schematic diagram illustrating the working principle of an indirect air-cooled system. Detailed Implementation

[0014] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0015] Figure 1 A flowchart of an indirect air-cooling system operation control method that balances antifreeze and economy, provided by the present invention, mainly includes: S1: Historical operational data preprocessing and cluster analysis S1.1: Clean the historical operating data from winter and handle missing and outlier values ​​in the database; S1.2: The density-based clustering algorithm DBSCAN is used to perform cluster analysis on similar working conditions to obtain an initial database; Specifically, the density-based clustering algorithm DBSCAN is as follows: For dataset D, given the parameter neighborhood radius e The minimum number of neighborhood samples to form the core point MinPTs Its clustering partitions satisfy: like and x j for e -reachable x i ,but and ;like x i for e -connected to x j and x j for e -connected to x l ,but x i and x l for e - Connected; Noise point identification: If sample points x m If a point is a non-core point and cannot be reached by any core point, then: In the formula: D is the sample dataset of winter historical operation data after preprocessing; x i , x j , x l , x m Let be any sample point in dataset D; N ε ( x i ) is the sample point x i of e - The set of sample points within the neighborhood; C k For the first k One cluster; e - Reachability refers to the neighborhood reachability relationship of sample points in the DBSCAN algorithm; e - Connected refers to the neighborhood connectivity of sample points in the DBSCAN algorithm; Noise The set of noise points is ultimately divided into various working condition clusters and their corresponding boundary conditions.

[0016] Specifically, the initial database should include the following parameters: load, circulating water flow rate, air-cooled tower circulating water inlet and outlet temperatures, ambient temperature, ambient wind speed, and louver opening combination.

[0017] S2: Construction of Operational Characteristic Analysis Model for Indirect Air-Cooled Systems Specifically, the operational characteristic analysis model of the indirect air-cooled system includes a turbine variable operating condition calculation model, a condenser heat balance model, and an air-cooled tower theoretical calculation model.

[0018] Specifically, the calculation model for the turbine under varying operating conditions is as follows: In the formula: D j For steam turbine j The actual steam flow rate of the stage group D j,0For steam turbine j The reference steam flow rate of the stage; P j For steam turbine j The actual steam pressure before the stage. P j+1 For steam turbine j The actual steam pressure after the stage, P j,0 For steam turbine j The baseline steam pressure before the stage is P j+1,0 For steam turbine j The reference steam pressure after the stage; T j For steam turbine j The actual steam temperature before the stage T j,0 For steam turbine j The baseline operating temperature of the steam before the stage; or The relative internal efficiency of the turbine stage; h 1 represents the steam inlet enthalpy of the turbine stage. h 2 represents the steam outlet enthalpy of the turbine stage. h 2s This represents the isentropic enthalpy value of the steam outlet of the turbine stage; h 2,j For steam turbine j Steam outlet enthalpy of the stage group, h 1,j For steam turbine j The steam inlet enthalpy of the stage group. or j For steam turbine j The relative internal efficiency of the class group H j For steam turbine j The isentropic enthalpy decrease of the class group; D st This refers to the exhaust steam flow rate of the steam turbine. D ms This refers to the main steam flow rate of the steam turbine. D r,j For steam turbine j Stage extraction steam flow rate; h c For the exhaust enthalpy of the steam turbine, h n This represents the enthalpy of the steam inlet of the last stage of the steam turbine. or n This refers to the relative internal efficiency of the last stage of the steam turbine. h n,s This is the isentropic enthalpy value of the steam outlet of the last stage of the steam turbine.

[0019] Specifically, the condenser heat balance model is as follows: In the formula: Q To exchange heat for the condenser D w The circulating water flow rate entering the condenser. h c ′ For the enthalpy of condensation, c p,w The specific heat capacity of the circulating water. t w1 This refers to the condenser circulating water inlet temperature. t w2 This refers to the condenser circulating water outlet temperature. t s Δ represents the saturation temperature corresponding to the condenser pressure. t The cooling water temperature rise is δ. t Condenser terminal temperature difference; P c This is back pressure.

[0020] Specifically, the theoretical calculation model for air-cooled towers is as follows: In the formula: R a It is the air equivalent number. r a,i air density, v For ambient wind speed, A 0 represents the inlet cross-sectional area of ​​the air-cooled tower. n The number of cooling triangles, c p,a The specific heat capacity of air; NTU The number of heat transfer units in a single cooling column; K The overall heat transfer coefficient of the air-cooled tower is... A y The frontal area of ​​a single cooling column; e t For the heat transfer efficiency of the air-cooled tower, C The heat capacity flow rate ratio; t w,o The outlet temperature of the air-cooled tower circulating water. t w,i The inlet temperature of the circulating water in the air-cooled tower.t a,i The ambient temperature.

[0021] S3: Determination of Optimal Back Pressure for Units under Multiple Boundary Conditions S3.1: Obtain the exhaust steam flow rate and exhaust steam enthalpy under the actual operating conditions of the unit based on the turbine variable operating condition calculation model; S3.2: Based on the obtained exhaust steam flow rate and exhaust steam enthalpy, combined with the back pressure of the condenser heat balance model computer group under different loads, circulating water flow rates and condenser circulating water inlet temperatures; S3.3: Calculate the condenser circulating water inlet temperature under different ambient wind speeds, ambient temperatures, and circulating water flow rates based on the air-cooled tower theoretical calculation model, i.e., the air-cooled tower circulating water outlet temperature; then replace the back pressure obtained in S3.2 under different loads, circulating water flow rates, and condenser circulating water inlet temperatures with parameters to obtain the back pressure under different loads, ambient temperatures, ambient wind speeds, and circulating water flow rates. S3.4: Construct the optimal back pressure calculation model, and perform global optimization of the back pressure under different unit loads, ambient temperatures, ambient wind speeds, and circulating water flow rates to obtain the optimal back pressure; S4: Determination of Fine-grained Control Strategies for Energy Saving and Freeze Prevention in Indirect Air-cooled Systems S4.1 integrates the optimal back pressure into the initial database to obtain an "economic database" containing load, circulating water flow, ambient temperature, ambient wind speed, and optimal back pressure, and an "energy-saving and anti-freeze database" containing load, circulating water flow, ambient temperature, ambient wind speed, optimal back pressure, and louver opening combination. S4.2: Normalize the data and use an ensemble learning algorithm based on Bayesian optimization to construct the optimal back pressure prediction model and the louver opening combination prediction model. Specifically, the Bayesian optimized ensemble learning algorithm is as follows: An ensemble prediction model is constructed using gradient boosting decision trees as base learners. In the formula: Back pressure prediction value Or predicted value of venetian blind opening combination ; x As the feature vector, in the optimal back pressure prediction model x =[ N , t a,i , v , D w In the venetian blind opening combination prediction model x =[ N , t a,i , v, D w , P c,opt ];, K The total number of base learners, For the first k Gradient boosting of decision tree output values; Define the hyperparameter optimization objective function: In the formula: Θ is the set of hyperparameters of the ensemble learning model, including the learning rate, regularization coefficient, split threshold, tree depth, and minimum number of samples in the leaf node; L (Θ) represents the model loss value corresponding to the hyperparameter Θ; N s The total number of samples in the training set; y i For the first i The true value of each sample For the model under hyperparameter Θ, the first i Predicted values ​​for each sample; l Ω is the penalty coefficient; f Θ The structure penalty term of the model is used to suppress overfitting. Bayesian optimization uses a Gaussian process to establish a surrogate model for the objective function: In the formula: GP It is a Gaussian process; m (Θ) is the mean function of the Gaussian process, which characterizes the overall trend of the objective function; k (Θ,Θ′) is the kernel function of the Gaussian process, which characterizes the correlation between the hyperparameter Θ and the objective function value corresponding to Θ′; The next evaluation point is determined by maximizing the expected improvement function: In the formula: m (Θ) is the mean of the surrogate model's predictions for the hyperparameter Θ; y best This represents the optimal objective function value corresponding to the hyperparameters evaluated in the current iteration. x To explore the factors, the exploration and utilization of the equilibrium model; Φ is the cumulative distribution function of the standard normal distribution; s (Θ) represents the standard deviation of the surrogate model's prediction of hyperparameter Θ; is the probability density function of the standard normal distribution; Latin hypercube sampling is used to generate the initial hyperparameter set. Calculate the corresponding objective function value; for each iteration, maximize... IEThe (Θ) function determines the next evaluation point. Train the model and calculate the objective function value, then update the observation set; when the maximum number of iterations is reached... T max or continuous S The improvement in the next iteration is less than the threshold. g When the time is right, training terminates and the optimal set of hyperparameters Θ is output. opt .

[0022] S4.3: Predict the optimal back pressure based on real-time data of the unit and the environment, and obtain a combination of louver opening angles that balances antifreeze and economy based on real-time data and predicted optimal back pressure. Figure 2 This is a schematic diagram of the working principle of an indirect air-cooled system. The exhaust steam from the turbine condenses and releases heat in the condenser. The circulating cooling water carries the heat released by the exhaust steam into the air-cooled tower, and releases the heat into the atmosphere through heat exchange between the air-cooled tower and the air. The cooled circulating water then re-enters the condenser for circulation.

Claims

1. A method for controlling the operation of an indirect air-cooled system that balances antifreeze and economy, characterized in that, Includes the following steps: S1: Historical operational data preprocessing and cluster analysis S1.1: Clean the historical operating data from winter and handle missing and outlier values ​​in the database; S1.2: The density-based clustering algorithm DBSCAN is used to perform cluster analysis on similar working conditions to obtain an initial database; The operating conditions obtained from cluster analysis should include the following parameters: load, circulating water flow rate, air-cooled tower circulating water inlet and outlet temperatures (i.e., condenser circulating water inlet and outlet temperatures), ambient temperature, ambient wind speed, and louver opening combination. S2: Construction of Operational Characteristic Analysis Model for Indirect Air-Cooled Systems The indirect air-cooled system operation characteristic analysis model includes a turbine variable operating condition calculation model, a condenser heat balance model, and an air-cooled tower theoretical calculation model. S3: Determination of Optimal Back Pressure for Units under Multiple Boundary Conditions S3.1: Obtain the exhaust steam flow rate and exhaust steam enthalpy under the actual operating conditions of the unit based on the turbine variable operating condition calculation model; S3.2: Based on the obtained exhaust steam flow rate and exhaust steam enthalpy, combined with the back pressure of the condenser heat balance model computer group under different loads, circulating water flow rates and condenser circulating water inlet temperatures; S3.3: Calculate the condenser circulating water inlet temperature under different ambient wind speeds, ambient temperatures, and circulating water flow rates based on the air-cooled tower theoretical calculation model, i.e., the air-cooled tower circulating water outlet temperature; then replace the back pressure obtained in S3.2 under different loads, circulating water flow rates, and condenser circulating water inlet temperatures with parameters to obtain the back pressure under different loads, ambient temperatures, ambient wind speeds, and circulating water flow rates. S3.4: Construct the optimal back pressure calculation model, and perform global optimization of the back pressure under different unit loads, ambient temperatures, ambient wind speeds, and circulating water flow rates to obtain the optimal back pressure; S4: Determination of Fine-grained Control Strategies for Energy Saving and Freeze Prevention in Indirect Air-cooled Systems S4.1 integrates the optimal back pressure into the initial database to obtain an "economic database" containing load, circulating water flow, ambient temperature, ambient wind speed, and optimal back pressure, and an "energy-saving and anti-freeze database" containing load, circulating water flow, ambient temperature, ambient wind speed, optimal back pressure, and louver opening combination. S4.2: Normalize the data and use an ensemble learning algorithm based on Bayesian optimization to construct the optimal back pressure prediction model and the louver opening combination prediction model. S4.3: Predict the optimal back pressure based on real-time data of the unit and the environment, and obtain a combination of louver opening degrees that balances antifreeze and economy based on real-time data and predicted optimal back pressure.

2. The method for controlling the operation of an indirect air-cooled system that balances antifreeze and economy according to claim 1, characterized in that, The density-based clustering algorithm DBSCAN in S1.2 is as follows: For dataset D, given the parameter neighborhood radius ε The minimum number of neighborhood samples to form the core point MinPTs Its clustering partitions satisfy: like and x j for ε -reachable x i ,but and ; like x i for ε -connected to x j and x j for ε -connected to x l ,but x i and x l for ε - Connected; Noise point identification: If sample points x m If a point is a non-core point and cannot be reached by any core point, then: In the formula: D is the sample dataset of winter historical operation data after preprocessing; x i , x j , x l , x m Let be any sample point in dataset D; N ε ( x i ) is the sample point x i of ε - The set of sample points within the neighborhood; C k For the first k One cluster; ε - Reachability refers to the neighborhood reachability relationship of sample points in the DBSCAN algorithm; ε - Connected refers to the neighborhood connectivity of sample points in the DBSCAN algorithm; Noise The set of noise points is ultimately divided into various working condition clusters and their corresponding boundary conditions.

3. The method for controlling the operation of an indirect air-cooled system that balances antifreeze and economy according to claim 1, characterized in that, The turbine variable operating condition calculation model in S2 is specifically as follows: In the formula: D j For steam turbine j The actual steam flow rate of the stage group D j,0 For steam turbine j The reference steam flow rate of the stage; P j For steam turbine j The actual steam pressure before the stage. P j+1 For steam turbine j The actual steam pressure after the stage, P j,0 For steam turbine j The baseline steam pressure before the stage is P j+1,0 For steam turbine j The reference steam pressure after the stage; T j For steam turbine j The actual steam temperature before the stage T j,0 For steam turbine j The baseline operating temperature of the steam before the stage; η The relative internal efficiency of the turbine stage; h 1 represents the steam inlet enthalpy of the turbine stage. h 2 represents the steam outlet enthalpy of the turbine stage. h 2s This represents the isentropic enthalpy value of the steam outlet of the turbine stage; h 2,j For steam turbine j Steam outlet enthalpy of the stage group, h 1,j For steam turbine j The steam inlet enthalpy of the stage group. η j For steam turbine j The relative internal efficiency of the class group H j For steam turbine j The isentropic enthalpy decrease of the class group; D st This refers to the exhaust steam flow rate of the steam turbine. D ms This refers to the main steam flow rate of the steam turbine. D r,j For steam turbine j Stage extraction steam flow rate; h c For the exhaust enthalpy of the steam turbine, h n This represents the enthalpy of the steam inlet of the last stage of the steam turbine. η n This refers to the relative internal efficiency of the last stage of the steam turbine. h n,s This is the isentropic enthalpy value of the steam outlet of the last stage of the steam turbine.

4. The method for controlling the operation of an indirect air-cooled system that balances antifreeze and economy according to claim 1, characterized in that, The condenser heat balance model in S2 is specifically as follows: In the formula: Q To exchange heat for the condenser D w The circulating water flow rate entering the condenser. h c ′ For the enthalpy of condensation, c p,w The specific heat capacity of the circulating water. t w1 This refers to the condenser circulating water inlet temperature. t w2 This refers to the condenser circulating water outlet temperature. t s Δ represents the saturation temperature corresponding to the condenser pressure. t The temperature rise of the cooling water is δ. t Condenser terminal temperature difference; P c This is back pressure.

5. The method for controlling the operation of an indirect air-cooled system that balances antifreeze and economy according to claim 1, characterized in that, The specific theoretical calculation model for the air-cooled tower in S2 is as follows: In the formula: R a It is the air equivalent number. ρ a,i air density, v For ambient wind speed, A 0 represents the inlet cross-sectional area of ​​the air-cooled tower. n The number of cooling triangles, c p,a The specific heat capacity of air; NTU The number of heat transfer units in a single cooling column; K The overall heat transfer coefficient of the air-cooled tower is... A y The frontal area of ​​a single cooling column; ε t For the heat transfer efficiency of the air-cooled tower, C The heat capacity flow rate ratio; t w,o The outlet temperature of the air-cooled tower circulating water. t w,i The inlet temperature of the circulating water in the air-cooled tower. t a,i The ambient temperature.

6. The method for controlling the operation of an indirect air-cooled system that balances antifreeze and economy according to claim 1, characterized in that, The Bayesian optimization-based ensemble learning algorithm in S4.2 is as follows: An ensemble prediction model is constructed using gradient boosting decision trees as base learners. In the formula: Back pressure prediction value Or predicted value of venetian blind opening combination ; x As the feature vector, in the optimal back pressure prediction model x =[ N , t a,i , v , D w In the prediction model of venetian blind opening combination x =[ N , t a,i , v , D w , P c,opt ];, K The total number of base learners, For the first k Gradient boosting of decision tree output values; Define the hyperparameter optimization objective function: In the formula: Θ is the set of hyperparameters of the ensemble learning model, including the learning rate, regularization coefficient, split threshold, tree depth, and minimum number of samples in the leaf node; L (Θ) represents the model loss value corresponding to the hyperparameter Θ; N s The total number of samples in the training set; y i For the first i The true value of each sample For the model under hyperparameter Θ, the first i Predicted values ​​for each sample; λ Ω is the penalty coefficient; f Θ The structure penalty term of the model is used to suppress overfitting. Bayesian optimization uses a Gaussian process to establish a surrogate model for the objective function: In the formula: GP It is a Gaussian process; m (Θ) is the mean function of the Gaussian process, which characterizes the overall trend of the objective function; k (Θ,Θ′) is the kernel function of the Gaussian process, which characterizes the correlation between the hyperparameter Θ and the objective function value corresponding to Θ′; The next evaluation point is determined by maximizing the expected improvement function: In the formula: μ (Θ) represents the mean of the surrogate model's predictions for the hyperparameter Θ; y best This represents the optimal objective function value corresponding to the hyperparameters evaluated in the current iteration; ξ To explore the factors, the exploration and utilization of the equilibrium model; Φ is the cumulative distribution function of the standard normal distribution; σ (Θ) represents the standard deviation of the surrogate model's prediction of hyperparameter Θ; is the probability density function of the standard normal distribution; Latin hypercube sampling is used to generate the initial hyperparameter set. Calculate the corresponding objective function value; for each iteration, maximize... EI The (Θ) function determines the next evaluation point. Train the model and calculate the objective function value, then update the observation set; When the maximum number of iterations is reached T max or continuous S The improvement in the next iteration is less than the threshold. g When the time is right, training terminates and the optimal set of hyperparameters Θ is output. opt .

Images