Air-cooled island anti-freezing intelligent simulation method based on digital twinning

Abstract

The application discloses an air-cooled island anti-freezing intelligent simulation method based on digital twinning, which comprises the following steps: obtaining air-cooled island operation real-time observation data to generate a structured observation vector; obtaining air-cooled island structure design data and thermal operation mechanism to construct a temperature field distribution simulation calculation unit; constructing a topological inversion framework based on the air-cooled island structure design data; generating an initial inversion temperature field distribution based on the structured observation vector in the topological inversion framework; generating a frozen boundary driving parameter set based on the initial inversion temperature field distribution; generating a frozen risk gradient field based on the frozen boundary driving parameter set; generating a double closed loop updating parameter set and a convergence judgment value based on the structured observation vector and the frozen risk gradient field; and generating a frozen risk distribution map and an anti-freezing control reference quantity based on the double closed loop updating parameter set. The application realizes topological constraint and risk driving collaborative updating.

Description

Technical Field

[0001] This invention relates to the field of power plant thermal system simulation and dynamic modeling, and in particular to an intelligent simulation method for antifreeze of air-cooled islands based on digital twins. Background Technology

[0002] In thermal power generating units, direct air-cooled systems, as an important alternative to water-cooled condensing systems, are widely used in large-capacity units in water-scarce regions. The air-cooled island is responsible for condensing turbine exhaust steam, and its operation is affected by multiple factors, including steam-side pressure, air-side temperature, fan speed, and zoned operation. In low-temperature winter environments, localized supercooling can easily occur within the finned tube bundles and zoned units, leading to freezing risks and impacting the safe operation of the unit and equipment lifespan.

[0003] Existing technologies mostly employ empirical rule-based anti-freezing operation strategies, controlling operation through manual adjustment of fan speed or zone switching, lacking the ability to fine-tune the overall temperature field distribution of the air-cooled island. Some technologies introduce single-mechanism models for thermodynamic calculations, but these typically fail to consider the coupling relationships between multiple zones and the influence of topology on parameter evolution, making it difficult to accurately reflect the spatial distribution characteristics of temperature state variables under complex operating conditions. Furthermore, traditional statistical update methods are mostly based on Gaussian assumptions, lacking adaptability to nonlinear and non-Gaussian risk distributions, and unable to effectively characterize the propagation characteristics of freezing risk within the spatial structure.

[0004] Furthermore, existing simulation technologies typically remain at the level of output results, lacking a systematic correlation with operational control parameters and failing to establish a clear mapping relationship between risk distribution and operational adjustment quantities, making it difficult to directly transform simulation results into engineering control basis.

[0005] Therefore, how to provide a digital twin-based intelligent simulation method for antifreeze of air-cooled islands is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent simulation method for antifreeze in air-cooled islands based on digital twins. This invention constructs a temperature field distribution simulation calculation unit and a topology inversion framework. Under the constraint of the topology block covariance tensor, it executes an ensemble Kalman inversion algorithm to update the parameter state vector. Furthermore, it introduces a freezing risk scalar field and a non-Gaussian variable flow equation to perform variable flow redistribution processing on the initial updated parameter set. Combined with the freezing risk gradient field, it achieves prediction-driven inversion updates, forming a dual-closed-loop updated parameter set. This method achieves the collaborative generation of convergent inverted temperature field distribution, freezing risk distribution map, and antifreeze control reference quantities, possessing advantages such as strong topology constraint capability, strong non-Gaussian risk characterization capability, and clear control correlation.

[0007] According to an embodiment of the present invention, a digital twin-based intelligent simulation method for antifreeze of air-cooled islands includes the following steps: Acquire real-time observation data of the air-cooled island operation, perform time alignment processing and anomaly removal processing, and splice the data in a fixed order to generate structured observation vectors; Obtain structural design data and thermodynamic operation mechanism of the air-cooled island, and construct a temperature field distribution simulation calculation unit; Based on the design data of the air-cooled island structure, the physical topology and topology embedding parameter space of the air-cooled island are constructed, the topology block covariance tensor is generated, and the topology inversion framework is constructed by combining the ensemble Kalman inversion algorithm. Within the topology inversion framework, based on the structured observation vectors, an ensemble update calculation is performed to generate an initial update parameter set, which is then loaded into the temperature field distribution simulation calculation unit to generate the initial inverted temperature field distribution. Based on the initial inverted temperature field distribution, a scalar field of freezing risk and a non-Gaussian variable flow equation are constructed, and variable flow redistribution processing is performed to generate a set of freezing boundary driving parameters. The parameter state vector is updated based on the parameter set driven by the frozen boundary, the forward calculation is performed to generate the temperature field distribution sequence, and the spatial gradient of the freezing risk scalar field is calculated to generate the freezing risk gradient field. Based on the structured observation vector and the frozen risk gradient field, the extended observation vector is generated by fusion, the prediction-driven inversion update calculation is performed to generate the prediction-driven update parameter set, the topological block covariance tensor is reconstructed, the variable flow redistribution processing is performed to generate the double closed-loop update parameter set, and the convergence judgment value is calculated. When the convergence judgment value meets the convergence judgment threshold, the double closed-loop update parameter set is loaded into the temperature field distribution simulation calculation unit to generate the converged inversion temperature field distribution, and the freezing risk distribution map and antifreeze control reference quantity are calculated.

[0008] Optionally, the generation of structured observation vectors includes: Collect real-time observation data of the air-cooled island operation, including steam side pressure, air side temperature, fan speed, and zone operation status parameters; Establish a unified time reference and map the sampling time of the real-time observation data of the air-cooled island operation to the unified time reference to generate time-aligned real-time observation data of the air-cooled island operation. Anomaly removal processing is performed on the time-aligned real-time observation data of the air-cooled island to generate pre-processed real-time observation data of the air-cooled island. The preprocessed real-time observation data of the air-cooled island are arranged according to a fixed field order, and the fields are concatenated in the execution order to generate a structured observation vector.

[0009] Optionally, the temperature field distribution simulation calculation unit includes: Obtain the structural design data of the air-cooled island, including the number of partition units, the arrangement relationship of fan units, and the connection relationship of heat exchange units. Sequentially establish the partition unit number set, fan unit number set, and heat exchange unit number set. Generate a connection relationship matrix according to the connection relationship of heat exchange units, and simultaneously generate the partition unit number order and fan unit number order. Each partition unit is divided into an independent thermal control body based on the set of partition unit numbers. Temperature state variables are established for each partition unit, pressure state variables are established based on the connection relationship of heat exchange units, and flow state variables are established based on the flow path between the fan unit and the partition unit, thus generating a set of state variables. The thermodynamic operation mechanism is obtained, including the steam-side condensation heat transfer process and the air-side convection heat transfer process. Convection heat transfer relationship is established based on the air-side convection heat transfer process. Pressure-flow balance relationship between pressure state variables and flow state variables is established based on the steam-side condensation heat transfer process. Coupled calculation relationship between temperature state variables, flow state variables and heat transfer is established based on the energy conservation relationship. Configure parameter state vectors for the coupled calculation relationship, including heat transfer attenuation coefficient, wind-induced disturbance transmission coefficient, and zonal coupling coefficient; By integrating the connection matrix, the set of state variables, the coupling calculation relationship, and the parameter state vector, a temperature field distribution simulation calculation unit is constructed.

[0010] Optionally, the construction of the topology inversion framework includes: Based on the air-cooled island structural design data, the set of partition unit numbers, the set of fan unit numbers, and the set of heat exchange unit numbers are uniformly indexed and mapped according to the connection relationship of heat exchange units and the arrangement relationship of fan units, generating a topology index sequence and generating an air-cooled island physical topology structure containing partition units, fan units, heat exchange units and connection relationships. Based on the physical topology of the air-cooled island, a topology embedding parameter space is established. Using the partition unit number as the block index, the parameter state vector is rearranged according to the partition unit number order. A parameter block structure is constructed for each partition unit. All parameter block structures are combined to generate a topology block parameter set. Based on the topological block parameter set, a centering process is performed on the sample dimension of the ensemble Kalman inversion algorithm, and the sample covariance matrix of the topological block parameter set is calculated on the sample dimension. Based on the connection relationship matrix in the physical topology of the air-cooled island, a topological adjacency weighting matrix is ​​constructed. Perform element-wise multiplication between the topological adjacency weighting matrix and the sample covariance matrix to generate the topological block covariance tensor; Based on the ensemble Kalman inversion algorithm and the topology block covariance tensor, a topology inversion framework is constructed. In the topology inversion framework, the covariance structure of the update gain is defined, and the topology block covariance tensor is used as the parameter covariance structure for the calculation of the update gain.

[0011] Optionally, generating the initial inversion temperature field distribution includes: Within the topology inversion framework, the current parameter state vector is loaded into the temperature field distribution simulation calculation unit, iterative calculation is performed, temperature state variables are generated, and the temperature field distribution corresponding to the current parameter state vector is formed. Based on the structured observation vector, the temperature state variable is extracted from the temperature field distribution corresponding to the current parameter state vector to generate the predicted temperature vector, and the difference between the predicted temperature vector and the structured observation vector is calculated to generate the observation residual vector. Within the topology inversion framework, based on the update gain and observation residual vector, a set update calculation is performed on the parameter state vector in the sample dimension to obtain the updated parameter sample set. Perform statistical summarization processing on the sample dimension of the updated parameter sample set, reconstruct and arrange it according to the original parameter dimension order of the parameter state vector, calculate the statistical expectation of each parameter on the sample dimension, and generate the initial updated parameter set. The initial set of updated parameters is loaded into the temperature field distribution simulation calculation unit, iterative calculation is performed, and the temperature state variables corresponding one-to-one with the set of partition unit numbers are output to generate the initial inverted temperature field distribution.

[0012] Optionally, the set of parameters for generating the frozen boundary includes: Based on the initial inverted temperature field distribution, a freezing critical temperature is set, and the temperature state variables of each partition unit are calculated with respect to the freezing critical temperature to obtain the temperature deviation. A freezing risk scalar field is then constructed based on the temperature deviation, with each partition unit in the freezing risk scalar field corresponding to a freezing risk scalar value. Based on the frozen risk scalar field, spatial smoothing calculation is performed according to the connection relationship matrix. The frozen risk scalar value of each partition unit is updated with neighborhood weighting to generate a smoothed frozen risk scalar field. Based on the smoothed freezing risk scalar field, the discrete gradient of the freezing risk scalar field along the direction of the connectivity matrix is ​​calculated to generate the freezing risk gradient vector. Based on the smoothed freezing risk scalar field and freezing risk gradient vector, a non-Gaussian variable flow equation is constructed to generate a non-Gaussian flow term. In the non-Gaussian variable flow equation, the topological block covariance tensor is introduced as a modulation term to scale and modulate the non-Gaussian flow term. Using the initial update parameter set as the initial condition, the non-Gaussian variable flow equation is substituted into it, and numerical integration update is performed under the set pseudo-time step control strategy to generate a parameter sample set after variable flow redistribution processing. Statistical summarization is performed on the parameter sample set after variable flow redistribution processing along the sample dimension, and the frozen boundary driving parameter set is reconstructed according to the original parameter dimension order of the parameter state vector.

[0013] Optionally, generating the freezing risk gradient field includes: Based on the frozen boundary driving parameter set, the coefficients in the parameter state vector are replaced and updated in the same order to obtain the updated parameter state vector; The updated parameter state vector is loaded into the temperature field distribution simulation calculation unit. Forward calculation is performed step by step according to the discrete time step within the preset time domain. The temperature state variables are iteratively updated in each time step, and a set of temperature state variables at multiple time steps is generated in the entire preset time domain. According to the order of the partition unit number set, the temperature state variable set corresponding to each moment in the preset time domain is arranged in chronological order to generate a temperature field distribution sequence; Based on the temperature field distribution sequence, the freezing risk scalar field at each time step is recalculated. Based on the scalar field of freezing risk at each time step, spatial difference calculation is performed according to the connection matrix to generate the freezing risk gradient vector at the corresponding time step. The freezing risk gradient vectors at each time point within the preset time domain are combined in chronological order to generate the freezing risk gradient field.

[0014] Optionally, generating the dual-loop update parameter set and calculating the convergence criterion value includes: Within the topology inversion framework, the structured observation vector and the frozen risk gradient field are spliced ​​and fused in the order of field dimensions, and the two vectors are extended and combined in the observation dimension to generate an extended observation vector. Based on the extended observation vector, prediction-driven inversion update calculation is performed within the topology inversion framework. The update gain is calculated based on the topology block covariance tensor. The set update calculation is performed on the parameter state vector in the sample dimension to generate the prediction-driven update parameter set. Based on the prediction-driven parameter set update, a centralization process is performed on the sample dimension to calculate the predicted sample covariance matrix; Element-wise multiplication is performed on the predicted sample covariance moment matrix and the topological adjacency weighted matrix to reconstruct the topological block covariance tensor, generating a reconstructed topological block covariance tensor that is linked to the frozen risk gradient field. Based on the reconstructed topological block covariance tensor, the non-Gaussian variable flow equation is called again within the topological inversion framework to perform numerical integration update and generate a parameter sample set after risk-driven redistribution. Perform statistical summarization processing on the sample dimension of the parameter sample set after risk-driven redistribution to generate a double closed-loop updated parameter set; Based on the double-loop update parameter set, the parameter difference norm of the parameter sample set in two consecutive double-loop update processes is calculated and compared with the convergence judgment threshold to generate a convergence judgment value.

[0015] Optionally, the calculation of the freezing risk distribution map and the antifreeze control reference quantity includes: Under the condition that the convergence judgment value meets the convergence judgment threshold, the double closed-loop update parameter set is loaded into the temperature field distribution simulation calculation unit in the order of the original parameter dimensions of the parameter state vector, and iterative calculation is performed to generate the converged inverse temperature field distribution. Based on the convergent inversion of the temperature field distribution, the deviation of the temperature state variables corresponding to each partition unit is calculated to generate a freezing risk scalar field. The freezing risk scalar field is then mapped into a two-dimensional matrix to generate a freezing risk distribution map. Based on the risk distribution map of freezing, risk levels are divided into each partition unit to generate a set of partition risk levels; Based on the risk level set of the partition and the physical topology of the air-cooled island, an initial antifreeze control reference quantity is generated according to the preset risk level corresponding control rules. Based on the spatial difference direction of the freezing risk scalar value between each partition unit in the freezing risk distribution map, the propagation direction of freezing risk in the physical topology of the air-cooled island is determined. Gradient direction correction is performed on the fan speed adjustment in the initial antifreeze control reference quantity to generate the antifreeze control reference quantity.

[0016] The beneficial effects of this invention are: First, this invention introduces the physical topology of the air-cooled island into the sample covariance calculation process by constructing a temperature field distribution simulation calculation unit and a topology inversion framework. This forms a topology block covariance tensor, which participates in parameter updates as the covariance structure for updating the gain in the ensemble Kalman inversion algorithm. This achieves a statistical correlation expression of the parameter state vector under physical structure constraints. Compared to update methods based solely on traditional statistical correlation, this method maintains the consistency of the parameter block structure under multi-partition coupling conditions, improving the stability and spatial consistency of the convergent inverted temperature field distribution.

[0017] Secondly, this invention constructs a freezing risk scalar field based on the initial updated parameter set, and introduces a non-Gaussian variable flow equation in conjunction with the freezing risk gradient field. This equation performs variable flow redistribution processing on the parameter samples, forming a risk-driven parameter reconstruction mechanism. This mechanism overcomes the limitations of the Gaussian assumption on the parameter distribution pattern, enabling the parameter update process to respond to the spatial variation characteristics of freezing risk between partitioned units. This enhances the ability to identify local extreme low-temperature regions and improves the sensitivity and accuracy of the freezing risk distribution map.

[0018] Furthermore, this invention constructs a prediction-driven inversion update mechanism by fusing structured observation vectors with the freezing risk gradient field. After reconstructing the topological block covariance tensor, it again invokes the non-Gaussian variable flow equation, forming a dual-closed-loop update parameter set that combines observation-driven and risk-driven approaches. This dual-closed-loop update mechanism achieves the coordinated generation of convergent inversion temperature field distribution and antifreeze control reference quantities, establishing a direct mapping relationship between the freezing risk distribution map and the zoned operating status parameters and fan speed adjustments. This enhances the correlation between simulation results and engineering control, improving the feasibility and reliability of antifreeze operation decisions for air-cooled islands. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of an intelligent simulation method for antifreeze of air-cooled islands based on digital twins proposed in this invention; Figure 2 This is a schematic diagram of the topology inversion framework structure in this invention; Figure 3 This is a schematic diagram of the freeze risk-driven dual closed-loop update mechanism in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figures 1-3 A digital twin-based intelligent simulation method for antifreeze protection of air-cooled islands includes the following steps: The real-time observation data of the air-cooled island operation is obtained. The real-time observation data of the air-cooled island operation includes steam side pressure, air side temperature, fan speed, and zone operation status parameters. Time alignment processing and anomaly removal processing are performed on the real-time observation data of the air-cooled island operation. The processed steam side pressure, air side temperature, fan speed, and zone operation status parameters are spliced ​​in a fixed order to generate a structured observation vector. The structural design data and thermodynamic operation mechanism of the air-cooled island are obtained. The structural design data of the air-cooled island includes the number of partition units, the arrangement relationship of fan units, and the connection relationship of heat exchange units. The thermodynamic operation mechanism includes the steam-side condensation heat transfer process and the air-side convection heat transfer process. Based on the structural design data of the air-cooled island, the connection relationship between partition units, fan units, and heat exchange units is established. Based on the thermodynamic operation mechanism, the coupling calculation relationship between temperature state variables, pressure state variables, and flow state variables is established. Parameter state vectors are configured for the coupling calculation relationship. Based on the connection relationship, the coupling calculation relationship, and the parameter state vector, a temperature field distribution simulation calculation unit is constructed. Based on the design data of the air-cooled island structure, a physical topology of the air-cooled island is constructed. The physical topology of the air-cooled island consists of partitioned units, fan units, heat exchange units, and connection relationships. A topology embedding parameter space is established based on the physical topology of the air-cooled island. The parameter state vector is mapped to the topology embedding parameter space according to the partitioning relationship of the physical topology of the air-cooled island to generate a topology block parameter set. A sample covariance matrix is ​​generated based on the topology block parameter set, and a topology adjacency weighting matrix is ​​constructed. The topology adjacency weighting matrix and the sample covariance matrix are coupled to generate a topology block covariance tensor. Finally, a topology inversion framework is constructed based on the ensemble Kalman inversion algorithm and the topology block covariance tensor. Within the topology inversion framework, the structured observation vector is used as the observation input, and the set update calculation is performed to generate the initial update parameter set. The initial update parameter set is then loaded into the temperature field distribution simulation calculation unit to generate the initial inverted temperature field distribution. Based on the initial inverted temperature field distribution, a freezing risk scalar field is constructed. The freezing risk scalar field is used to characterize the degree of deviation between the temperature state variables of each partition unit and the freezing critical temperature. Based on the freezing risk scalar field, a non-Gaussian variable flow equation is constructed, and the variable flow redistribution process is performed on the initial update parameter set to generate the freezing boundary driving parameter set. The parameter state vector is updated based on the parameter set driven by the frozen boundary. The updated parameter state vector is loaded into the temperature field distribution simulation calculation unit. Forward calculation is performed at multiple times within the preset time domain to generate the temperature field distribution sequence. The spatial gradient of the freezing risk scalar field is calculated based on the temperature field distribution sequence to generate the freezing risk gradient field. The structured observation vector is fused with the frozen risk gradient field to generate an extended observation vector. Within the topology inversion framework, the extended observation vector is used to perform prediction-driven inversion update calculations to generate a prediction-driven update parameter set. Based on the prediction-driven update parameter set, the prediction sample covariance matrix is ​​generated and the topology block covariance tensor is reconstructed. Based on the reconstructed topology block covariance tensor and combined with the non-Gaussian variable flow equation, variable flow redistribution processing is performed to generate a double-closed-loop update parameter set. The convergence judgment value is calculated based on the double-closed-loop update parameter set. When the convergence judgment value meets the convergence judgment threshold, the double closed-loop update parameter set is loaded into the temperature field distribution simulation calculation unit to generate the converged inversion temperature field distribution. Based on the converged inversion temperature field distribution, the freezing risk distribution map and antifreeze control reference quantity are calculated.

[0022] In this embodiment, generating structured observation vectors includes: Real-time observation data of the air-cooled island operation is collected. The real-time observation data of the air-cooled island operation includes steam side pressure, air side temperature, fan speed, and zone operation status parameters. Among them, the steam side pressure is the measured value of the steam side pressure of the air-cooled island, the air side temperature is the measured value of the air side temperature of each zone unit, the fan speed is the measured value of the operating speed of each fan unit, and the zone operation status parameters are the measured values ​​of the operation status of each zone unit. A unified time reference is established, and the sampling time of the real-time observation data of the air-cooled island is mapped to the unified time reference. For observation sequences with a sampling frequency lower than the unified time reference, linear interpolation is used to fill in the time nodes. For observation sequences with a sampling frequency higher than the unified time reference, a preservation strategy is used to perform downsampling processing to generate time-aligned real-time observation data of the air-cooled island. For the time-aligned real-time observation data of the air-cooled island, anomaly removal processing is performed. The anomaly removal processing includes boundary filtering based on a preset physical threshold range and outlier removal based on sliding window statistics, generating pre-processed real-time observation data of the air-cooled island. The preprocessed real-time observation data of the air-cooled island is arranged according to a fixed field order. The steam-side pressure is arranged in the first field position, the air-side temperature is arranged according to the observation index order corresponding to the partition unit, the fan speed is arranged according to the observation index order corresponding to the fan unit, and the partition operation status parameters are arranged according to the observation index order corresponding to the partition unit. The real-time observation data of the air-cooled island is then concatenated according to the field arrangement order to generate a structured observation vector. In the structured observation vector, a one-to-one correspondence index relationship is established between the air-side temperature field and the partition operation status parameter field, and a fixed association relationship is established between the fan speed field and the partition unit, forming a fixed observation arrangement structure to ensure that the observation input order of the structured observation vector remains consistent in the subsequent topology inversion framework.

[0023] In this embodiment, the temperature field distribution simulation calculation unit includes: Obtain the structural design data of the air-cooled island, including the number of partition units, the arrangement relationship of fan units, and the connection relationship of heat exchange units. Based on the structural design data of the air-cooled island, establish a set of partition unit numbers, a set of fan unit numbers, and a set of heat exchange unit numbers in sequence. Generate a connection relationship matrix between partition units, fan units, and heat exchange units according to the connection relationship of heat exchange units. Simultaneously generate the partition unit number order and the fan unit number order. Specifically, assign integer numbers sequentially according to the physical arrangement order of partition units in the air-cooled island structural design data to generate a set of partition unit numbers with a fixed arrangement order. Assign integer numbers sequentially according to the arrangement relationship of fan units to generate a set of fan unit numbers. Assign integer numbers sequentially according to the connection relationship of heat exchange units to generate a set of heat exchange unit numbers. Based on the set of partition unit numbers, each partition unit is divided into an independent thermal control body. Temperature state variables are established for each partition unit, pressure state variables are established based on the connection relationship of heat exchange units, and flow state variables are established based on the flow path between the fan unit and the partition unit. A set of state variables consistent with the connection relationship matrix is ​​generated. The thermodynamic operation mechanism is obtained, including the steam-side condensation heat transfer process and the air-side convection heat transfer process. Based on the air-side convection heat transfer process, a convection heat transfer relationship is established. In the convection heat transfer relationship, the heat transfer is determined by the heat transfer coefficient, heat transfer area, and the difference between the air-side temperature and the temperature state variable. Based on the steam-side condensation heat transfer process, a pressure-flow balance relationship is established between the pressure state variable and the flow state variable. In the pressure-flow balance relationship, the flow state variable is determined by the difference between the pressure state variables and the flow resistance parameter. Based on the energy conservation relationship, a coupled calculation relationship is established between the temperature state variable, the flow state variable, and the heat transfer. In the coupled calculation relationship, the change in the temperature state variable of each partition unit is determined by the heat transfer corresponding to the flow state variable and the heat capacity parameter of the partition unit. Configure parameter state vectors for the coupled calculation relationship. The parameter state vectors include heat transfer attenuation coefficient, wind-induced disturbance transmission coefficient, and partition coupling coefficient. The heat transfer attenuation coefficient acts on the heat transfer coefficient in the convective heat transfer relationship, the wind-induced disturbance transmission coefficient acts on the flow resistance parameter in the pressure-flow balance relationship, and the partition coupling coefficient acts on the heat transfer intensity term between adjacent partition units. The connection matrix, the set of state variables, the coupling calculation relationship and the parameter state vector are integrated to construct a temperature field distribution simulation calculation unit. The parameter state vector participates in the iterative update of temperature state variables, pressure state variables and flow state variables as a coefficient term in the coupling calculation relationship. The temperature field distribution simulation calculation unit performs iterative calculations when given a parameter state vector and steam-side pressure, air-side temperature, fan speed, and zoned operation state parameters as input conditions. It outputs temperature state variables that correspond one-to-one with the set of zoned unit numbers, generating the temperature field distribution.

[0024] In this embodiment, constructing the topology inversion framework includes: Based on the air-cooled island structural design data, the set of partition unit numbers, the set of fan unit numbers, and the set of heat exchange unit numbers are uniformly indexed and mapped according to the connection relationship of heat exchange units and the arrangement relationship of fan units, generating a topology index sequence consistent with the connection relationship matrix. The topology index sequence uses the partition unit number as the main index, and each partition unit corresponds to the associated fan unit index and heat exchange unit index, generating an air-cooled island physical topology structure that includes partition units, fan units, heat exchange units and connection relationships. Based on the physical topology of the air-cooled island, a topology embedding parameter space is established. In the topology embedding parameter space, the partition unit number is used as the block index. The parameter state vector is rearranged according to the partition unit number order. A parameter block structure is constructed for each partition unit. The heat transfer attenuation coefficient, wind-induced disturbance transmission coefficient, and partition coupling coefficient are arranged in a fixed order within each parameter block structure. All parameter block structures are combined according to the partition unit number order to generate a topology block parameter set. Based on the topology block parameter set, a centering process is performed on the sample dimension of the ensemble Kalman inversion algorithm. The sample covariance matrix of the topology block parameter set is calculated on the sample dimension. The row and column index order of the sample covariance matrix is ​​consistent with the partition unit numbering order of the topology block parameter set. Each covariance sub-block corresponds to the degree of correlation between the parameter block structure of two partition units. The sample covariance matrix is ​​used to characterize the parameter statistical correlation under the condition of no topology structure constraint. Based on the connection relationship matrix in the physical topology of the air-cooled island, a topological adjacency weighting matrix is ​​constructed. The row and column index order of the topological adjacency weighting matrix is ​​consistent with the partition unit numbering order of the topological block parameter set. When two partition units have a direct connection relationship in the connection relationship matrix, the corresponding element of the topological adjacency weighting matrix takes the value normalized according to the heat transfer intensity parameter. When two partition units do not have a direct connection relationship in the connection relationship matrix, the corresponding element of the topological adjacency weighting matrix takes the value of 0. The topological adjacency weighting matrix is ​​a symmetric matrix with diagonal elements taking the value of 1, which is used to characterize the parameter correlation of the partition unit itself. The topological adjacency weighting matrix and the sample covariance matrix are multiplied element-wise to generate a topological block covariance tensor. The topological block covariance tensor retains the covariance sub-blocks in the sample covariance matrix that are consistent with the connection relationship matrix according to the parameter block structure in the topological block parameter set. The parameter block structure covariance sub-blocks corresponding to partition units that do not have a connection relationship are suppressed. Thus, the physical topological structure constraint of the air-cooled island is introduced on the basis of statistical correlation. The block structure of the topological block covariance tensor is consistent with the parameter block structure in the topological block parameter set. Based on the ensemble Kalman inversion algorithm and the topology block covariance tensor, a topology inversion framework is constructed. Within this framework, a covariance structure for the update gain is defined, and the topology block covariance tensor is used as the parametric covariance structure for calculating the update gain. The update gain is calculated in the following form: ; in, Indicates the update gain. Represents the matrix form of the topological block covariance tensor. Represents the observation mapping matrix, This represents the transpose of the observation mapping matrix, which is constructed based on the field order of the structured observation vectors. The observation error covariance matrix is ​​represented by the topology block covariance tensor. By introducing the topology block covariance tensor into the definition of the updated gain structure, the topology inversion framework is constructed.

[0025] In this embodiment, generating the initial inversion temperature field distribution includes: Within the topology inversion framework, the current parameter state vector is loaded into the temperature field distribution simulation calculation unit. Under the input conditions of steam side pressure, air side temperature, fan speed, and partition operation state parameters, iterative calculation is performed to generate temperature state variables that correspond one-to-one with the partition unit number set, thus forming the temperature field distribution corresponding to the current parameter state vector. Based on the air-side temperature field in the structured observation vector, and according to the arrangement rules of the partition unit numbering order, the temperature state variables of the corresponding partition unit are extracted from the temperature field distribution corresponding to the current parameter state vector, and a predicted temperature vector with the same dimension as the air-side temperature field in the structured observation vector is generated. The difference between the predicted temperature vector and the air-side temperature field in the structured observation vector is calculated to generate the observation residual vector. Within the topology inversion framework, based on the update gain and observation residual vector, the parameter state vector is subjected to set update calculation in the sample dimension to obtain the updated parameter sample set. During the update process, the parameter dimension of the parameter sample is kept consistent with the parameter state vector, and the parameter type is kept as heat transfer attenuation coefficient, wind-induced disturbance transfer coefficient, and zonal coupling coefficient. Statistical summarization is performed on the updated parameter sample set in the sample dimension. The heat transfer attenuation coefficient, wind-induced disturbance transfer coefficient, and zonal coupling coefficient are reconstructed and arranged according to the original parameter dimension order of the parameter state vector. The statistical expectation of each parameter in the sample dimension is calculated to generate the initial updated parameter set. The initial updated parameter set is consistent with the parameter state vector in terms of parameter structure and parameter dimension. The initial set of updated parameters is loaded into the temperature field distribution simulation calculation unit. Iterative calculations are performed with steam side pressure, air side temperature, fan speed, and zoned operation status parameters as inputs. The output is the temperature state variables that correspond one-to-one with the zoned unit number set, generating the initial inverted temperature field distribution.

[0026] In this embodiment, the set of parameters for generating the frozen boundary driving mechanism includes: Based on the temperature state variables in the initial inverted temperature field distribution that correspond one-to-one with the set of partition unit numbers, a freezing critical temperature is set. The freezing critical temperature is a fixed temperature threshold. The temperature state variables of each partition unit are calculated with respect to the freezing critical temperature to obtain the temperature deviation. A freezing risk scalar field is then constructed based on the temperature deviation. Each partition unit in the freezing risk scalar field corresponds to a freezing risk scalar value. The freezing risk scalar value is obtained by mapping the temperature deviation through a monotonically nonlinear function. When the temperature state variable is lower than the freezing critical temperature, the freezing risk scalar value increases with the degree of deviation. When the temperature state variable is higher than the freezing critical temperature, the freezing risk scalar value is zero. Based on the frozen risk scalar field, spatial smoothing calculation is performed according to the connection relation matrix. The spatial smoothing calculation updates the frozen risk scalar value of each partition unit according to the partition unit number order. The neighborhood range is determined by the partition unit with a value of 1 in the corresponding row vector in the connection relation matrix. The smoothing weight is determined by the value of the corresponding element in the connection relation matrix. The frozen risk scalar value of the current partition unit and the frozen risk scalar values ​​of its neighboring partition units are weighted and summed and normalized to generate a smoothed frozen risk scalar field. The smoothed frozen risk scalar field maintains numerical continuity and an arrangement structure consistent with the partition unit number order in the topology corresponding to the partition unit number set. Based on the smoothed frozen risk scalar field, the discrete gradient of the frozen risk scalar field along the direction of the connection matrix is ​​calculated in the order of partition unit numbering, and the frozen risk gradient vector is generated. The frozen risk gradient vector is consistent with the partition unit block index of the topology block parameter set in the parameter dimension. Based on the smoothed freezing risk scalar field and freezing risk gradient vector, a non-Gaussian variable flow equation is constructed. The non-Gaussian variable flow equation defines the evolution direction of the parameter sample under continuous pseudo-time variables. The flow direction of the parameter sample is determined by the freezing risk gradient vector. The flow intensity is modulated by the freezing risk scalar value of the corresponding partition unit of the freezing risk scalar field. A nonlinear saturation function is superimposed on the flow direction to generate a non-Gaussian flow term. In the non-Gaussian variable flow equation, a topological block covariance tensor is introduced as a modulation term. The topological block covariance tensor scales and modulates the non-Gaussian flow term according to the parameter block structure in the topological block parameter set. When the corresponding element of two partition units in the topological adjacency weighting matrix has a large value, the non-Gaussian flow term between their corresponding parameter block structures is synchronously enhanced. When the corresponding element of two partition units in the topological adjacency weighting matrix has a value of 0, the non-Gaussian flow coupling between their corresponding parameter block structures does not occur, thus forming a non-Gaussian variable flow equation constrained by the physical topology of the air-cooled island. Using the initial update parameter set as the initial condition, the initial update parameter set is substituted into the non-Gaussian variable flow equation in the sample dimension. Numerical integration update is performed under the set pseudo-time step control strategy. The pseudo-time step is adaptively scaled according to the maximum freezing risk scalar value of the freezing risk scalar field to generate the parameter sample set after variable flow redistribution. Statistical summarization is performed on the parameter sample set after the variable flow redistribution process in the sample dimension. The heat transfer attenuation coefficient, wind-induced disturbance transmission coefficient, and zonal coupling coefficient are reconstructed according to the original parameter dimension order of the parameter state vector to generate the frozen boundary driving parameter set. The frozen boundary driving parameter set is consistent with the parameter state vector in terms of parameter structure and parameter dimension.

[0027] In this embodiment, generating the freezing risk gradient field includes: Based on the set of parameters driven by the frozen boundary, the heat transfer attenuation coefficient, wind-induced disturbance transmission coefficient and partition coupling coefficient in the parameter state vector are replaced and updated in the same order to obtain the updated parameter state vector. The updated parameter state vector is consistent with the original parameter state vector in terms of parameter arrangement order and parameter structure. The updated parameter state vector is loaded into the temperature field distribution simulation calculation unit. Under the input conditions of steam side pressure, air side temperature, fan speed, and zone operation state parameters, the temperature field distribution simulation calculation unit performs forward calculation step by step according to the discrete time step within the preset time domain. In each time step, the temperature state variables are iteratively updated based on the coupled calculation relationship and the updated parameter state vector, and the temperature state variables corresponding to the corresponding time and the set of zone unit numbers are output. Thus, a set of temperature state variables at multiple times is generated in the entire preset time domain for subsequent construction of temperature field distribution sequence. The preset time domain is defined as the time interval from the current time to the end of the target prediction termination time. The preset time domain is determined by the start time, the end time, and the discrete time step. According to the order of the partition unit number set, the temperature state variable set corresponding to each moment in the preset time domain is arranged in chronological order to generate a temperature field distribution sequence. The arrangement structure of the temperature state variable at each moment in the temperature field distribution sequence is consistent with the partition unit number set. Based on the temperature field distribution sequence, at each time step, the freezing risk scalar field at the corresponding time step is recalculated according to the calculation rule of the difference between the freezing critical temperature and the temperature state variable. The freezing risk scalar field is generated in the order of the partition unit number set and corresponds one-to-one with the temperature state variable at the corresponding time step. Based on the scalar field of freezing risk at each time step, spatial difference calculation is performed according to the connection relationship matrix in the order of partition unit numbering. The difference operation is performed on the scalar value of freezing risk of adjacent partition units to generate the freezing risk gradient vector at the corresponding time step. The freezing risk gradient vector is arranged in the order of partition unit numbering and is consistent with the parameter block structure in the topology block parameter set. The frozen risk gradient vectors at each time point within the preset time domain are combined in chronological order to generate a frozen risk gradient field. The frozen risk gradient field is consistent with the partition unit number set in spatial structure and consistent with the temperature field distribution sequence in temporal structure. It is used to fuse with structured observation vectors within the topology inversion framework and participate in prediction-driven inversion update calculations.

[0028] In this embodiment, generating a dual-loop update parameter set and calculating the convergence criterion value includes: Within the topology inversion framework, the structured observation vector and the frozen risk gradient field are spliced ​​and fused according to the field dimension order. The structured observation vector retains the original field arrangement structure, while the frozen risk gradient field is expanded into a vector structure consistent with the partition block index of the topology block parameter set according to the partition unit number set order. The two vectors are extended and combined in the observation dimension to generate an extended observation vector. The extended observation vector retains the structured observation vector field at the front and adds the frozen risk gradient field field at the back to form a fixed extended observation arrangement structure. Based on the extended observation vector, prediction-driven inversion update calculation is performed within the topology inversion framework. During the update gain calculation, the observation mapping matrix is ​​extended to include the mapping relationship of the frozen risk gradient field. The front of the extended observation mapping matrix corresponds to the mapping relationship of the structured observation vector, and the back corresponds to the mapping relationship of the frozen risk gradient field in the parameter block structure. The update gain is calculated based on the topology block covariance tensor. The set update calculation is performed on the parameter state vector in the sample dimension to generate the prediction-driven update parameter set. The prediction-driven update parameter set is consistent with the parameter state vector in terms of parameter structure and parameter dimension. Based on the prediction-driven parameter set update, a centralization process is performed on the sample dimension. The predicted sample covariance matrix is ​​calculated according to the partitioning and block index order of the topological block parameter set. The row and column index order of the predicted sample covariance matrix is ​​consistent with that of the topological block parameter set. The predicted sample covariance matrix reflects the statistical correlation of the parameter sample distribution under the extended observation vector constraint. Compared with the initial sample covariance matrix, the predicted sample covariance matrix produces a covariance contraction effect in the direction of the partition unit corresponding to the frozen risk gradient field. The topological block covariance tensor is reconstructed by performing element-wise multiplication operations on the predicted sample covariance moment matrix and the topological adjacency weighting matrix. During the reconstruction process, the corresponding elements of the topological adjacency weighting matrix are proportionally modulated according to the gradient magnitude of each partition unit in the frozen risk gradient field. When the gradient magnitude of the frozen risk gradient field is large at a certain partition unit, the weighting coefficient between the corresponding parameter block structure of that partition unit and the adjacent partition units is increased. When the gradient magnitude of the frozen risk gradient field is zero at a certain partition unit, the weights of the original topological adjacency weighting matrix remain unchanged, thereby generating a reconstructed topological block covariance tensor that is linked to the frozen risk gradient field. Based on the reconstructed topological block covariance tensor, the non-Gaussian variable flow equation is invoked again within the topology inversion framework. The prediction-driven update parameter set is used as the initial flow condition in the sample dimension. The frozen risk scalar value of the partition unit corresponding to the frozen risk scalar field is used as the intensity modulation term of the non-Gaussian variable flow equation. The reconstructed topological block covariance tensor is used as the coupling modulation term between parameter block structures. Numerical integration update is performed under continuous pseudo-time variables to generate the parameter sample set after risk-driven redistribution. Statistical summarization processing is performed on the parameter sample set after risk-driven redistribution in the sample dimension. The heat transfer attenuation coefficient, wind-induced disturbance transmission coefficient, and zonal coupling coefficient are reconstructed according to the original parameter dimension order of the parameter state vector to generate a double-closed-loop updated parameter set. The first closed loop is a prediction-driven inversion update based on the extended observation vector, and the second closed loop is a non-Gaussian variable flow redistribution update based on the joint constraints of the frozen risk scalar field and the reconstructed topology block covariance tensor. Based on the double-loop update parameter set, the parameter difference norm of the parameter sample set in two consecutive double-loop update processes is calculated. The parameter difference norm is compared with the convergence judgment threshold to generate a convergence judgment value. The convergence judgment value is used to determine whether the double-loop update parameter set has reached a stable state.

[0029] In this embodiment, the calculation of the freezing risk distribution map and the antifreeze control reference quantity includes: Under the condition that the convergence judgment value meets the convergence judgment threshold, the double closed-loop update parameter set is loaded into the temperature field distribution simulation calculation unit according to the original parameter dimension order of the parameter state vector. Iterative calculation is performed with steam side pressure, air side temperature, fan speed, and partition operation state parameters as input. The output is the temperature state variable corresponding to the partition unit number set one by one, generating the converged inversion temperature field distribution. The converged inversion temperature field distribution is arranged in the order of the partition unit number set. Based on the convergent inversion of the temperature field distribution, and according to the calculation rules for the difference between the freezing critical temperature and the temperature state variable, the deviation of the temperature state variable corresponding to each partition unit is calculated. A freezing risk scalar field is generated in the order of the partition unit number set. The freezing risk scalar field is mapped into a two-dimensional matrix form according to the spatial arrangement relationship of the partition units in the physical topology of the air-cooled island, and a freezing risk distribution map is generated. Each matrix unit in the freezing risk distribution map corresponds to the freezing risk scalar value of a partition unit. Based on the freezing risk distribution map, according to the correspondence between the freezing risk scalar value and the preset risk level range, each partition unit is divided into risk levels. When the freezing risk scalar value is greater than the first risk threshold, the corresponding partition unit is marked as a high-risk partition unit. When the freezing risk scalar value is between the second risk threshold and the first risk threshold, the corresponding partition unit is marked as a medium-risk partition unit. When the freezing risk scalar value is less than the second risk threshold, the corresponding partition unit is marked as a low-risk partition unit, thus generating a set of partition risk levels. Based on the risk level set of the partitions and the connection relationship between partition units and fan units in the physical topology of the air-cooled island, an initial antifreeze control reference quantity is generated according to the control rules corresponding to the preset risk level. The initial antifreeze control reference quantity is arranged in the order of the partition unit number set and the fan unit number set. The initial antifreeze control reference quantity includes the partition operation status parameter adjustment quantity and the fan speed adjustment quantity. Among them, the partition unit with a freezing risk scalar value higher than the high risk threshold corresponds to the increase of the fan speed adjustment quantity of the associated fan unit. The partition unit with a freezing risk scalar value between the medium risk threshold and the high risk threshold corresponds to the fan speed adjustment quantity that maintains the current fan speed. The partition unit with a freezing risk scalar value lower than the medium risk threshold corresponds to the fan speed adjustment quantity that decreases the fan speed of the associated fan unit. Based on the spatial difference direction of the scalar value of freezing risk between each partition unit in the freezing risk distribution map, and combined with the connection relationship matrix, the propagation direction of freezing risk in the physical topology of the air-cooled island is determined. Under the sequence of the partition unit number set, the fan speed adjustment amount in the initial antifreeze control reference amount is corrected by gradient direction. When the freezing risk shows an increasing trend along the direction of the connection relationship matrix, the fan speed adjustment amount of the corresponding fan unit is increased by the correction increment. When the freezing risk shows a decreasing trend along the direction of the connection relationship matrix, the fan speed adjustment amount of the corresponding fan unit is decreased by the correction increment. After correction, the antifreeze control reference amount is generated. The antifreeze control reference quantity is output according to the arrangement structure of the partition unit number set and the fan unit number set, which is used by the air-cooled island operation control system to perform partition operation status parameter adjustment and fan speed adjustment.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a low-temperature winter operation scenario of a direct air-cooled system in a coal-fired power generating unit. This unit adopts a multi-zone air-cooled island structure, with the air-cooled island consisting of 24 zone units, each equipped with an independent fan unit and heat exchange unit. During winter operation, the ambient air temperature dropped to a minimum of -18℃, the steam pressure fluctuated between 8.5 kPa and 12.3 kPa, and the fan speed control range was 420 rpm to 980 rpm. Historical operating records show that under continuous low-temperature and load fluctuation conditions, the temperature state variable of the finned tube bundle in some zone units dropped to 1.2℃, approaching the freezing critical temperature of 0℃, posing a freezing risk. Operators mainly relied on experience to adjust the fan speed and zone operating parameters, resulting in insufficient control precision and high energy consumption.

[0031] In this scenario, real-time observation data of the air-cooled island operation, including steam-side pressure, air-side temperature, fan speed, and zoned operating status parameters, are integrated to generate structured observation vectors. A temperature field distribution simulation calculation unit is constructed based on the air-cooled island structural design data and thermodynamic operating mechanism. The physical topology of the air-cooled island and its embedded parameter space are established, forming a topological block covariance tensor and a topological inversion framework. During a continuous 30-day low-temperature operation cycle, a set update calculation is performed every 5 minutes to generate an initial inverted temperature field distribution. A freezing risk scalar field is constructed based on the initial inverted temperature field distribution, and a variable flow redistribution process is performed using a non-Gaussian variable flow equation to generate a set of freezing boundary driving parameters. Subsequently, a forward calculation is performed to generate a temperature field distribution sequence and calculate the freezing risk gradient field. The structured observation vectors and the freezing risk gradient field are fused to form a double-closed-loop update parameter set. Under the condition that the convergence criteria value meets the convergence threshold, a converged inverted temperature field distribution and freezing risk distribution map are generated, outputting anti-freezing control reference quantities to guide the adjustment of fan speed and zoned operating status parameters.

[0032] During the verification period, the method of this invention was compared with traditional empirical adjustment methods and conventional mechanism simulation methods. Indicators such as freezing risk warning accuracy, mean square error of the temperature field, average fan power, number of freezing risk exceedances, and number of parameter convergence iterations were collected. Experimental data are shown in the table below.

[0033] Table 1 Comparison of Experimental Data on Anti-freezing Performance of Air-cooled Island

[0034] Based on the data analysis in Table 1, it can be seen that the accuracy of the freezing risk warning improved from 82.4% using the traditional empirical adjustment method to 90.2%, an increase of 7.8 percentage points; the mean square error of the temperature field decreased from 2.85℃ to 1.21℃, a reduction of approximately 57.5%, indicating that under the constraints of the topology inversion framework and the topology block covariance tensor, the parameter state vector update process is more consistent with the physical topology of the air-cooled island; the average fan power decreased from 4120kW to 3840kW, a reduction of approximately 6.8%, indicating that the antifreeze control reference quantity formed under the driving force of the freezing risk scalar field and the freezing risk gradient field can reduce the increase in energy consumption caused by over-adjustment.

[0035] The number of freezing risk exceedances decreased from 9 to 2, and the average deviation of high-risk zone identification decreased from 2.1℃ to 0.9℃. This indicates that after coupling the non-Gaussian variable flow equation with the reconstructed topology block covariance tensor, the redistribution of parameter samples in the risk region direction is more concentrated, enhancing the ability to characterize the low-temperature critical region. The number of parameter convergence iterations decreased from 18 to 11, indicating that the double-closed-loop updated parameter set converges faster under the dual constraints of observation-driven and risk-driven approaches. The fluctuation range of the wind turbine speed adjustment decreased from 180 rpm to 85 rpm, indicating that the wind turbine adjustment process is more stable after the freezing risk gradient field participates in the generation of control variables.

[0036] Under continuous low-temperature conditions, the lowest temperature reached by traditional methods drops to -0.8℃, resulting in freezing; the lowest temperature reached by conventional mechanism simulation methods is 0.6℃; the lowest temperature reached by the method of this invention is 1.3℃, remaining above the freezing critical temperature, and no freezing was recorded. This invention achieves reasonable energy consumption and high prediction accuracy while maintaining operational safety margin, demonstrating the practical engineering value of topological constraints and risk-driven redistribution mechanisms in intelligent simulation of air-cooled island antifreeze.

[0037] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A digital twin-based intelligent simulation method for antifreeze protection of air-cooled islands, characterized in that, Includes the following steps: Acquire real-time observation data of the air-cooled island operation, perform time alignment processing and anomaly removal processing, and splice the data in a fixed order to generate structured observation vectors; Obtain structural design data and thermodynamic operation mechanism of the air-cooled island, and construct a temperature field distribution simulation calculation unit; Based on the design data of the air-cooled island structure, the physical topology and topology embedding parameter space of the air-cooled island are constructed, the topology block covariance tensor is generated, and the topology inversion framework is constructed by combining the ensemble Kalman inversion algorithm. Within the topology inversion framework, based on the structured observation vectors, an ensemble update calculation is performed to generate an initial update parameter set, which is then loaded into the temperature field distribution simulation calculation unit to generate the initial inverted temperature field distribution. Based on the initial inverted temperature field distribution, a scalar field of freezing risk and a non-Gaussian variable flow equation are constructed, and variable flow redistribution processing is performed to generate a set of freezing boundary driving parameters. The parameter state vector is updated based on the parameter set driven by the frozen boundary, the forward calculation is performed to generate the temperature field distribution sequence, and the spatial gradient of the freezing risk scalar field is calculated to generate the freezing risk gradient field. Based on the structured observation vector and the frozen risk gradient field, the extended observation vector is generated by fusion, the prediction-driven inversion update calculation is performed to generate the prediction-driven update parameter set, the topological block covariance tensor is reconstructed, the variable flow redistribution processing is performed to generate the double closed-loop update parameter set, and the convergence judgment value is calculated. When the convergence judgment value meets the convergence judgment threshold, the double closed-loop update parameter set is loaded into the temperature field distribution simulation calculation unit to generate the converged inversion temperature field distribution, and the freezing risk distribution map and antifreeze control reference quantity are calculated.

2. The intelligent simulation method for antifreeze of air-cooled islands based on digital twins according to claim 1, characterized in that, The generated structured observation vectors include: Collect real-time observation data of the air-cooled island operation, including steam side pressure, air side temperature, fan speed, and zone operation status parameters; Establish a unified time reference and map the sampling time of the real-time observation data of the air-cooled island operation to the unified time reference to generate time-aligned real-time observation data of the air-cooled island operation. Anomaly removal processing is performed on the time-aligned real-time observation data of the air-cooled island to generate pre-processed real-time observation data of the air-cooled island. The preprocessed real-time observation data of the air-cooled island are arranged according to a fixed field order, and the fields are concatenated in the execution order to generate a structured observation vector.

3. The intelligent simulation method for anti-freezing of air-cooled islands based on digital twins according to claim 1, characterized in that, The temperature field distribution simulation calculation unit includes: Obtain the structural design data of the air-cooled island, including the number of partition units, the arrangement relationship of fan units, and the connection relationship of heat exchange units. Sequentially establish the partition unit number set, fan unit number set, and heat exchange unit number set. Generate a connection relationship matrix according to the connection relationship of heat exchange units, and simultaneously generate the partition unit number order and fan unit number order. Each partition unit is divided into an independent thermal control body based on the set of partition unit numbers. Temperature state variables are established for each partition unit, pressure state variables are established based on the connection relationship of heat exchange units, and flow state variables are established based on the flow path between the fan unit and the partition unit, thus generating a set of state variables. The thermodynamic operation mechanism is obtained, including the steam-side condensation heat transfer process and the air-side convection heat transfer process. Convection heat transfer relationship is established based on the air-side convection heat transfer process. Pressure-flow balance relationship between pressure state variables and flow state variables is established based on the steam-side condensation heat transfer process. Coupled calculation relationship between temperature state variables, flow state variables and heat transfer is established based on the energy conservation relationship. Configure parameter state vectors for the coupled calculation relationship, including heat transfer attenuation coefficient, wind-induced disturbance transmission coefficient, and zonal coupling coefficient; By integrating the connection matrix, the set of state variables, the coupling calculation relationship, and the parameter state vector, a temperature field distribution simulation calculation unit is constructed.

4. The intelligent simulation method for antifreeze of air-cooled islands based on digital twins according to claim 1, characterized in that, The construction of the topology inversion framework includes: Based on the air-cooled island structural design data, the set of partition unit numbers, the set of fan unit numbers, and the set of heat exchange unit numbers are uniformly indexed and mapped according to the connection relationship of heat exchange units and the arrangement relationship of fan units, generating a topology index sequence and generating an air-cooled island physical topology structure containing partition units, fan units, heat exchange units and connection relationships. Based on the physical topology of the air-cooled island, a topology embedding parameter space is established. Using the partition unit number as the block index, the parameter state vector is rearranged according to the partition unit number order. A parameter block structure is constructed for each partition unit. All parameter block structures are combined to generate a topology block parameter set. Based on the topological block parameter set, a centering process is performed on the sample dimension of the ensemble Kalman inversion algorithm, and the sample covariance matrix of the topological block parameter set is calculated on the sample dimension. Based on the connection relationship matrix in the physical topology of the air-cooled island, a topological adjacency weighting matrix is ​​constructed. Perform element-wise multiplication between the topological adjacency weighting matrix and the sample covariance matrix to generate the topological block covariance tensor; Based on the ensemble Kalman inversion algorithm and the topology block covariance tensor, a topology inversion framework is constructed. In the topology inversion framework, the covariance structure of the update gain is defined, and the topology block covariance tensor is used as the parameter covariance structure for the calculation of the update gain.

5. The intelligent simulation method for anti-freezing of air-cooled islands based on digital twins according to claim 1, characterized in that, The generation of the initial inversion temperature field distribution includes: Within the topology inversion framework, the current parameter state vector is loaded into the temperature field distribution simulation calculation unit, iterative calculation is performed, temperature state variables are generated, and the temperature field distribution corresponding to the current parameter state vector is formed. Based on the structured observation vector, the temperature state variable is extracted from the temperature field distribution corresponding to the current parameter state vector to generate the predicted temperature vector, and the difference between the predicted temperature vector and the structured observation vector is calculated to generate the observation residual vector. Within the topology inversion framework, based on the update gain and observation residual vector, a set update calculation is performed on the parameter state vector in the sample dimension to obtain the updated parameter sample set. Perform statistical summarization processing on the sample dimension of the updated parameter sample set, reconstruct and arrange it according to the original parameter dimension order of the parameter state vector, calculate the statistical expectation of each parameter on the sample dimension, and generate the initial updated parameter set. The initial set of updated parameters is loaded into the temperature field distribution simulation calculation unit, iterative calculation is performed, and the temperature state variables corresponding one-to-one with the set of partition unit numbers are output to generate the initial inverted temperature field distribution.

6. The intelligent simulation method for antifreeze of air-cooled islands based on digital twins according to claim 1, characterized in that, The set of parameters for generating the frozen boundary includes: Based on the initial inverted temperature field distribution, a freezing critical temperature is set, and the temperature state variables of each partition unit are calculated with respect to the freezing critical temperature to obtain the temperature deviation. A freezing risk scalar field is then constructed based on the temperature deviation, with each partition unit in the freezing risk scalar field corresponding to a freezing risk scalar value. Based on the frozen risk scalar field, spatial smoothing calculation is performed according to the connection relationship matrix. The frozen risk scalar value of each partition unit is updated with neighborhood weighting to generate a smoothed frozen risk scalar field. Based on the smoothed freezing risk scalar field, the discrete gradient of the freezing risk scalar field along the direction of the connectivity matrix is ​​calculated to generate the freezing risk gradient vector. Based on the smoothed freezing risk scalar field and freezing risk gradient vector, a non-Gaussian variable flow equation is constructed to generate a non-Gaussian flow term. In the non-Gaussian variable flow equation, the topological block covariance tensor is introduced as a modulation term to scale and modulate the non-Gaussian flow term. Using the initial update parameter set as the initial condition, the non-Gaussian variable flow equation is substituted into it, and numerical integration update is performed under the set pseudo-time step control strategy to generate a parameter sample set after variable flow redistribution processing. Statistical summarization is performed on the parameter sample set after variable flow redistribution processing along the sample dimension, and the frozen boundary driving parameter set is reconstructed according to the original parameter dimension order of the parameter state vector.

7. The intelligent simulation method for antifreeze of air-cooled islands based on digital twins according to claim 1, characterized in that, The generation of the freezing risk gradient field includes: Based on the frozen boundary driving parameter set, the coefficients in the parameter state vector are replaced and updated in the same order to obtain the updated parameter state vector; The updated parameter state vector is loaded into the temperature field distribution simulation calculation unit. Forward calculation is performed step by step according to the discrete time step within the preset time domain. The temperature state variables are iteratively updated in each time step, and a set of temperature state variables at multiple time steps is generated in the entire preset time domain. According to the order of the partition unit number set, the temperature state variable set corresponding to each moment in the preset time domain is arranged in chronological order to generate a temperature field distribution sequence; Based on the temperature field distribution sequence, the freezing risk scalar field at each time step is recalculated. Based on the scalar field of freezing risk at each time step, spatial difference calculation is performed according to the connection matrix to generate the freezing risk gradient vector at the corresponding time step. The freezing risk gradient vectors at each time point within the preset time domain are combined in chronological order to generate the freezing risk gradient field.

8. The intelligent simulation method for antifreeze of air-cooled islands based on digital twins according to claim 1, characterized in that, The generation of the dual-loop update parameter set and the calculation of the convergence determination value include: Within the topology inversion framework, the structured observation vector and the frozen risk gradient field are spliced ​​and fused in the order of field dimensions, and the two vectors are extended and combined in the observation dimension to generate an extended observation vector. Based on the extended observation vector, prediction-driven inversion update calculation is performed within the topology inversion framework. The update gain is calculated based on the topology block covariance tensor. The set update calculation is performed on the parameter state vector in the sample dimension to generate the prediction-driven update parameter set. Based on the prediction-driven parameter set update, a centralization process is performed on the sample dimension to calculate the predicted sample covariance matrix; Element-wise multiplication is performed on the predicted sample covariance moment matrix and the topological adjacency weighted matrix to reconstruct the topological block covariance tensor, generating a reconstructed topological block covariance tensor that is linked to the frozen risk gradient field. Based on the reconstructed topological block covariance tensor, the non-Gaussian variable flow equation is called again within the topological inversion framework to perform numerical integration update and generate a parameter sample set after risk-driven redistribution. Perform statistical summarization processing on the sample dimension of the parameter sample set after risk-driven redistribution to generate a double closed-loop updated parameter set; Based on the double-loop update parameter set, the parameter difference norm of the parameter sample set in two consecutive double-loop update processes is calculated and compared with the convergence judgment threshold to generate a convergence judgment value.

9. The intelligent simulation method for anti-freezing of air-cooled islands based on digital twins according to claim 1, characterized in that, The calculation of the freezing risk distribution map and the antifreeze control reference values ​​include: Under the condition that the convergence judgment value meets the convergence judgment threshold, the double closed-loop update parameter set is loaded into the temperature field distribution simulation calculation unit in the order of the original parameter dimensions of the parameter state vector, and iterative calculation is performed to generate the converged inversion temperature field distribution. Based on the convergent inversion of the temperature field distribution, the deviation of the temperature state variables corresponding to each partition unit is calculated to generate a freezing risk scalar field. The freezing risk scalar field is then mapped into a two-dimensional matrix to generate a freezing risk distribution map. Based on the risk distribution map of freezing, risk levels are divided into each partition unit to generate a set of partition risk levels; Based on the risk level set of the partition and the physical topology of the air-cooled island, an initial antifreeze control reference quantity is generated according to the preset risk level corresponding control rules. Based on the spatial difference direction of the freezing risk scalar value between each partition unit in the freezing risk distribution map, the propagation direction of freezing risk in the physical topology of the air-cooled island is determined. Gradient direction correction is performed on the fan speed adjustment in the initial antifreeze control reference quantity to generate the antifreeze control reference quantity.

Images