CN-121980944-A - Critical heat flow density prediction system and method based on mechanism-data fusion

CN121980944ACN 121980944 ACN121980944 ACN 121980944ACN-121980944-A

Abstract

The invention discloses a critical heat flow density prediction system and a method based on mechanism-data fusion, wherein the method firstly obtains multidimensional feature vectors of a narrow rectangular runner to be predicted, and the multidimensional feature vectors comprise geometric structures, operation conditions, local thermal engineering and geometric deformation feature parameters; a parallel fusion architecture is then constructed that includes a mechanism prediction channel and a machine learning compensation channel. The mechanism channel solves a mass conservation equation based on a liquid film drying physical model to output a reference value, the machine learning channel calculates residual errors between the mechanism model and a true value by utilizing XGBoost algorithm and the like to output a compensation value, and finally a final predicted value is obtained through linear superposition. The method effectively solves the problem that the critical heat flow density under the fuel plate shape forming working condition cannot be accurately predicted by the traditional method by introducing the geometric deformation parameters and utilizing the data driving model to compensate systematic deviation of the mechanism model, remarkably improves the prediction precision and physical consistency, and is suitable for safety evaluation of an advanced nuclear reactor.

Inventors

LIU HAIDONG
JIANG HONGFU
LIU DINGKAI
HE JIANG
WANG YANGYANG
LIU MINGXIA

Assignees

重庆理工大学

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. The critical heat flow density prediction method based on mechanism-data fusion is characterized by comprising the following steps of: S1, acquiring a multidimensional feature vector of a narrow rectangular runner to be predicted, wherein the feature vector at least comprises a geometric structure parameter, a system operation condition parameter, a local thermal parameter and a geometric deformation feature parameter; s2, constructing a parallel fusion prediction framework, wherein the framework comprises a mechanism prediction channel and a machine learning compensation channel which are mutually independent; s3, inputting the characteristic vector into the mechanism prediction channel, solving a liquid film mass conservation equation based on a liquid film drying physical model, and outputting a critical heat flow density reference value; S4, inputting the feature vector into the machine learning compensation channel, calculating residual errors between a mechanism model and a true value by using a pre-trained machine learning model, and outputting a critical heat flow density compensation value; And S5, linearly superposing the reference value and the compensation value by adopting an addition and fusion strategy to obtain a final critical heat flow density predicted value.
2. The method for predicting critical heat flux density based on mechanism-data fusion as set forth in claim 1, wherein the geometric deformation characteristic parameter in the step S1 is used for describing non-uniform deformation of the fuel plate under irradiation or thermal load, and specifically comprises deformation area, deformation height ratio and deformation type; the geometric structure parameters in the step S1 comprise runner gap size, runner width and heating length; the system operation condition parameters in the step S1 comprise system pressure, mass flow rate and inlet supercooling degree; the local thermal parameters in the step S1 include a local heat balance air content and a power peak factor.
3. The method for predicting critical heat flux density based on mechanism-data fusion as set forth in claim 1, wherein the equation of conservation of mass of liquid film in the mechanism predicting channel in step S2 is defined as: And carrying out integral solution on the mass flow rate m lf of the liquid film along the axial direction of the flow channel: Wherein z is an axial coordinate, P total is the total geometric perimeter of the flow passage section, D is the droplet deposition rate, E is the droplet entrainment rate, q' is the heat flux density, P heat is the actual heated perimeter of the flow passage section, and h fg is the latent heat of vaporization; The equation characterizes the lateral supplementary effect of the non-heating side wall of the narrow rectangular flow channel on the liquid film by introducing the difference between P total and P heat .
4. The method for predicting critical heat flux density based on mechanism-data fusion as set forth in claim 1, wherein in the step S2, a machine learning compensation channel is constructed by XGBoost algorithm, and an objective function of the machine learning compensation channel comprises a loss function and a regularization term; and in the model training process, performing second-order Taylor expansion on the loss function, and guiding the growth of a tree structure by using a step degree statistic and a second-order gradient statistic so as to minimize the mechanism model prediction residual error.
5. The critical heat flux density prediction method based on mechanism-data fusion as set forth in claim 1, wherein the super parameters of the machine learning model are globally optimized by a Bayesian optimization algorithm, and the Bayesian optimization algorithm performs iterative search on a super parameter space including a learning rate, a maximum depth of a tree, a subsampled sampling rate and a regularization coefficient with a root mean square error minimization on a cross validation set as a target based on a TPE process.
6. The method for predicting critical heat flux density based on mechanism-data fusion as set forth in claim 1, further comprising an interpretation analysis step of the predicted result: And integrating a SHAP interpreter, calculating marginal contribution of the geometric deformation characteristic parameter to the critical heat flow density compensation value based on a game theory shape value, and quantifying a specific influence value of deformation to the critical heat flow density.
7. The critical heat flow density prediction method based on mechanism-data fusion according to claim 1, wherein training data of a machine learning model in the step S2 is derived from a mixed high-fidelity database, the database comprises experimental data measured by a narrow rectangular runner thermal hydraulic experimental loop and supplementary data generated by two-phase flow computational fluid mechanics simulation under extremely high pressure or extremely low flow speed working conditions, and before training, a Z-Score standardization method is adopted to carry out dimensionless treatment on the feature vectors.
8. A critical heat flux density prediction system based on mechanism-data fusion, comprising: The multi-source sensing module is used for collecting geometric structure parameters and operation condition data of the narrow rectangular flow channel and obtaining geometric deformation characteristic parameters of the fuel plate through on-line monitoring or soft measurement; The mechanism prediction module is used for running a liquid film drying physical model, solving a liquid film mass conservation equation considering the narrow slit cold wall effect through numerical integration, and outputting a critical heat flow density reference value; The intelligent compensation module is embedded with a packaged XGBoost inference engine and is used for calculating residual error compensation values relative to the mechanism reference value in parallel according to the input characteristic parameters; And the fusion output module is used for executing addition fusion operation, adding the reference value and the compensation value, executing physical boundary inspection and outputting a final critical heat flow density predicted value.
9. The critical heat flux density prediction system based on mechanism-data fusion as set forth in claim 8, wherein said system is trained and packaged as a dynamic link library or API interface in an offline phase, and real-time prediction is implemented in an online phase and integrated into a nuclear reactor safety analysis program or digital control system; The boundary checking module is used for checking the physical rationality of the final predicted value; The machine learning compensation module integrates a SHAP interpreter for providing feature importance and contribution analysis.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any one of claims 1 to 7.

Description

Critical heat flow density prediction system and method based on mechanism-data fusion Technical Field The invention belongs to the technical field of nuclear reactor thermal safety, and particularly relates to a critical heat flow density prediction system and method based on mechanism-data fusion. Background In the design of many advanced nuclear power systems, fuel assemblies composed of parallel fuel plates are commonly employed for the purpose of achieving high power density and core miniaturization. The coolant channels formed between these fuel plates are typically narrow rectangular flow channels, see fig. 1. This particular geometry (typically with a channel gap of less than 3mm and a large aspect ratio) results in a complex boiling two-phase flow phenomenon inside it, which is significantly different from conventional circular pipes. In particular, the vapor bubbles in the flow channels are strongly limited by the wall surfaces and are in a flattened hat-shaped or elongated bullet-shaped form, so that the inter-phase drag force and the heat transfer area are greatly changed, and meanwhile, the fluid dynamics of the four corner areas of the flow channels can lead to non-uniform distribution and local stagnation of a liquid film, so that the stability of boiling heat transfer is a serious challenge. Currently, CHF predictions for such flow channels rely primarily on empirical relationships (e.g., bowing, sudo formulas, etc.) established based on specific experimental data or look-up tables (e.g., groeneveld CHF Look-up tables) based on a generic fluid model, see fig. 2. However, these methods exhibit significant limitations in applicability and accuracy in the face of different geometries, heating conditions (e.g., single-sided heating versus double-sided heating), and non-uniform power distribution. These conventional methods fail to capture the resulting local flow rate and enthalpy changes, resulting in significant prediction errors, particularly when the fuel plate is geometrically deformed under irradiation and thermal hydraulic loading. The emerging pure machine learning method in recent years shows strong data fitting capability, but the 'black box' nature of the method makes the prediction result lack of necessary physical constraint, and a conclusion against the physical rule can be output when the method faces working conditions outside the training set, which is unacceptable in the field of nuclear safety with high reliability requirements. Therefore, developing a new method capable of accurately and reliably predicting narrow rectangular-runner CHF has become an urgent need for advanced reactor technology development. Disclosure of Invention In order to make up for the defects of the prior art, a critical heat flow density prediction system and a method based on mechanism-data fusion are provided. A critical heat flow density prediction method based on mechanism-data fusion comprises the following steps: S1, acquiring a multidimensional feature vector of a narrow rectangular runner to be predicted, wherein the feature vector at least comprises a geometric structure parameter, a system operation condition parameter, a local thermal parameter and a geometric deformation feature parameter; s2, constructing a parallel fusion prediction framework, wherein the framework comprises a mechanism prediction channel and a machine learning compensation channel which are mutually independent; s3, inputting the characteristic vector into the mechanism prediction channel, solving a liquid film mass conservation equation based on a liquid film drying physical model, and outputting a critical heat flow density reference value; S4, inputting the feature vector into the machine learning compensation channel, calculating residual errors between a mechanism model and a true value by using a pre-trained machine learning model, and outputting a critical heat flow density compensation value; And S5, linearly superposing the reference value and the compensation value by adopting an addition and fusion strategy to obtain a final critical heat flow density predicted value. Further, the geometric deformation characteristic parameter is used for describing the non-uniform deformation of the fuel plate under irradiation or thermal load, and specifically comprises a deformation area, a deformation height ratio and a deformation type; the geometric structure parameters comprise a runner gap size, a runner width and a heating length; the system operation condition parameters comprise system pressure, mass flow rate and inlet supercooling degree; the local thermal parameters include local heat balance air content and power peak factor. Further, the mechanism predicts the liquid film mass conservation equation in the channel to be defined as: And carrying out integral solution on the mass flow rate m lf of the liquid film along the axial direction of the flow channel: Wherein z is an axial coordinate, P total is the total