A cylinder head fire face heat exchange coefficient calibration method based on a multi-constraint optimization algorithm

Abstract

The application discloses a cylinder head fire surface heat exchange coefficient calibration method based on a multi-constraint optimization algorithm, and relates to the technical field of computer simulation. The method comprises the following steps: acquiring partition information, measured temperature, temperature measuring position and heat exchange constraint reference; determining partition radial characteristics, establishing continuous heat exchange distribution and calibration parameters; generating samples by disturbing parameters, obtaining temperature and constraint state through thermal simulation; inputting an improved Cubist model to generate temperature mapping rules and boundary markers; calculating predicted temperature by the mapping rules and forming radial residual attribution; calculating three types of heat exchange constraint margins to form a constraint set; adopting a COBYLA algorithm with constraint margin retraction and residual partition scaling to optimize parameters; forming a target heat exchange distribution, and outputting calibration results through thermal simulation verification. The application improves the accuracy, physical constraint consistency and parameter optimization stability of the cylinder head fire surface heat exchange coefficient calibration through collaborative optimization of the improved Cubist model and the COBYLA algorithm.

Description

Technical Field

[0001] This invention relates to the field of computer simulation technology, and in particular to a method for calibrating the heat transfer coefficient of the cylinder head fire surface based on a multi-constraint optimization algorithm. Background Technology

[0002] As diesel engines develop towards higher power density and enhanced performance, the thermal load on the cylinder head's firing face is continuously increasing. Its temperature field distribution directly affects the cylinder head's thermal stress, thermal fatigue life, and overall engine reliability. In cylinder head thermal analysis, the heat transfer coefficient of the firing face is a crucial parameter for determining the thermal boundary conditions on the combustion side, and its accuracy directly impacts the thermal simulation results. Current techniques typically divide the firing face into regions based on experimental experience, empirical correlations, or existing engine data, assigning corresponding heat transfer coefficients to different regions. Then, finite element thermal simulation is used to obtain the cylinder head temperature field, and the heat transfer coefficient is adjusted based on the deviation between the simulated and measured temperatures to calibrate the thermal boundary parameters of the firing face.

[0003] Existing heat transfer coefficient calibration methods often use discrete heat transfer coefficients corresponding to different fire surface zones as the adjustment objects. Since the parameters of each zone are relatively independent, they fail to fully reflect the distribution characteristics of the heat transfer coefficient as it continuously changes radially from the center of the injector to the periphery of the fire surface. Abrupt changes in heat transfer parameters are prone to occur at the boundaries between adjacent zones. Furthermore, some methods primarily use the deviation between simulated and measured temperatures as the basis for parameter adjustment, failing to adequately consider the relative heat transfer strength between different regions, the overall heat transfer energy of the fire surface, and the radial heat transfer variation. This means that while some parameter combinations may achieve small temperature deviations at measurement points, the corresponding overall heat transfer state or radial heat transfer distribution may deviate from actual physical laws, affecting the rationality of the heat transfer coefficient calibration results.

[0004] In addition, the existing calibration process usually requires repeated cylinder head thermal simulation for different candidate heat transfer parameters and repeated adjustment of parameters based on simulation results. When the number of parameters to be calibrated increases or multiple physical constraints exist at the same time, the coupling relationship between parameters is enhanced, which can easily lead to problems such as a large iterative search range, unstable parameter adjustment near the constraint boundary, and increased computational load of repeated simulations. It is difficult to take into account the temperature matching accuracy, heat transfer distribution continuity and multiple physical constraint requirements.

[0005] Therefore, how to provide a method for calibrating the heat transfer coefficient of the cylinder head fire surface based on a multi-constraint optimization algorithm 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 a method for calibrating the heat transfer coefficient of the cylinder head fire surface based on a multi-constraint optimization algorithm. This invention fully utilizes thermal simulation, an improved Cubist model, and the COBYLA algorithm, and describes in detail a method for continuous modeling and joint iterative calibration of the heat transfer coefficient based on the radial position of the fire surface, regional heat transfer relationship, energy conservation relationship, radial variation law, and measured temperature. It has the advantages of good continuity of heat transfer coefficient distribution, high physical constraint compliance, and high calibration accuracy.

[0007] A method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to an embodiment of the present invention includes: Obtain the geometric information of the firepower surface zones, zone area, injector center position, measured temperature, temperature measurement location, and heat transfer constraint benchmark; Determine the radial range and equivalent radial position of each fire zone, and establish the continuous distribution relationship of heat transfer coefficients and parameters to be calibrated; The parameters to be calibrated are disturbed to generate parameter samples, the corresponding heat transfer distribution is loaded into the thermal simulation model to obtain the simulation temperature, and the constraint state is generated according to the constraint benchmark. The Cubist model is improved by inputting parameter samples, constraint states and simulation temperatures to generate temperature mapping rules and boundary markers. The improved Cubist model includes a thermal boundary constraint rule generation module, a rule linear regression module and a boundary residual committee module. Using the parameters to be calibrated as candidate parameters, the predicted temperature is obtained according to the temperature mapping rule and compared with the measured temperature to generate temperature residuals. The temperature measurement position and the equivalent radial position are matched to form radial residual assignment. Based on candidate parameters and constraint benchmarks, regional heat transfer, energy conservation, and radial variation margins are determined to form a constraint set; The candidate parameters are iteratively optimized using the COBYLA algorithm, which introduces a constraint margin shrinkage operator and a temperature residual partition scaling factor. The boundary components are shrunk according to the boundary marker and constraint set, and the search scale is adjusted according to the residual assignment to obtain the target parameters. Substitute the target parameters into the continuous distribution relationship of heat transfer coefficients, and output the continuous heat transfer coefficient distribution after thermal simulation verification.

[0008] Optionally, the geometric information of the fire surface partitions includes the boundary position, contour information and adjacency relationship of each partition; the partition area is the surface area of ​​the fire surface corresponding to each partition; the injector center position is the position of the injector center in the fire surface coordinate system; the measured temperature is the temperature data of each temperature measuring point under the steady-state thermal condition of the cylinder head, and the temperature measuring position is the spatial position of the corresponding temperature measuring point in the cylinder head; the heat transfer constraint benchmark includes the heat transfer strength relationship benchmark between each partition, the overall heat transfer energy benchmark of the fire surface, the heat transfer temperature difference benchmark, and the radial variation benchmark of the heat transfer coefficient.

[0009] Optionally, establishing the continuous distribution relationship of the heat transfer coefficient and the parameters to be calibrated includes: Using the center position of the fuel injector as the radial reference, determine the radial distance corresponding to the boundary of the firepower surface zone, and form the radial range of each firepower surface zone; The centroid of the area is determined based on the boundary position and area of ​​the fire zone, and the radial position of the centroid of the area relative to the center of the fuel injector is determined as the equivalent radial position of the corresponding fire zone. Set the heat exchange control position according to the equivalent radial position, and configure the heat exchange parameters to be calibrated for the heat exchange control position; Based on the heat exchange control position and the heat exchange parameters to be calibrated, a distribution relationship of heat transfer coefficients that continuously varies with radial position is established, and the heat exchange parameters to be calibrated are combined into calibration parameters.

[0010] Optionally, generating the constraint state based on the constraint criterion includes: The parameters to be calibrated are perturbed within a preset range to generate multiple sets of parameter samples, and the continuous distribution of heat transfer coefficients corresponding to the parameter samples is determined. The heat transfer coefficient is continuously distributed and loaded into the thermal simulation model. Thermal simulation is performed, and the corresponding simulation temperature is extracted according to the temperature measurement location. The overall heat transfer of the fire surface is also extracted. By comparing the heat transfer coefficients corresponding to the parameter samples with the heat transfer constraint benchmark according to the radial order of the fire surface partition, the regional heat transfer relationship state and radial variation state are determined. The overall heat exchange of the firepower surface is compared with the overall heat exchange energy benchmark to determine the energy conservation state. The heat exchange relationship state of the collection area, the radial change state, and the energy conservation state form the constraint state.

[0011] Optionally, the generated temperature mapping rules and boundary markers include: The rule generation, rule-based linear regression, and committee prediction process based on the Cubist model are as follows: In the rule generation process, a candidate segmentation and screening process based on the thermal boundary constraint state is added to form a thermal boundary constraint rule generation module; the linear regression processing corresponding to the rules forms a rule-based linear regression module; in the committee prediction process, a residual correction weight allocation process based on the constraint boundary margin is added to form a boundary residual committee module; the output of the thermal boundary constraint rule generation module is passed to the rule-based linear regression module, and the output of the rule-based linear regression module is passed to the boundary residual committee module, forming an improved Cubist model. Input the parameter samples, constraint states, and simulation temperatures into the thermal boundary constraint rule generation module. Use the regional heat transfer relationship state, radial change state, and energy conservation state as thermal boundary constraint states. Traverse the candidate cutting positions of the parameter samples, filter the candidate cutting positions based on the differences in thermal boundary constraint states, and determine the cutting positions according to the fitting deviation of the simulation temperature to generate rule partitions and constraint boundary information. Input the rule partition, parameter samples and simulation temperature into the rule linear regression module, establish the local linear response relationship between the parameter to be calibrated and the simulation temperature within the rule partition, generate the initial temperature mapping rule, and compare the corresponding predicted temperature with the simulation temperature to generate the temperature prediction residual. The constraint boundary information, constraint state, and temperature prediction residual are input into the boundary residual committee module. The boundary samples are determined based on the margin between the constraint state and the constraint boundary. The residual correction weights are allocated according to the constraint boundary margin. The training response of the committee is corrected using the weighted temperature prediction residual, and temperature mapping rules and boundary labels are generated. The improved Cubist model was trained using parameter samples, constraint states, and simulation temperatures. The parameters for rule segmentation, local linear response, and committee residual correction were optimized to obtain the trained improved Cubist model.

[0012] Optionally, training the improved Cubist model using parameter samples, constraint states, and simulation temperatures includes: Based on the correspondence between parameter samples, the constraint states and simulation temperatures are associated, and the associated samples are divided into training samples and validation samples. The training samples are input into the thermal boundary constraint rule generation module. Candidate segments are selected based on the regional heat transfer relationship, radial change state, and energy conservation state. The rule segmentation position is determined according to the simulation temperature fitting deviation corresponding to the candidate segmentation, and the training rule partition and constraint boundary information are obtained. The training rule partitions and corresponding training samples are input into the rule linear regression module. The parameters to be calibrated are used as regression variables and the simulated temperature is used as the target response. The local linear response parameters corresponding to the rule partitions are updated to obtain the training temperature mapping rules and temperature prediction residuals. The temperature prediction residual, constraint state, and constraint boundary information are input into the boundary residual committee module. The residual correction weights are allocated according to the margin between the training samples and the constraint boundary. The training response of the next committee is updated using the weighted temperature prediction residuals, and the committee residual correction parameters are obtained iteratively. The validation sample is input into the improved Cubist model that completes the current iteration to obtain the validation predicted temperature. The boundary sample in the validation sample is determined according to the constraint boundary information. The temperature prediction deviation between the validation sample and the boundary sample is calculated respectively. When the temperature prediction deviation of the two types meets the preset convergence condition, the trained improved Cubist model is obtained.

[0013] Optionally, the step of obtaining the predicted temperature according to the temperature mapping rule and comparing it with the measured temperature to generate a temperature residual, and matching the temperature measurement location and the equivalent radial location to form a radial residual assignment, includes: The candidate parameters are matched with the temperature mapping rules to determine the rule partition to which the candidate parameters belong, and the predicted temperature of the temperature measurement location is obtained according to the corresponding temperature mapping rule. According to the correspondence of temperature measurement locations, the predicted temperature is compared with the actual temperature one by one to determine the direction and amount of temperature deviation and generate temperature residual. Using the center position of the fuel injector as the radial reference, determine the radial position of the temperature measurement position, and match it with the equivalent radial position of the fire surface zone to determine the radial zone corresponding to the temperature residual. The temperature residuals are associated with the corresponding radial partitions to the parameter to be calibrated, and the temperature residuals associated with the same parameter to be calibrated are aggregated to form radial residual assignment.

[0014] Optionally, forming the constraint set includes: Substitute the candidate parameters into the continuous distribution relationship of heat transfer coefficients to determine the heat transfer coefficients corresponding to the fire surface zones, and compare them with the benchmark of regional heat transfer strength to generate regional heat transfer margin. The overall heat transfer is determined by combining the heat transfer coefficient, zone area and heat transfer temperature difference benchmark corresponding to the fire surface zone, and compared with the overall heat transfer energy benchmark to generate energy conservation margin. The changes in heat transfer coefficients of adjacent fire surface zones are compared according to the equivalent radial position and compared with the radial change benchmark to generate radial change margin. The heat transfer margin, energy conservation margin, and radial variation margin of the aggregated area are used to form a constraint set.

[0015] Optionally, the obtained target parameters include: Using the candidate parameters as the current parameters, a linear approximate relationship is established between the temperature residual and the constraint set within the trust region of the COBYLA algorithm, generating the candidate parameter increment; Based on the boundary markers, the regional heat transfer margin, energy conservation margin, and radial variation margin before and after the incremental action of candidate parameters are compared to determine the parameter components that touch the constraint boundary and their corresponding shrinkage magnitude, and construct the constraint margin shrinkage operator. The constraint margin shrinkage operator is used to correct the increment of the candidate parameters, shrinking the parameter components that touch the constraint boundary to generate corrected candidate parameters. The temperature residuals corresponding to the parameter components are determined based on the radial residual assignment, and the search scale scaling amount is determined in combination with the corresponding constraint margin. A temperature residual partition scaling factor is constructed, and the trust region search scale of the corresponding parameter components in the next iteration is adjusted using the scaling factor. The candidate parameters are rematched with the temperature mapping rules, the temperature residuals and boundary markers are updated, and the constraint set is updated according to the revised candidate parameters. The process of generating candidate parameters, shrinking parameters, and adjusting the search scale is repeated until the preset convergence conditions are met, and the target parameters are obtained.

[0016] Optionally, the continuous heat transfer coefficient distribution output after thermal simulation verification includes: Substitute the target parameters into the continuous distribution relationship of heat transfer coefficients to determine the heat transfer coefficients corresponding to the radial positions of the fire surface, thus forming the target heat transfer coefficient distribution. The target heat transfer coefficient distribution is loaded into the thermal simulation model, the simulation temperature is extracted and verified according to the temperature measurement location, and the temperature verification result is obtained by comparing it with the measured temperature. Based on the target heat transfer coefficient distribution and heat transfer constraint benchmark, the regional heat transfer relationship, the overall energy conservation relationship and the radial variation relationship are checked respectively to obtain the constraint check results; When both the temperature verification result and the constraint verification result meet the preset verification conditions, the target heat transfer coefficient distribution is determined as the continuous heat transfer coefficient distribution of the cylinder head fire surface and output.

[0017] The beneficial effects of this invention are: This invention determines the radial range and equivalent radial position of the fire surface partition by acquiring the geometric information, area, and center position of the fuel injector of the cylinder head fire surface partition. It also establishes a distribution relationship in which the heat transfer coefficient changes continuously with the radial position, transforming the heat transfer coefficient calibration from discrete partition assignment to continuous parameterized expression. This reduces the impact of abrupt changes in heat transfer parameters between adjacent partitions on the description of the thermal boundary and improves the consistency between the heat transfer distribution of the cylinder head fire surface and the actual spatial heat transfer characteristics.

[0018] This invention establishes temperature mapping rules between the parameters to be calibrated and the simulation temperature by improving the Cubist model. In the rule generation process, candidate cutting positions are screened in combination with the thermal boundary constraint state. In the committee prediction process, the residual correction weight is adjusted in combination with the constraint boundary margin. This allows the surrogate temperature response to take into account parameter changes, regional heat transfer relationships, radial changes and energy conservation states, reduce the number of times the thermal simulation model is repeatedly called in the parameter iteration process, and improve the applicability of temperature response prediction near the constraint boundary.

[0019] This invention further employs the COBYLA algorithm, which introduces a constraint margin shrinkage operator and a temperature residual partition scaling factor, for parameter optimization. Parameter components affecting the boundaries of regional heat transfer, energy conservation, and radial variation constraints are shrunk, and the search scale of the corresponding parameters is adjusted according to the radial attribution of the temperature residual. This ensures that parameter updates are simultaneously controlled by temperature matching results and multiple heat transfer physical constraints. Finally, thermal simulation is used to verify the target parameters, achieving a balance between temperature calibration accuracy, heat transfer coefficient distribution continuity, and physical constraint consistency, thereby improving the accuracy, stability, and engineering applicability of the cylinder head fire surface heat transfer coefficient calibration results. Attached Figure Description

[0020] 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 a flowchart of a method for calibrating the heat transfer coefficient of the cylinder head fire surface based on a multi-constraint optimization algorithm proposed in this invention; Figure 2 This is a schematic diagram of the improved Cubist model for a cylinder head fire surface heat transfer coefficient calibration method based on a multi-constraint optimization algorithm proposed in this invention. Figure 3 This is a flowchart of the COBYLA algorithm for a cylinder head fire surface heat transfer coefficient calibration method based on a multi-constraint optimization algorithm proposed in this invention. Detailed Implementation

[0021] 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.

[0022] refer to Figure 1 , Figure 2 and Figure 3 A method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm, comprising: Obtain the geometric information of the firepower surface zones, zone area, injector center position, measured temperature, temperature measurement location, and heat transfer constraint benchmark; Determine the radial range and equivalent radial position of each fire zone, and establish the continuous distribution relationship of heat transfer coefficients and parameters to be calibrated; The parameters to be calibrated are disturbed to generate parameter samples, the corresponding heat transfer distribution is loaded into the thermal simulation model to obtain the simulation temperature, and the constraint state is generated according to the constraint benchmark. The Cubist model is improved by inputting parameter samples, constraint states and simulation temperatures to generate temperature mapping rules and boundary markers. The improved Cubist model includes a thermal boundary constraint rule generation module, a rule linear regression module and a boundary residual committee module. Using the parameters to be calibrated as candidate parameters, the predicted temperature is obtained according to the temperature mapping rule and compared with the measured temperature to generate temperature residuals. The temperature measurement position and the equivalent radial position are matched to form radial residual assignment. Based on candidate parameters and constraint benchmarks, regional heat transfer, energy conservation, and radial variation margins are determined to form a constraint set; The candidate parameters are iteratively optimized using the COBYLA algorithm, which introduces a constraint margin shrinkage operator and a temperature residual partition scaling factor. The boundary components are shrunk according to the boundary marker and constraint set, and the search scale is adjusted according to the residual assignment to obtain the target parameters. Substitute the target parameters into the continuous distribution relationship of heat transfer coefficients, and output the continuous heat transfer coefficient distribution after thermal simulation verification.

[0023] In this embodiment, the geometric information of the fire surface partitions includes the boundary position, contour information and adjacency relationship of each partition; the partition area is the surface area of ​​the fire surface corresponding to each partition; the injector center position is the position of the injector center in the fire surface coordinate system; the measured temperature is the temperature data of each temperature measuring point under the steady-state thermal condition of the cylinder head, and the temperature measuring position is the spatial position of the corresponding temperature measuring point in the cylinder head; the heat transfer constraint benchmark includes the heat transfer strength relationship benchmark between each partition, the overall heat transfer energy benchmark of the fire surface, the heat transfer temperature difference benchmark, and the radial variation benchmark of the heat transfer coefficient.

[0024] In this embodiment, establishing the continuous distribution relationship of the heat transfer coefficient and the parameters to be calibrated includes: Using the center position of the fuel injector as the radial reference, the radial distance corresponding to the boundary of the fire control zone is determined, and the radial range of each fire control zone is formed. Specifically, determining the radial distance corresponding to the boundary of the fire control zone and forming the radial range of the fire control zone is as follows: Read the coordinates of the fuel injector center and the boundary of the fire surface in the fire surface coordinate system. Calculate the coordinate difference between each boundary position and the fuel injector center in each coordinate direction. Square each coordinate difference and sum them. Take the square root of the sum to obtain the radial distance of the corresponding boundary position. Traverse all boundary positions of the same fire surface partition and extract the minimum and maximum radial distances. Use these as the inner and outer radial boundaries of the corresponding fire surface partition, respectively. Use the distance interval between the inner and outer radial boundaries as the radial range of the corresponding fire surface partition. The centroid of the area is determined based on the boundary position and area of ​​the fire-fighting surface zone. The radial position of the centroid relative to the center of the fuel injector is then defined as the equivalent radial position of the corresponding fire-fighting surface zone. Specifically: The fire zone coverage area is defined by the boundary positions. Connecting the non-adjacent boundary positions within the coverage area, the coverage area is divided into non-overlapping triangular area units. The average coordinates of the three vertices of each triangular area unit are calculated to obtain the center position of the area unit. The area of ​​each area unit is calculated using the same method as the area of ​​a triangle. The ratio of the area of ​​each area unit to the area of ​​the corresponding zone is used as the area weight. Each coordinate component of the center position of the area unit is multiplied by the area weight and summed to obtain the centroid position of the fire zone. The coordinate difference between the centroid position and the center position of the fuel injector in each coordinate direction is calculated. The squares of these differences are summed, and the square root of the sum is taken to obtain the radial distance between the centroid and the center of the fuel injector. This radial distance is determined as the equivalent radial position of the corresponding fire zone. The heat transfer control positions are set according to the equivalent radial position, and the heat transfer parameters to be calibrated are configured for each heat transfer control position. Specifically, the process involves setting the heat transfer control positions according to the equivalent radial position and configuring the heat transfer parameters to be calibrated for each heat transfer control position. Fire surface partitions are arranged in ascending order of equivalent radial position. The equivalent radial position corresponding to each fire surface partition is determined as the heat transfer control position. The initial thermal simulation boundary data corresponding to the current steady-state thermal condition is read, and the initial heat transfer coefficient corresponding to the fire surface partition is extracted. The initial heat transfer coefficient is configured as the initial value of the heat transfer parameter to be calibrated at the corresponding heat transfer control position. The historical heat transfer coefficient calibration data of the same type of engine is read, and the minimum and maximum values ​​of the heat transfer coefficient of the corresponding fire surface partition are extracted. These values ​​are used as the lower limit and upper limit of the heat transfer parameter to be calibrated, respectively, forming a one-to-one correspondence between the heat transfer control position and the heat transfer parameter to be calibrated. Based on the heat transfer control position and the heat transfer parameters to be calibrated, a heat transfer coefficient distribution relationship that continuously varies with radial position is established, and the heat transfer parameters to be calibrated are then combined into calibration parameters. Specifically, the establishment of the heat transfer coefficient distribution relationship that continuously varies with radial position based on the heat transfer control position and the heat transfer parameters to be calibrated, and the combination of these parameters into calibration parameters, is as follows: Adjacent control sections are divided according to the radial order of heat exchange control positions from smallest to largest. For radial positions located between adjacent heat exchange control positions, the distance from the radial position to each of the two heat exchange control positions is calculated. The proportion of one distance to the sum of the two distances is used as the combination weight of the heat exchange parameter to be calibrated corresponding to the other heat exchange control position. The proportion of the other distance to the sum of the two distances is used as the combination weight of the heat exchange parameter to be calibrated corresponding to the current heat exchange control position. The two heat exchange parameters to be calibrated are weighted and combined to obtain the heat transfer coefficient of the corresponding radial position. The heat transfer coefficient between the inner radial boundary of the fire surface and the first heat exchange control position is set as the first heat transfer parameter to be calibrated, and the heat transfer coefficient between the last heat exchange control position and the outer radial boundary of the fire surface is set as the last heat transfer parameter to be calibrated. Connecting all control sections forms a continuous distribution relationship of heat transfer coefficients covering the radial range of the fire surface. The heat exchange parameters to be calibrated are arranged according to the radial order of the heat exchange control positions to form the calibration parameters.

[0025] In this embodiment, generating the constraint state based on the constraint criterion includes: The parameters to be calibrated are perturbed within a preset range to generate multiple sets of parameter samples, and the continuous distribution of the heat transfer coefficients corresponding to these parameter samples is determined. Specifically, generating multiple sets of parameter samples and determining the continuous distribution of the heat transfer coefficients corresponding to these parameter samples involves: Historical heat transfer coefficient calibration data of the same type of engine is read. The minimum and maximum historical heat transfer coefficient values ​​at the corresponding heat transfer control positions are used as the lower and upper limits of the heat transfer parameters to be calibrated, respectively. The heat transfer coefficient in the initial thermal simulation boundary data corresponding to the current steady-state thermal condition is used as the initial value of the heat transfer parameter to be calibrated. For each heat transfer control position, the lower limit, the average of the lower limit and the initial value, the initial value, the average of the initial value and the upper limit, and the upper limit are selected as parameter perturbation values. The parameter perturbation values ​​at different positions are fully combined according to the radial order of the heat transfer control positions. Candidate parameter combinations are formed; the normalized absolute deviations of the heat transfer parameters to be calibrated relative to their corresponding initial values ​​in each candidate parameter combination are calculated and summed, and the candidate parameter combinations are arranged according to the summation results. The arranged candidate parameter combinations are divided into multiple continuous intervals, and the candidate parameter combination at the middle position of each continuous interval is selected as the parameter sample; the heat transfer parameters to be calibrated in each parameter sample are configured to the corresponding heat transfer control positions, and the heat transfer coefficients between adjacent heat transfer control positions are weighted according to the radial distance ratio to form a continuous distribution of heat transfer coefficients covering the radial range of the fire surface; The heat transfer coefficient is continuously distributed and applied to the thermal simulation model. Thermal simulation is then performed, and the corresponding simulation temperature is extracted according to the temperature measurement location. The overall heat transfer of the fire-fighting surface is also extracted. Specifically, the thermal simulation process involves extracting the corresponding simulation temperature according to the temperature measurement location and extracting the overall heat transfer of the fire-fighting surface. The heat transfer coefficients corresponding to the parameter samples are continuously distributed and mapped to the fire surface grid cells of the cylinder head thermal simulation model according to the radial position of the fire surface, serving as the combustion-side heat transfer boundary. The thermal simulation settings corresponding to the steady-state conditions of the measured temperature are read, and the cylinder head material thermophysical parameters, cooling-side heat transfer boundary, combustion gas temperature boundary, and environmental heat transfer boundary are configured into the thermal simulation model, keeping the boundary conditions consistent with the measured temperature acquisition conditions, and the steady-state temperature field of the cylinder head is solved. The nearest thermal simulation grid node is found according to the spatial coordinates of the temperature measurement location, and the temperature of the corresponding node is used as the simulation temperature of the corresponding temperature measurement location. For each fire surface grid cell, the temperature difference between the combustion gas temperature and the surface temperature of the grid cell is calculated. The heat transfer of the grid cell is obtained by multiplying the temperature difference, the corresponding heat transfer coefficient of the grid cell, and the surface area of ​​the grid cell. The heat transfer of all fire surface grid cells is accumulated to obtain the overall heat transfer of the fire surface. By comparing the heat transfer coefficients corresponding to the parameter samples in the radial order of the fire zone with the heat transfer constraint benchmark, the regional heat transfer relationship state and radial variation state are determined. Specifically, the determination of the regional heat transfer relationship state and radial variation state is as follows: The fire surfaces are arranged in ascending order of equivalent radial position. The heat transfer coefficients for each equivalent radial position are extracted from the continuous distribution of heat transfer coefficients corresponding to the parameter samples. The pairing relationships and order of heat transfer strength in the heat transfer strength benchmark are read. The difference between the heat transfer coefficients of the strong and weak heat transfer zones in each pair is calculated. A difference greater than zero is considered a satisfied state, a difference equal to zero is considered a boundary state, and a difference less than zero is considered a violation state. The comparison results of all paired zones are collected to form the regional heat transfer relationship state. Adjacent fire surfaces are selected sequentially according to the order of equivalent radial position. The process involves dividing the region into zones, calculating the difference between the heat transfer coefficient of the next zone and the heat transfer coefficient of the previous zone, and dividing the difference by the distance between two equivalent radial positions to obtain the radial heat transfer rate of change. The minimum and maximum values ​​of the heat transfer rate of change in adjacent radial intervals are obtained from the historical calibration data of thermal simulation and measured temperature verification of the same type of engine in the radial change benchmark. A radial heat transfer rate between the minimum and maximum values ​​is defined as a satisfied state, equal to the minimum or maximum value is defined as a boundary state, and below the minimum or above the maximum value is defined as a violation state. The comparison results of all adjacent fire surface zones are then compiled to form the radial change state. The energy conservation state is determined by comparing the overall heat transfer of the fire-generating surface with the overall heat transfer energy benchmark. The heat transfer relationship state, radial variation state, and energy conservation state of the collecting region form a constraint state, specifically: The overall heat transfer of the fire surface is read from the historical heat transfer coefficient calibration data of the same type of engine, and verified by measured temperature and thermal balance. The minimum and maximum values ​​of the historical overall heat transfer of the fire surface are determined as the lower limit and upper limit of the overall heat transfer energy benchmark, respectively. The overall heat transfer of the fire surface corresponding to the parameter sample is compared with the overall heat transfer energy benchmark. When the overall heat transfer of the fire surface is between the lower limit and the upper limit, it is determined as a satisfied state. When it is equal to the lower limit or the upper limit, it is determined as a boundary state. When it is below the lower limit or above the upper limit, it is determined as a violated state, forming an energy conservation state. According to the one-to-one correspondence between the parameter sample and the simulation results, the regional heat transfer relationship state, radial change state and energy conservation state corresponding to the same parameter sample are associated to form the constraint state of the corresponding parameter sample.

[0026] In this embodiment, the generated temperature mapping rules and boundary markers include: The rule generation, rule-based linear regression, and committee prediction process based on the Cubist model are as follows: In the rule generation process, a candidate segmentation and screening process based on the thermal boundary constraint state is added to form a thermal boundary constraint rule generation module; the linear regression processing corresponding to the rules forms a rule-based linear regression module; in the committee prediction process, a residual correction weight allocation process based on the constraint boundary margin is added to form a boundary residual committee module; the output of the thermal boundary constraint rule generation module is passed to the rule-based linear regression module, and the output of the rule-based linear regression module is passed to the boundary residual committee module, forming an improved Cubist model. This invention structurally optimizes the traditional Cubist model by introducing thermal boundary constraint information into the rule generation and committee prediction processes, thereby improving the model's ability to express the relationship between heat transfer parameters and simulation temperature. To address the problem that traditional rule-based segmentation mainly relies on fitting error and is difficult to reflect differences in physical constraints, a candidate segmentation screening process based on thermal boundary constraint states is introduced, enabling rule-based partitioning to simultaneously reflect differences in temperature response and heat transfer constraints. To enhance the ability to characterize temperature response relationships within different parameter ranges, a local linear response relationship is established within a regular partition to improve the fitting accuracy of temperature prediction results corresponding to different combinations of heat transfer parameters. To address the issue of insufficient prediction stability for samples near constraint boundaries, a residual correction weight allocation process based on constraint boundary margin is introduced to enable boundary samples to receive targeted residual correction and improve the prediction reliability of the constraint boundary region. The improved Cubist model can balance temperature mapping accuracy, physical constraint consistency, and boundary prediction stability, providing a reliable temperature response basis for multi-constraint iterative optimization of heat transfer coefficient; The parameter samples, constraint states, and simulation temperatures are input into the thermal boundary constraint rule generation module. The regional heat transfer relationship state, radial variation state, and energy conservation state are used as the thermal boundary constraint states. The module iterates through candidate segmentation positions of the parameter samples, filters candidate segmentation positions based on differences in thermal boundary constraint states, and determines the segmentation position according to the simulation temperature fitting deviation. This generates rule partitions and constraint boundary information. The thermal boundary constraint rule generation module includes: Constraint State Processing Unit: According to the one-to-one correspondence of parameter samples, the heat transfer relationship state, radial change state and energy conservation state of the region are associated. Satisfactory state, boundary state and violation state are encoded into different state tags to form the thermal boundary constraint state corresponding to the parameter sample. Candidate segmentation generation unit: Reads parameter samples dimension by dimension according to the parameter to be calibrated, arranges the parameter values ​​of the same parameter dimension in ascending order of value, takes the midpoint of adjacent parameter values ​​with different values ​​as the candidate segmentation position, and divides the parameter samples into two candidate sample groups according to the candidate segmentation position; Constraint Difference Screening Unit: The unit counts the number of samples corresponding to each state label in the two candidate sample groups for the regional heat transfer relationship state, radial change state, and energy conservation state. It divides the number of samples for each state label by the total number of samples in its respective candidate sample group to obtain the state percentage. The absolute difference between the state percentages of the two candidate sample groups is calculated and summed to obtain the constraint state difference value. Candidate splitting positions with constraint state difference values ​​greater than zero are retained. When all constraint state difference values ​​are equal to zero, all candidate splitting positions for the current parameter dimension are retained. Segmentation Judgment Unit: For the retained candidate segmentation positions, calculate the average value of the simulated temperature at each temperature measurement position in the two candidate sample groups, calculate the squared difference between each simulated temperature and the corresponding average value of the candidate sample group, sum up all the squared differences to obtain the simulation temperature fitting deviation, determine the candidate segmentation position with the smallest simulation temperature fitting deviation as the current rule segmentation position, and record the corresponding parameter dimension, parameter segmentation value and thermal boundary constraint status on both sides. In the thermal boundary constraint rule generation module, the constraint state processing unit converts the three types of constraint states corresponding to the parameter samples into comparable thermal boundary constraint states. The candidate segmentation generation unit generates candidate segmentation positions along the dimensions of each parameter to be calibrated and forms corresponding candidate sample groups. The constraint difference screening unit uses the state proportion difference between candidate sample groups to exclude segmentation positions that cannot distinguish thermal boundary constraint states. The segmentation determination unit compares the simulation temperature fitting deviation among the retained candidate segmentation positions, determines the rule segmentation position, and continues to divide the parameter samples according to the rule segmentation position until there are no candidate segmentation positions with different values ​​in the current sample group or the number of samples is less than 4, forming a rule partition. The parameter dimension, parameter segmentation value, and thermal boundary constraint states on both sides corresponding to the rule segmentation position are combined to form constraint boundary information. The regular partition, parameter samples, and simulation temperature are input into the regular linear regression module. Within the regular partition, a local linear response relationship between the parameter to be calibrated and the simulation temperature is established, generating an initial temperature mapping rule. The corresponding predicted temperature is then compared with the simulation temperature to generate a temperature prediction residual. The regular linear regression module includes: Rule matching unit: Reads the parameter dimensions and parameter segmentation values ​​corresponding to the rule partition, compares each parameter component of the parameter sample with the parameter segmentation value in each group, and classifies the parameter samples and corresponding simulation temperatures that meet the same group of segmentation conditions into the same rule partition; Regression variable processing unit: Arrange the heat transfer parameters to be calibrated in each group of parameter samples within the partition according to the radial order of the heat transfer control position. Use the heat transfer parameters to be calibrated as regression variables and arrange the corresponding simulation temperatures according to the temperature measurement position to form local regression data corresponding to each temperature measurement position. Local response fitting unit: For each regular partition and each temperature measurement position, the heat transfer parameter to be calibrated in the regular partition is multiplied by the corresponding regression coefficient and the bias term is accumulated to obtain the fitting temperature. The square of the difference between the fitting temperature and the simulation temperature is calculated and accumulated for all parameter samples in the regular partition. The local linear response relationship between the parameter to be calibrated and the simulation temperature of the corresponding temperature measurement position is established with the regression coefficient and bias term corresponding to the minimum accumulation result. The parameter segmentation conditions, regression coefficients and bias terms of the regular partition are combined to form the initial temperature mapping rule. Residual calculation unit: Substitute each set of parameter samples in the regular partition into the corresponding local linear response relationship to obtain the predicted temperature of each temperature measurement location, subtract the corresponding predicted temperature from the simulated temperature to obtain the temperature prediction residual, and associate the temperature prediction residual with the parameter sample identifier and the temperature measurement location identifier. In the regular linear regression module, the rule matching unit determines the rule partition to which the parameter sample belongs based on the parameter segmentation conditions. The regression variable organization unit organizes the heat transfer parameters to be calibrated according to a unified radial order and associates them with the corresponding simulation temperatures. The local response fitting unit adjusts the regression coefficients and bias terms for each rule partition and each temperature measurement location. The minimum sum of the squared differences between the fitted temperature and the simulation temperature within the rule partition is used as the coefficient determination condition to establish a local linear response relationship and form an initial temperature mapping rule. The residual calculation unit uses the initial temperature mapping rule to calculate the predicted temperature corresponding to the parameter sample and forms the temperature prediction residual through the difference between the simulation temperature and the predicted temperature. The constraint boundary information, constraint state, and temperature prediction residual are input into the boundary residual committee module. Boundary samples are determined based on the margin between the constraint state and the constraint boundary. Residual correction weights are assigned according to the constraint boundary margin. The weighted temperature prediction residual is used to correct the training response of the committee, generating temperature mapping rules and boundary labels. The boundary residual committee module includes: Boundary margin calculation unit: Reads the parameter dimension, parameter segmentation value and thermal boundary constraint status on both sides from the constraint boundary information. According to the correspondence between parameter samples and constraint status, it calculates the distance from the heat transfer coefficient difference corresponding to the regional heat transfer relationship to the zero boundary, the minimum distance from the radial rate of change to the corresponding lower or upper limit of the rate of change, and the minimum distance from the overall heat transfer of the fire surface to the lower or upper limit of the overall heat transfer energy. The corresponding distances are used as the regional heat transfer constraint margin, the radial change constraint margin and the energy conservation constraint margin. Boundary Sample Determination Unit: Within each rule partition, parameter samples are arranged in ascending order of the absolute value of the three types of constraint margins. The parameter sample with the smallest absolute margin in each type of constraint is determined as the corresponding constraint boundary sample. When multiple parameter samples have the same minimum absolute margin, they are all determined as boundary samples, and the constraint type and constraint boundary corresponding to the boundary sample are recorded. Residual weight allocation unit: For the three types of constraint boundary samples, the sorting number is determined according to the absolute value of the constraint margin from small to large. The weight base is obtained by subtracting the sorting number from the total number of boundary samples of the same type and adding 1. The weight base of each boundary sample is divided by the sum of the weight bases of all boundary samples of the same type to obtain the residual correction weight. The residual correction weight is multiplied by the temperature prediction residual of the corresponding temperature measurement position to obtain the weighted temperature prediction residual. Committee Response Correction Unit: The predicted temperature output by the current committee for the boundary sample is added to the corresponding weighted temperature prediction residual to obtain the corrected temperature response. The corrected temperature response is used as the response value of the next committee for the corresponding boundary sample. The non-boundary sample retains the original simulation temperature as the response value, and the residual correction process is repeated according to the committee arrangement order. In the boundary residual committee module, the boundary margin calculation unit determines the distance of the parameter samples relative to the regional heat transfer constraint boundary, radial variation constraint boundary, and energy conservation constraint boundary, respectively. The boundary sample determination unit determines the parameter samples close to the constraint boundary within the rule partition according to the absolute value of each type of constraint margin. The residual weight allocation unit allocates residual correction weights according to the proximity of the boundary samples to the constraint boundary. The committee response correction unit uses the weighted temperature prediction residual to correct the committee response step by step. The rule partition conditions and local linear response parameters formed by each committee are collected, and the arithmetic mean of the predicted temperatures of each committee is taken to form a temperature mapping rule. The parameter dimension, constraint type, and constraint boundary corresponding to the boundary sample are combined to form a boundary marker. The improved Cubist model was trained using parameter samples, constraint states, and simulation temperatures. The parameters for rule segmentation, local linear response, and committee residual correction were optimized to obtain the trained improved Cubist model.

[0027] In this embodiment, training the improved Cubist model using parameter samples, constraint states, and simulation temperature includes: Based on the correspondence between parameter samples and the constraint states and simulation temperatures, the correlated samples are divided into training samples and validation samples. Specifically, the division of the correlated samples into training samples and validation samples is as follows: Based on the parameter sample identifiers, each group of parameter samples is associated with the corresponding regional heat transfer relationship state, radial change state, energy conservation state, and simulated temperature at each temperature measurement location to form a complete sample. The complete samples are then grouped according to the combinations of satisfied, boundary, and violated states in the regional heat transfer relationship state, radial change state, and energy conservation state, forming sample groups with different constraint state combination types. Each sample group is arranged in ascending order of parameter sample identifiers. 80% of the total number of sample groups is calculated and rounded down to obtain the number of training samples. Complete samples within the training sample number range are included in the training samples, and the remaining complete samples are included in the validation samples. The training and validation samples corresponding to each constraint state combination type are then collected to form a training sample set and a validation sample set, ensuring that both the training and validation sample sets retain the parameter samples and simulated temperatures corresponding to different thermal boundary constraint states. The training samples are input into the thermal boundary constraint rule generation module. Candidate segments are selected based on the regional heat transfer relationship, radial change state, and energy conservation state. The rule segmentation position is determined according to the simulation temperature fitting deviation corresponding to the candidate segmentation, and the training rule partition and constraint boundary information are obtained. The training rule partitions and corresponding training samples are input into the rule-based linear regression module. Using the parameters to be calibrated as regression variables and the simulated temperature as the target response, the local linear response parameters corresponding to the rule partitions are updated, resulting in the training temperature mapping rule and the temperature prediction residual. Specifically, the training temperature mapping rule and the temperature prediction residual are obtained as follows: According to the training rule partition, the corresponding training samples are read separately. The heat transfer parameters to be calibrated at each heat transfer control position are arranged in radial order as regression variables, and the simulated temperature is read as the target response according to the temperature measurement position. For each training rule partition and each temperature measurement position, the heat transfer parameters to be calibrated are multiplied by the corresponding regression coefficient and summed with the bias term to obtain the training prediction temperature. The squared difference between the training prediction temperature and the simulated temperature is calculated and accumulated over all training samples in the training rule partition. The partial derivatives of the accumulated result with respect to each regression coefficient and bias term are calculated. The partial derivatives are set to zero to form a system of linear equations. The system of linear equations is solved to obtain the regression coefficients and bias terms that minimize the accumulated squared difference. The corresponding local linear response parameters are updated with the solution results. The parameter segmentation conditions of the training rule partition are associated with the updated local linear response parameters to form a training temperature mapping rule. The training samples are substituted into the corresponding training temperature mapping rule to obtain the predicted temperature at each temperature measurement position. The temperature prediction residual is obtained by subtracting the corresponding predicted temperature from the simulated temperature and associated with the training sample identifier and the temperature measurement position identifier. The temperature prediction residual, constraint state, and constraint boundary information are input into the boundary residual committee module. Residual correction weights are assigned based on the margin between the training samples and the constraint boundaries. The weighted temperature prediction residual is used to update the training response of the next committee, iteratively obtaining the committee residual correction parameters. Specifically, the iteratively obtained committee residual correction parameters are as follows: Five committees were set up. The first committee used the simulated temperature corresponding to the training sample as the training response. Based on the constraint boundary information, it calculated the regional heat transfer constraint margin, radial variation constraint margin, and energy conservation constraint margin corresponding to the training sample, and determined the boundary samples and residual correction weights according to the constraint margins. The temperature prediction residual of the boundary sample was multiplied by the corresponding residual correction weight to obtain the weighted temperature prediction residual. The predicted temperature of the boundary sample corresponding to the first committee was added to the weighted temperature prediction residual to form the corrected training response. The non-boundary samples maintained the corresponding simulated temperature as the training response. The resulting training response was input into the second committee. The residual correction of the second to fifth committees was completed in the same way. Each committee calculated the temperature prediction residual based on the training response formed by the previous committee and updated the boundary sample training response. After the processing of the five committees was completed, the residual correction weights of the boundary samples corresponding to each committee and the training response correction amount were recorded to form the committee residual correction parameters. The validation sample is input into the improved Cubist model that has completed the current iteration to obtain the validation predicted temperature. Boundary samples in the validation sample are determined based on the constraint boundary information. The temperature prediction deviations of the validation sample and the boundary samples are calculated separately. When the temperature prediction deviations of the two types meet the preset convergence condition, the trained improved Cubist model is obtained, in which the validation predicted temperature is obtained. The model's convergence condition is then determined based on the validation predicted temperature, the simulated temperature, and the prediction deviations corresponding to the boundary samples. Specifically: The calibration parameters from the validation samples are input into the improved Cubist model that completes the current iteration. The validation predicted temperature for each temperature measurement location is obtained according to the corresponding partitioning and temperature mapping rules. The absolute value of the difference between the validation predicted temperature and the corresponding simulated temperature for each temperature measurement location of each validation sample is calculated. All absolute values ​​are summed and divided by the product of the number of validation samples and the number of temperature measurement locations to obtain the validation average absolute deviation. Based on the constraint boundary information, the regional heat transfer constraint margin, radial variation constraint margin, and energy conservation constraint margin corresponding to the validation sample are calculated. The validation sample with the smallest absolute margin among all constraints is determined as the boundary sample. The absolute value of the difference between the validation predicted temperature and the corresponding simulated temperature for each temperature measurement location of the boundary sample is calculated. All absolute values ​​are summed and divided by the number of boundary samples and the number of temperature measurement locations. The product is used to obtain the boundary mean absolute deviation. After the first iteration, the verification mean absolute deviation and the boundary mean absolute deviation are recorded without determining the decrease ratio. Starting from the second iteration, the decrease ratio of the two types of deviations relative to the previous iteration is calculated. The ratio of the decrease in the verification mean absolute deviation of the current iteration relative to the verification mean absolute deviation of the previous iteration to the verification mean absolute deviation of the previous iteration, and the ratio of the decrease in the boundary mean absolute deviation of the current iteration relative to the boundary mean absolute deviation of the previous iteration to the boundary mean absolute deviation of the previous iteration are calculated to obtain the decrease ratio of the two types of deviations. When the decrease ratio of both types of deviations is less than 1%, the convergence condition is determined to be met, and the improved Cubist model is obtained after training. When either of the decrease ratios of the two types of deviations is not less than 1%, the model parameters are updated and the next round of training is executed. In this embodiment, the step of obtaining the predicted temperature according to the temperature mapping rule and comparing it with the measured temperature to generate the temperature residual, and matching the temperature measurement position and the equivalent radial position to form the radial residual assignment includes: The candidate parameters are matched with temperature mapping rules to determine the rule partition to which the candidate parameters belong, and the predicted temperature of the temperature measurement location is obtained according to the corresponding temperature mapping rule. Specifically, determining the rule partition to which the candidate parameters belong and obtaining the predicted temperature of the temperature measurement location according to the corresponding temperature mapping rule involves: Read the parameter dimensions, parameter segmentation values, and parameter value ranges recorded by each committee in the temperature mapping rules. Extract the corresponding parameter components from the candidate parameters according to the parameter dimensions. Compare each parameter component with the parameter segmentation value. When a parameter component falls within the value range defined by the rule, it is determined that the corresponding segmentation condition is met. The rule partition that simultaneously meets all segmentation conditions is determined as the rule partition to which the candidate parameter belongs. Read the regression coefficients and bias terms corresponding to each temperature measurement position in the rule partition. Multiply each parameter component of the candidate parameter by the corresponding regression coefficient. Sum all products and add the bias term to obtain the predicted temperature of a single committee at the corresponding temperature measurement position. Calculate the predicted temperatures of the five committees at the same temperature measurement position in the same way. Sum the five predicted temperatures and divide by 5 to obtain the predicted temperature of the corresponding temperature measurement position. Based on the correspondence of temperature measurement locations, the predicted temperature is compared with the measured temperature one by one to determine the direction and amount of temperature deviation, and to generate the temperature residual. Specifically, determining the direction and amount of temperature deviation and generating the temperature residual involves: According to the temperature measurement location markings, the predicted temperature at each measurement location is associated with the corresponding measured temperature. The signed temperature difference value for the corresponding measurement location is obtained by subtracting the predicted temperature from the measured temperature. This signed temperature difference value is used as the temperature residual. When the temperature residual is greater than zero, the temperature deviation direction is determined to be that the predicted temperature is lower than the measured temperature; when the temperature residual is less than zero, the temperature deviation direction is determined to be that the predicted temperature is higher than the measured temperature; when the temperature residual is equal to zero, no temperature deviation is determined. The absolute value of the temperature residual is taken to obtain the deviation amount, and the temperature deviation direction, deviation amount, and temperature residual are recorded according to the temperature measurement location markings. Using the center position of the fuel injector as the radial reference, the radial position corresponding to the temperature measurement location is determined and matched with the equivalent radial position of the fire surface zone to determine the radial zone corresponding to the temperature residual. Specifically, determining the radial position corresponding to the temperature measurement location and matching it with the equivalent radial position of the fire surface zone to determine the radial zone corresponding to the temperature residual is as follows: Read the position of the origin of the fire surface coordinate system and the directions of the two plane coordinate axes. Project the spatial coordinate difference between the temperature measurement position and the origin of the fire surface coordinate system onto the directions of the two plane coordinate axes to obtain the projected coordinates of the temperature measurement position on the fire surface reference plane. Calculate the coordinate difference between the projected coordinates and the center position of the fuel injector in the two plane coordinate directions. Square the two coordinate differences and sum them. Take the square root of the sum to obtain the radial position of the temperature measurement. Compare the radial positions of the temperature measurement zone by zone according to the radial inner and radial outer boundaries of the fire surface zone. When the radial position of the temperature measurement is located between the corresponding radial inner and radial outer boundaries, the corresponding fire surface zone is determined as the radial zone corresponding to the temperature residual. When the radial position of the temperature measurement falls within the radial range of two adjacent fire surface zones, calculate the distance between the radial position of the temperature measurement and the equivalent radial positions of the two fire surface zones. The fire surface zone corresponding to the equivalent radial position with the smaller distance is determined as the radial zone corresponding to the temperature residual. The temperature residuals are associated with the corresponding radial partitions to the parameter to be calibrated. The temperature residuals associated with the same parameter to be calibrated are aggregated to form radial residual attribution. The formation of radial residual attribution is specifically as follows: Read the equivalent radial position of each radial zone, and determine the heat transfer parameter to be calibrated for each radial zone according to the one-to-one correspondence between the equivalent radial position and the heat transfer control position. Associate the temperature residuals corresponding to each temperature measurement position in the radial zone with the corresponding heat transfer parameter to be calibrated according to the temperature measurement position identifier. When multiple temperature residuals are associated with the same heat transfer parameter to be calibrated, take the absolute value of all temperature residuals and sum them. Divide the sum by the number of temperature residuals to obtain the zone residual amount. At the same time, sum all signed temperature residuals and divide them by the number of temperature residuals to obtain the zone average residual. When the zone average residual is greater than zero, the deviation direction is determined to be that the predicted temperature is too low. When the zone average residual is less than zero, the deviation direction is determined to be that the predicted temperature is too high. When the zone average residual is equal to zero, the temperature deviation is determined to be balanced. Associate the corresponding temperature measurement position identifier, temperature residual, zone residual amount and deviation direction with the heat transfer parameter identifier to be calibrated to form the radial residual assignment.

[0028] In this embodiment, forming the constraint set includes: Substituting the candidate parameters into the continuous distribution relationship of heat transfer coefficients, the heat transfer coefficients corresponding to the fire surface zones are determined, and compared with the benchmark of regional heat transfer strength to generate regional heat transfer margins. Specifically, generating regional heat transfer margins involves: According to the radial order of the heat exchange control positions, each heat exchange parameter to be calibrated in the candidate parameters is configured to the corresponding heat exchange control position, and the heat exchange coefficient of each radial position of the fire surface is determined according to the radial distance ratio between adjacent heat exchange control positions. The equivalent radial position of each fire surface partition is read, and the heat exchange coefficient of the corresponding equivalent radial position is extracted from the continuous distribution of heat exchange coefficients as the partition heat exchange coefficient of the corresponding fire surface partition. The partition pairing relationship and heat exchange strength order predefined in the regional heat exchange strength relationship benchmark are read. For each pairing partition, the partition heat exchange coefficient of the strong heat exchange partition is subtracted from the partition heat exchange coefficient of the weak heat exchange partition to obtain the regional heat exchange margin of the corresponding partition pairing. When the regional heat exchange margin is greater than zero, it is determined that the regional heat exchange strength relationship is satisfied. When the regional heat exchange margin is equal to zero, it is determined that it is at the regional heat exchange constraint boundary. When the regional heat exchange margin is less than zero, it is determined that the regional heat exchange strength relationship is violated. The regional heat exchange margin corresponding to each partition pairing is retained respectively. The overall heat transfer is determined by combining the heat transfer coefficient, zone area, and heat transfer temperature difference benchmark corresponding to the fire surface zone. This is then compared with the overall heat transfer energy benchmark to generate an energy conservation margin. Specifically, the energy conservation margin is generated as follows: Historical calibration data of the same type of engine, verified by measured temperature and thermal balance, are read. The difference between the combustion gas temperature and the average surface temperature of each fire face zone is calculated. The arithmetic mean of the historical differences for the same fire face zone is used as the heat transfer temperature difference benchmark. The heat transfer coefficient, zone area, and heat transfer temperature difference benchmark corresponding to each fire face zone are multiplied to obtain the zone heat transfer of the corresponding fire face zone. The heat transfer of all zones is summed to obtain the candidate overall heat transfer corresponding to the candidate parameters. The lower limit and upper limit of the overall heat transfer energy benchmark are read. The lower limit of the candidate overall heat transfer is obtained by subtracting the lower limit from the candidate overall heat transfer, and the upper limit of the candidate overall heat transfer is obtained by subtracting the candidate overall heat transfer from the upper limit. The smaller value between the lower limit and the upper limit of the energy is taken as the energy conservation margin. When the energy conservation margin is greater than zero, the candidate overall heat transfer is determined to be within the allowable range of the overall heat transfer energy. When the energy conservation margin is equal to zero, the candidate overall heat transfer is determined to be at the energy conservation constraint boundary. When the energy conservation margin is less than zero, the candidate overall heat transfer exceeds the allowable range of the overall heat transfer energy. The changes in heat transfer coefficients of adjacent fire surface zones are compared according to their equivalent radial positions and compared with the radial change benchmark to generate a radial change margin. Specifically, the generation of the radial change margin is as follows: Fire surface zones are arranged in ascending order of equivalent radial position. The heat transfer coefficients of adjacent fire surface zones are read sequentially. The difference between the heat transfer coefficient of the next fire surface zone and the heat transfer coefficient of the previous fire surface zone is calculated, and this difference is divided by the distance between the two equivalent radial positions to obtain the radial heat transfer rate of change for the corresponding radial interval. Historical heat transfer coefficient calibration data of the same type of engine, verified by measured temperature and thermal balance, are read. The historical radial heat transfer rate of change for each group is calculated according to the same radial interval. The minimum and maximum values ​​of the historical radial heat transfer rate of change are used as the lower limit and the baseline for radial change, respectively. The upper limit is determined by subtracting the lower limit of the radial change reference from the radial heat transfer rate corresponding to the candidate parameter to obtain the lower limit of the radial change reference, and by subtracting the radial heat transfer rate corresponding to the candidate parameter from the upper limit of the radial change reference to obtain the upper limit of the radial change reference. The smaller value between the lower limit of the radial change reference and the upper limit of the radial change reference is taken as the radial change margin of the corresponding radial interval. When the radial change margin is greater than zero, the radial change constraint is satisfied. When the radial change margin is equal to zero, the radial change constraint boundary is determined. When the radial change margin is less than zero, the radial change constraint is determined to be violated. The radial change margin corresponding to each radial interval is recorded respectively. The heat transfer margin, energy conservation margin, and radial variation margin of the collection area form a constraint set, wherein the heat transfer margin, energy conservation margin, and radial variation margin of the collection area are specifically as follows: The candidate parameter identifiers are used to associate the regional heat transfer margin corresponding to each partition pair, the energy conservation margin corresponding to the candidate total heat transfer, and the radial variation margin corresponding to each adjacent radial interval. The partition pair identifier is recorded for the regional heat transfer margin, the total energy constraint identifier is recorded for the energy conservation margin, and the adjacent radial interval identifier is recorded for the radial variation margin. The value and positive / negative status of each margin are also recorded. All margin data are arranged in a fixed order of regional heat transfer constraint, energy conservation constraint, and radial variation constraint to form the constraint set corresponding to the candidate parameter.

[0029] In this embodiment, obtaining the target parameters includes: Using the candidate parameters as the current parameters, a linear approximation relationship is established between the temperature residual and the constraint set within the trust region of the COBYLA algorithm, generating candidate parameter increments. Specifically, generating candidate parameter increments involves: According to the temperature measurement location identifier, read the temperature residual corresponding to the current candidate parameter, square and sum the temperature residuals, and divide the sum by the number of temperature measurement locations to obtain the target value of the temperature residual. Calculate the difference between the upper and lower limits of the value of each heat transfer parameter to be calibrated, and use 10% of the difference as the initial trust region radius of the corresponding parameter component. Centered on the current candidate parameter, increase the corresponding trust region radius along each parameter component to form the parameter dimension plus one interpolation parameter point. Match each interpolation parameter point with the temperature mapping rule to obtain the predicted temperature and calculate the target value of the temperature residual. At the same time, calculate the corresponding regional heat transfer margin, energy conservation margin, and radial variation margin. For each parameter component, use interpolation... The parameter difference between the parameter point and the current candidate parameter is used as the denominator, and the difference of the corresponding temperature residual target value is used as the numerator to obtain the target change coefficient. The difference of each constraint margin is used as the numerator to obtain the corresponding constraint change coefficient. The sum of the current temperature residual target value and each parameter increment multiplied by the corresponding target change coefficient is used as the target linear approximation value. The sum of the current constraint margin and each parameter increment multiplied by the corresponding constraint change coefficient is used as the constraint linear approximation value. The absolute value of each parameter increment is limited to not exceeding the corresponding trust region radius, and all constraint linear approximation values ​​are limited to not being less than zero. Among the parameter increments that meet the limitations, the parameter increment with the smallest target linear approximation value is selected as the candidate parameter increment. Based on the boundary markers, the regional heat transfer margin, energy conservation margin, and radial variation margin are compared before and after the incremental application of candidate parameters. The parameter components that touch the constraint boundary and their corresponding shrinkage magnitudes are determined, and a constraint margin shrinkage operator is constructed. Specifically, the construction of the constraint margin shrinkage operator is as follows: The incremental candidate parameter is added to the corresponding current parameter to form the incremental parameter. The regional heat transfer margin, energy conservation margin, and radial variation margin corresponding to the current parameter and the incremental parameter are calculated respectively. For the same constraint, if the margin before application is greater than zero and the margin after application is less than or equal to zero, the incremental candidate parameter is determined to have touched the corresponding constraint boundary. Similarly, if the margin before application is equal to zero and the margin after application is less than zero, the corresponding constraint boundary is also determined to have been touched. The association between parameter components and corresponding constraints is determined based on the parameter dimension, constraint type, and constraint boundary in the boundary markers. This is combined with the partition pairing identifier, overall energy constraint identifier, and adjacent radial interval identifier recorded in the constraint set to determine the contact point. The parameter components of the constraint boundary are defined as follows: When the pre-action margin is greater than zero, the allowable increment ratio is obtained by dividing the pre-action margin by the sum of the absolute values ​​of the pre-action margin and the post-action margin. When the pre-action margin is equal to zero, the allowable increment ratio is set to 0. The allowable increment ratio is obtained by multiplying the candidate parameter increment of the boundary parameter component by the allowable increment ratio. The corresponding shrinkage amplitude is obtained by subtracting the boundary allowable increment from the candidate parameter increment. When the same parameter component is associated with multiple boundary constraints, the allowable increment ratio of each boundary constraint is calculated separately, and the minimum allowable increment ratio is used to determine the corresponding shrinkage amplitude. The parameter component identifier, allowable increment ratio, and shrinkage amplitude are combined to form the constraint margin shrinkage operator. The constraint margin shrinkage operator is used to correct the increment of candidate parameters, shrinking the parameter components that touch the constraint boundary, and generating corrected candidate parameters. Specifically, the generation of corrected candidate parameters involves: Read the parameter component identifiers and corresponding shrinkage amplitudes in the constraint margin shrinkage operator, and match the parameter components in the candidate parameter increments one by one; the parameter components that do not match the boundary identifiers keep the original candidate parameter increments unchanged, and the parameter components that match the boundary identifiers have their original candidate parameter increments reduced by the corresponding shrinkage amplitudes to obtain the corrected parameter increments; add each corrected parameter increment to the corresponding parameter component in the current parameter to obtain the corrected heat transfer parameters to be calibrated, and arrange them in radial order according to the heat transfer control position to form corrected candidate parameters; Based on the radial residual assignment, the temperature residual corresponding to the parameter component is determined, and the search scale scaling is determined in conjunction with the corresponding constraint margin. A temperature residual partition scaling factor is constructed, and the trust region search scale of the corresponding parameter component in the next iteration is adjusted using the scaling factor. Specifically, the construction of the temperature residual partition scaling factor and the adjustment of the trust region search scale of the corresponding parameter component in the next iteration are as follows: According to the identifier of the heat transfer parameter to be calibrated, read the zone residuals in the radial residual assignment. Summate the zone residuals corresponding to all heat transfer parameters to be calibrated. Divide the zone residual corresponding to each heat transfer parameter to be calibrated by the summation result to obtain the temperature residual ratio. When the sum of all zone residuals is equal to zero, set the temperature residual ratio of each parameter component to zero. According to the zone pairing identifier and the adjacent radial interval identifier, read the regional heat transfer margin and radial variation margin associated with each parameter component, and read the energy conservation margin. When there is a value less than or equal to zero in the associated margin, set the constraint margin value of the corresponding parameter component to zero. When all associated margins are greater than zero, take the minimum value. The margin serves as the constraint margin value for the corresponding parameter component. The maximum value among all parameter component constraint margin values ​​is extracted, and the constraint margin value for each parameter component is divided by the maximum value to obtain the constraint margin ratio. When the maximum value is zero, the constraint margin ratio for each parameter component is set to zero. The temperature residual ratio is added to the constraint margin ratio and divided by 2, then added to 0.5 to obtain the temperature residual partition scaling factor for the corresponding parameter component, ensuring the temperature residual partition scaling factor is between 0.5 and 1.5. The current trust region search scale for each parameter component is multiplied by the corresponding temperature residual partition scaling factor to obtain the trust region search scale for the corresponding parameter component in the next iteration. The candidate parameters are re-matched to the temperature mapping rules, the temperature residuals and boundary markers are updated, and the constraint set is updated based on the revised candidate parameters. This process of candidate parameter generation, parameter shrinkage, and search scale adjustment is repeated until the preset convergence condition is met, yielding the target parameters. Specifically, the target parameters are: The process involves reading the parameter dimensions, parameter segmentation values, and parameter value ranges recorded in the temperature mapping rules. Each candidate parameter is compared with the parameter segmentation conditions to determine its corresponding rule partition. The predicted temperature at each measurement location is calculated using the corresponding local linear response parameters. The measured temperature is then subtracted from the predicted temperature to update the temperature residual. Next, the process reads the parameter dimensions, parameter segmentation values, and thermal boundary constraint states from the constraint boundary information. The magnitudes of the candidate parameter components and parameter segmentation values ​​are compared. The boundary markers of the corresponding parameter components are updated based on the regional heat transfer relationship, radial variation state, and energy conservation state. The regional heat transfer margin, energy conservation margin, and radial variation margin corresponding to the candidate parameters are recalculated, and an updated constraint set is formed according to the constraint type and associated identifier. Finally, the updated temperature at each measurement location is updated. The residuals are squared and summed. The summation result is divided by the number of temperature measurement locations to obtain the target value of the temperature residual for the current iteration. The target value of the temperature residual for the current iteration is subtracted from the target value of the temperature residual for the previous iteration, and the difference is divided by the target value of the temperature residual for the previous iteration to obtain the target value reduction ratio. When the regional heat transfer margin, energy conservation margin, and radial variation margin are all not less than zero, and the target value reduction ratio is less than 1%, the convergence condition is determined, and the current modified candidate parameter is determined as the target parameter. If the convergence condition is not met, the current modified candidate parameter is used as the new current parameter. The candidate parameter increment is generated according to the adjusted trust region search scale, the parameter shrinkage is performed, and the search scale is adjusted. When the number of iterations reaches 100, the parameter with the smallest temperature residual target value is selected from the iterative parameters that satisfy all constraint margins not less than zero as the target parameter.

[0030] In this embodiment, the continuous heat transfer coefficient distribution output after thermal simulation verification includes: Substituting the target parameters into the continuous distribution relationship of heat transfer coefficients, the heat transfer coefficients corresponding to the radial positions of the fire surface are determined, thus forming the target heat transfer coefficient distribution. Specifically, the formation of the target heat transfer coefficient distribution is as follows: According to the radial order of heat exchange control positions from smallest to largest, each heat exchange parameter to be calibrated in the target parameters is configured to the corresponding heat exchange control position. For any radial position between adjacent heat exchange control positions, the distance from the radial position to the two heat exchange control positions is calculated. The proportion of one distance to the sum of the two distances is used as the weight of the target heat exchange parameter corresponding to the other heat exchange control position, and the proportion of the other distance to the sum of the two distances is used as the weight of the target heat exchange parameter corresponding to this heat exchange control position. The two target heat exchange parameters are weighted and summed to obtain the heat transfer coefficient of the corresponding radial position. The first target heat transfer parameter is used between the inner radial boundary of the fire surface and the first heat exchange control position, and the last target heat transfer parameter is used between the last heat exchange control position and the outer radial boundary of the fire surface, forming a target heat transfer coefficient distribution covering the radial range of the fire surface. The target heat transfer coefficient distribution is loaded into the thermal simulation model. The simulation temperature is extracted and verified according to the temperature measurement location, and compared with the measured temperature to obtain the temperature verification result. Specifically, the extraction and comparison of the simulation temperature according to the temperature measurement location to obtain the temperature verification result is as follows: Based on the radial position corresponding to the center of the fire surface mesh element, the corresponding heat transfer coefficient is read from the target heat transfer coefficient distribution and configured into the fire surface mesh element of the thermal simulation model to form the combustion-side heat transfer boundary. The thermal simulation settings corresponding to the measured temperature acquisition steady-state condition are read, and the cylinder head material thermophysical parameters, cooling-side heat transfer boundary, combustion gas temperature boundary, and environmental heat transfer boundary are configured into the thermal simulation model to solve the cylinder head steady-state temperature field. The nearest thermal simulation mesh node is found according to the spatial coordinates of the temperature measurement location, and the mesh node temperature is read as the verification simulation temperature for the corresponding temperature measurement location. The absolute value of the difference between the verification simulation temperature and the measured temperature at each temperature measurement location is calculated and aggregated to form the absolute temperature deviation. All absolute temperature deviations are summed and divided by the number of temperature measurement locations to obtain the average absolute temperature deviation. The calibration results from each heat transfer coefficient calibration in the historical heat transfer coefficient calibration data of the same type of engine are read, and the average absolute temperature deviation corresponding to each calibration result is calculated. The maximum average absolute temperature deviation is determined as the temperature verification threshold. The average absolute temperature deviation and the comparison result between the average absolute temperature deviation and the temperature verification threshold are combined to form the temperature verification result. Based on the target heat transfer coefficient distribution and heat transfer constraint benchmark, the regional heat transfer relationship, the overall energy conservation relationship, and the radial variation relationship are checked respectively, and the constraint check results are obtained. Specifically, the constraint check results are as follows: Based on the equivalent radial position of each fire surface zone, the corresponding heat transfer coefficient is extracted from the target heat transfer coefficient distribution. The zone pairing relationship and heat transfer strength order in the regional heat transfer strength benchmark are read. The heat transfer coefficients of strong and weak heat transfer zones are compared group by group. If the heat transfer coefficient of a strong heat transfer zone is greater than that of a weak heat transfer zone, the corresponding region's heat transfer relationship is determined to meet the benchmark; otherwise, the corresponding region's heat transfer relationship is determined not to meet the benchmark. The surface temperature, heat transfer coefficient, and surface area of ​​each fire surface grid cell in the thermal simulation model are read. The difference between the combustion gas temperature and the grid cell surface temperature is calculated. The temperature difference, heat transfer coefficient, and surface area are multiplied to obtain the grid cell heat transfer. The heat transfer of all fire surface grid cells is summed to obtain the overall verification heat transfer. The overall verification heat transfer is compared with the lower and upper limits of the overall heat transfer energy benchmark. If the overall heat transfer is between the lower and upper limits, the overall energy conservation relationship is deemed to meet the benchmark; otherwise, it is deemed not to meet the benchmark. The fire surfaces are arranged in ascending order of equivalent radial position. The difference in heat transfer coefficient between adjacent fire surface zones is calculated and divided by the distance between the corresponding equivalent radial positions to obtain the radial heat transfer rate. This rate is compared with the lower and upper limits of the radial change benchmark for the corresponding radial interval. If the radial heat transfer rate is between the corresponding lower and upper limits, the radial change relationship is deemed to meet the benchmark; otherwise, it is deemed not to meet the benchmark. If all regional heat transfer relationships, the overall energy conservation relationship, and all radial change relationships meet the corresponding benchmarks, the constraint check result is determined to be passed; if any relationship does not meet the corresponding benchmark, the constraint check result is determined to be failed. When both the temperature verification results and the constraint verification results meet the preset verification conditions, the target heat transfer coefficient distribution is determined as the continuous heat transfer coefficient distribution of the cylinder head fire surface and output. Specifically, determining the target heat transfer coefficient distribution as the continuous heat transfer coefficient distribution of the cylinder head fire surface and outputting it is as follows: The average absolute temperature deviation and temperature verification threshold in the temperature verification results are read, and the constraint verification results are also read. If the average absolute temperature deviation is not greater than the temperature verification threshold and the constraint verification result is passed, the verification condition is determined to be met. The target heat transfer coefficient distribution is determined as the continuous heat transfer coefficient distribution of the cylinder head fire surface, and output according to the one-to-one correspondence between radial position and heat transfer coefficient. If the average absolute temperature deviation is greater than the temperature verification threshold or the constraint verification result is failed, the target heat transfer coefficient distribution is not output. The corresponding target parameter is re-used as the current candidate parameter. The temperature residual is updated using the difference between the verification simulation temperature and the measured temperature. The regional heat transfer margin, energy conservation margin, and radial variation margin are recalculated based on the target parameters to form a constraint set. The incremental generation of candidate parameters, parameter shrinkage, and search scale adjustment are continued to obtain new target parameters and thermal simulation verification is performed again until the average absolute temperature deviation is not greater than the temperature verification threshold and the constraint verification result is passed.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to the calibration of the heat transfer coefficient of the cylinder head fire surface of a certain type of high-strength diesel engine. The research object was a six-cylinder diesel engine developed for high power density power units. The rated power steady-state operating condition was selected as the calibration condition, with an engine speed of 2100 r / min and a single-cylinder average effective pressure of 2.15 MPa. Combining cylinder head structural design data, combustion chamber layout data, and thermal load test data, the fire surface was divided into 5 heat transfer zones, with areas of 2860 mm², 3540 mm², 4210 mm², 4960 mm², and 5380 mm², respectively. The center of the fuel injector was used as a unified radial reference.

[0032] The radial distance from the boundary of each zone to the center of the fuel injector was calculated based on the fire surface coordinate system. The equivalent radial positions corresponding to the five zones were 18.6 mm, 31.4 mm, 45.7 mm, 60.9 mm, and 77.3 mm, respectively, and these five equivalent radial positions were set as heat exchange control positions. Historical thermal simulation and thermal balance test data of the same type of diesel engine were read, and the initial heat transfer coefficients of the five heat exchange control positions were 2860 W / (m²·K), 2640 W / (m²·K), 2310 W / (m²·K), 1980 W / (m²·K), and 1710 W / (m²·K), respectively. The range of each parameter value was determined by the minimum and maximum values ​​in the previous valid calibration results.

[0033] In the continuous distribution construction process, the heat transfer parameters to be calibrated are configured in radial order from smallest to largest according to the heat transfer control positions. Between adjacent control positions, the two heat transfer parameters are weighted according to the radial distance ratio, and the heat transfer coefficient corresponding to any radial position is calculated. The first heat transfer parameter is used within the first heat transfer control position, and the last heat transfer parameter is used outside the last heat transfer control position, thus forming a continuous heat transfer coefficient distribution covering the entire radial range of the fire surface. Compared to directly assigning fixed heat transfer coefficients to each of the five fire surface zones, continuous distribution can avoid obvious abrupt changes in thermal boundaries at the boundaries of adjacent zones.

[0034] To establish parameter samples, for each heat transfer parameter to be calibrated, the lower limit, the average of the lower limit and the initial value, the initial value, the average of the initial value and the upper limit, and the upper limit were selected as disturbance values ​​within the value range. These disturbance values ​​were then combined across five heat transfer control positions, resulting in 3125 candidate parameter combinations. The absolute deviation of each heat transfer parameter to be calibrated relative to its corresponding initial value was calculated for each candidate parameter combination. Each absolute deviation was normalized by dividing by the difference between the upper and lower limits of the corresponding parameter value. The sum of the five normalized deviations yielded the combined disturbance amplitude. The 3125 candidate parameter combinations were arranged in ascending order of combined disturbance amplitude, and the sequence was divided into 625 continuous intervals, each containing five candidate parameter combinations. The candidate parameter combination located in the middle of each interval was selected as the parameter sample, resulting in a total of 625 valid parameter samples. Each parameter sample was substituted with a continuous distribution relationship and loaded into the cylinder head thermal simulation model. The thermal simulation uses the cylinder head material thermal properties, coolant side heat transfer boundary, combustion gas temperature boundary and environmental heat transfer boundary that have been verified by experiments in the existing engine thermal analysis process of the research institute, to ensure that the calculation conditions are consistent with the cylinder head steady-state temperature measurement test.

[0035] Eight temperature measurement points were arranged on the cylinder head. The measured temperatures obtained from the steady-state test were 214.6℃, 228.3℃, 241.7℃, 253.4℃, 247.9℃, 232.8℃, 218.5℃, and 205.1℃, respectively. Thermal simulation was performed on 625 sets of parameter samples. The simulated temperatures of the eight temperature measurement points were extracted for each set, and the heat transfer of all fire surface grid units was accumulated to obtain the overall heat transfer of the corresponding fire surface. Based on the historical calibration results of the same type of diesel engine in the research institute, the stronger heat transfer of the radial inner zone than the adjacent outer zone was used as the benchmark for the regional heat transfer strength relationship. 74.5kW to 82.0kW was taken as the allowable range of overall heat transfer energy. The upper and lower limits of the heat transfer change rate of each radial zone were determined based on historical data that passed temperature verification and thermal balance verification.

[0036] For 625 sets of samples, regional heat transfer relationship states, radial variation states, and energy conservation states were generated respectively. Parameter samples, constraint states, and simulation temperatures were then input into the improved Cubist model. Training and validation samples were divided in an 8:2 ratio, with 500 sets used as training samples and 125 sets as validation samples, maintaining a relatively consistent proportion of different constraint state combinations in both types of samples. The thermal boundary constraint rule generation module traversed candidate segmentation positions according to the dimension of the parameters to be calibrated, filtering out segmentation positions that could not distinguish thermal boundary states using the differences among the three types of constraint states, and determining the rule segmentation positions based on the fitting deviation of the simulation temperature, ultimately forming 24 effective rule partitions.

[0037] The regular linear regression module establishes local linear response relationships between five heat transfer parameters to be calibrated and the simulated temperatures at eight temperature measurement locations within each regular partition. The boundary residual committee module sets up five committees to assign residual correction weights to samples near the boundary of heat transfer, energy conservation, and radial variation constraints, and uses weighted temperature prediction residuals to progressively correct the committee training responses. After training, the average absolute temperature prediction bias for the 125 validation samples was 1.84℃, the average absolute prediction bias for the constraint boundary samples was 1.71℃, and the average absolute prediction bias for the boundary samples of the ordinary Cubist model on the same validation samples was 3.12℃.

[0038] Five initial heat transfer parameters were used as the current candidate parameters for the COBYLA algorithm, and the initial trust region search scale for each parameter component was set to 10% of the width of the corresponding value range. The target value of the temperature residual was calculated based on the temperature residuals at eight temperature measurement locations, and a linear approximation relationship between the target value and three types of constraint margins was established within the trust region to generate candidate parameter increments. In the 7th iteration, the heat transfer margin corresponding to the 2nd heat transfer control location changed from 126 W / (m²·K) to -18 W / (m²·K). The constraint margin reduction operator determined that the corresponding parameter component touched the constraint boundary, calculated the allowable increment ratio to be 0.875, and reduced the original candidate increment of 160 W / (m²·K) to 140 W / (m²·K).

[0039] During the search scale adjustment process, the partition residuals corresponding to the five parameter components were 8.4℃, 6.7℃, 5.1℃, 3.9℃, and 2.8℃, respectively. The temperature residual partition scaling factors were calculated by combining the regional heat transfer margin, energy conservation margin, and radial variation margin, resulting in five scaling factors of 1.28, 1.17, 1.03, 0.86, and 0.72. These factors were then applied to the trust region search scale of the corresponding parameter components in the next round, ensuring that the radial region with larger temperature deviations received a larger search scale, while the parameter components closer to the physical constraint boundary adopted a smaller search scale.

[0040] After 26 iterations, the regional heat transfer margin, energy conservation margin, and radial variation margin were all no less than 0, and the decrease in the target value of the temperature residual was less than 1% for two consecutive rounds. Five target parameters were obtained: 3146 W / (m²·K), 2818 W / (m²·K), 2437 W / (m²·K), 2049 W / (m²·K), and 1736 W / (m²·K). These target parameters were reconstructed into a target heat transfer coefficient distribution and loaded into the cylinder head thermal simulation model. The verification simulation temperatures at eight temperature measurement locations were 215.3℃, 227.5℃, 242.1℃, 252.6℃, 248.4℃, 233.1℃, 217.9℃, and 205.7℃, with an average absolute temperature deviation of 0.59℃.

[0041] The final thermal simulation yielded an overall heat transfer of 78.8 kW for the fire surfaces. The heat transfer strength relationships among the five fire surface zones all conformed to historical calibration patterns, and the heat transfer change rates of adjacent radial intervals were all within acceptable limits. The constraint verification result was passed. Compared to the existing manual zone adjustment method used in the research institute, 53 thermal simulations were required, with a final average absolute temperature deviation of 2.47℃. When using the COBYLA algorithm without introducing constraint margin shrinkage and temperature residual zone scaling, 41 iterations were required, with a final average absolute temperature deviation of 1.36℃ and two instances of radial change constraint out-of-bounds errors.

[0042] As can be seen from the embodiments, the present invention can be integrated with the existing cylinder head thermal analysis, thermal balance test and structural design process of engine research institutions. Under the condition that the heat transfer relationship of the diesel engine fire surface area, the overall energy conservation and the radial heat transfer change law meet the engineering requirements, the accuracy of heat transfer coefficient calibration is improved, the number of repeated thermal simulations is reduced, and the stability of parameter search near the constraint boundary is enhanced. It is suitable for R&D scenarios such as cylinder head thermal boundary calibration, thermal load matching and thermal design optimization of high power density diesel engines.

[0043] The above description is only a preferred embodiment 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 method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm, characterized in that, include: Obtain the geometric information of the firepower surface partition, partition area, fuel injector center position, measured temperature, temperature measurement location, and heat transfer constraint benchmark; Determine the radial range and equivalent radial position of each fire zone, and establish the continuous distribution relationship of heat transfer coefficients and parameters to be calibrated; The parameters to be calibrated are disturbed to generate parameter samples, the corresponding heat transfer distribution is loaded into the thermal simulation model to obtain the simulation temperature, and the constraint state is generated according to the constraint benchmark. The Cubist model is improved by inputting parameter samples, constraint states and simulation temperatures to generate temperature mapping rules and boundary markers. The improved Cubist model includes a thermal boundary constraint rule generation module, a rule linear regression module and a boundary residual committee module. Using the parameters to be calibrated as candidate parameters, the predicted temperature is obtained according to the temperature mapping rule and compared with the measured temperature to generate temperature residuals. The temperature measurement position and the equivalent radial position are matched to form the radial residual assignment. Based on candidate parameters and constraint benchmarks, regional heat transfer, energy conservation, and radial variation margins are determined to form a constraint set; The candidate parameters are iteratively optimized using the COBYLA algorithm, which introduces a constraint margin shrinkage operator and a temperature residual partition scaling factor. The boundary components are shrunk according to the boundary marker and constraint set, and the search scale is adjusted according to the residual assignment to obtain the target parameters. Substitute the target parameters into the continuous distribution relationship of heat transfer coefficients, and output the continuous heat transfer coefficient distribution after thermal simulation verification.

2. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The geometric information of the fire surface partitions includes the boundary position, contour information and adjacency relationship of each partition; the partition area is the surface area of ​​the fire surface corresponding to each partition; the injector center position is the position of the injector center in the fire surface coordinate system; the measured temperature is the temperature data of each temperature measuring point under the steady-state thermal condition of the cylinder head, and the temperature measuring position is the spatial position of the corresponding temperature measuring point in the cylinder head; the heat transfer constraint benchmark includes the heat transfer strength relationship benchmark between each partition, the overall heat transfer energy benchmark of the fire surface, the heat transfer temperature difference benchmark, and the radial variation benchmark of the heat transfer coefficient.

3. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The establishment of the continuous distribution relationship of the heat transfer coefficient and the parameters to be calibrated includes: Using the center position of the fuel injector as the radial reference, determine the radial distance corresponding to the boundary of the firepower surface zone, and form the radial range of each firepower surface zone; The centroid of the area is determined based on the boundary position and area of ​​the fire zone, and the radial position of the centroid of the area relative to the center of the fuel injector is determined as the equivalent radial position of the corresponding fire zone. Set the heat exchange control position according to the equivalent radial position, and configure the heat exchange parameters to be calibrated for the heat exchange control position; Based on the heat exchange control position and the heat exchange parameters to be calibrated, a distribution relationship of heat transfer coefficients that continuously varies with radial position is established, and the heat exchange parameters to be calibrated are combined into calibration parameters.

4. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The step of generating constraint states based on constraint benchmarks includes: The parameters to be calibrated are perturbed within a preset range to generate multiple sets of parameter samples, and the continuous distribution of heat transfer coefficients corresponding to the parameter samples is determined. The heat transfer coefficient is continuously distributed and loaded into the thermal simulation model. Thermal simulation is performed, and the corresponding simulation temperature is extracted according to the temperature measurement location. The overall heat transfer of the fire surface is also extracted. By comparing the heat transfer coefficients corresponding to the parameter samples with the heat transfer constraint benchmark according to the radial order of the fire surface partition, the regional heat transfer relationship state and radial variation state are determined. The overall heat exchange of the firepower surface is compared with the overall heat exchange energy benchmark to determine the energy conservation state. The heat exchange relationship state, radial change state, and energy conservation state of the collection area form a constraint state.

5. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The generated temperature mapping rules and boundary markers include: The rule generation, rule-based linear regression, and committee prediction process based on the Cubist model are as follows: In the rule generation process, a candidate segmentation and screening process based on the thermal boundary constraint state is added to form a thermal boundary constraint rule generation module; the linear regression processing corresponding to the rules forms a rule-based linear regression module; in the committee prediction process, a residual correction weight allocation process based on the constraint boundary margin is added to form a boundary residual committee module; the output of the thermal boundary constraint rule generation module is passed to the rule-based linear regression module, and the output of the rule-based linear regression module is passed to the boundary residual committee module, forming an improved Cubist model. Input the parameter samples, constraint states, and simulation temperatures into the thermal boundary constraint rule generation module. Use the regional heat transfer relationship state, radial change state, and energy conservation state as thermal boundary constraint states. Traverse the candidate cutting positions of the parameter samples, filter the candidate cutting positions based on the differences in thermal boundary constraint states, and determine the cutting positions according to the fitting deviation of the simulation temperature to generate rule partitions and constraint boundary information. Input the rule partition, parameter samples and simulation temperature into the rule linear regression module, establish the local linear response relationship between the parameter to be calibrated and the simulation temperature within the rule partition, generate the initial temperature mapping rule, and compare the corresponding predicted temperature with the simulation temperature to generate the temperature prediction residual. The constraint boundary information, constraint state, and temperature prediction residual are input into the boundary residual committee module. The boundary samples are determined based on the margin between the constraint state and the constraint boundary. The residual correction weights are allocated according to the constraint boundary margin. The training response of the committee is corrected using the weighted temperature prediction residual, and temperature mapping rules and boundary labels are generated. The improved Cubist model was trained using parameter samples, constraint states, and simulation temperatures. The parameters for rule segmentation, local linear response, and committee residual correction were optimized to obtain the trained improved Cubist model.

6. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 5, characterized in that, The training of the improved Cubist model using parameter samples, constraint states, and simulation temperatures includes: Based on the correspondence between parameter samples, the constraint states and simulation temperatures are associated, and the associated samples are divided into training samples and validation samples. The training samples are input into the thermal boundary constraint rule generation module. Candidate segments are selected based on the regional heat transfer relationship, radial change state, and energy conservation state. The rule segmentation position is determined according to the simulation temperature fitting deviation corresponding to the candidate segmentation, and the training rule partition and constraint boundary information are obtained. The training rule partitions and corresponding training samples are input into the rule linear regression module. The parameters to be calibrated are used as regression variables and the simulated temperature is used as the target response. The local linear response parameters corresponding to the rule partitions are updated to obtain the training temperature mapping rules and temperature prediction residuals. The temperature prediction residual, constraint state, and constraint boundary information are input into the boundary residual committee module. The residual correction weights are allocated according to the margin between the training samples and the constraint boundary. The training response of the next committee is updated using the weighted temperature prediction residuals, and the committee residual correction parameters are obtained iteratively. The validation sample is input into the improved Cubist model that completes the current iteration to obtain the validation predicted temperature. The boundary sample in the validation sample is determined according to the constraint boundary information. The temperature prediction deviation between the validation sample and the boundary sample is calculated respectively. When the temperature prediction deviation of the two types meets the preset convergence condition, the trained improved Cubist model is obtained.

7. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The process of obtaining the predicted temperature according to the temperature mapping rule and comparing it with the measured temperature to generate a temperature residual, and matching the temperature measurement location and the equivalent radial location to form the radial residual assignment, includes: The candidate parameters are matched with the temperature mapping rules to determine the rule partition to which the candidate parameters belong, and the predicted temperature of the temperature measurement location is obtained according to the corresponding temperature mapping rule. According to the correspondence of temperature measurement locations, the predicted temperature is compared with the actual temperature one by one to determine the direction and amount of temperature deviation and generate temperature residual. Using the center position of the fuel injector as the radial reference, determine the radial position of the temperature measurement position, and match it with the equivalent radial position of the fire surface zone to determine the radial zone corresponding to the temperature residual. The temperature residuals are associated with the corresponding radial partitions to the parameter to be calibrated, and the temperature residuals associated with the same parameter to be calibrated are aggregated to form radial residual assignment.

8. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The formation of the constraint set includes: Substitute the candidate parameters into the continuous distribution relationship of heat transfer coefficients to determine the heat transfer coefficients corresponding to the fire surface zones, and compare them with the benchmark of regional heat transfer strength to generate regional heat transfer margin. The overall heat transfer is determined by combining the heat transfer coefficient, zone area and heat transfer temperature difference benchmark corresponding to the fire surface zone, and compared with the overall heat transfer energy benchmark to generate energy conservation margin. The changes in heat transfer coefficients of adjacent fire surface zones are compared according to the equivalent radial position and compared with the radial change benchmark to generate radial change margin. The heat transfer margin, energy conservation margin, and radial variation margin of the aggregated area are used to form a constraint set.

9. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The obtained target parameters include: Using the candidate parameters as the current parameters, a linear approximate relationship is established between the temperature residual and the constraint set within the trust region of the COBYLA algorithm, generating the candidate parameter increment; Based on the boundary markers, the regional heat transfer margin, energy conservation margin, and radial variation margin before and after the incremental action of candidate parameters are compared to determine the parameter components that touch the constraint boundary and their corresponding shrinkage magnitude, and construct the constraint margin shrinkage operator. The constraint margin shrinkage operator is used to correct the increment of the candidate parameters, shrinking the parameter components that touch the constraint boundary to generate corrected candidate parameters. The temperature residuals corresponding to the parameter components are determined based on the radial residual assignment, and the search scale scaling amount is determined in combination with the corresponding constraint margin. A temperature residual partition scaling factor is constructed, and the trust region search scale of the corresponding parameter components in the next iteration is adjusted using the scaling factor. The candidate parameters are corrected and rematched with the temperature mapping rules, the temperature residuals and boundary markers are updated, and the constraint set is updated according to the corrected candidate parameters. The process of generating candidate parameters, shrinking parameters and adjusting the search scale is repeated until the preset convergence conditions are met, and the target parameters are obtained.

10. The method for calibrating the heat transfer coefficient of the cylinder head firing surface based on a multi-constraint optimization algorithm according to claim 1, characterized in that, The continuous heat transfer coefficient distribution output by thermal simulation verification includes: Substitute the target parameters into the continuous distribution relationship of heat transfer coefficients to determine the heat transfer coefficients corresponding to the radial positions of the fire surface, thus forming the target heat transfer coefficient distribution. The target heat transfer coefficient distribution is loaded into the thermal simulation model, the simulation temperature is extracted and verified according to the temperature measurement location, and the temperature verification result is obtained by comparing it with the measured temperature. Based on the target heat transfer coefficient distribution and heat transfer constraint benchmark, the regional heat transfer relationship, the overall energy conservation relationship and the radial variation relationship are checked respectively to obtain the constraint check results; When both the temperature verification result and the constraint verification result meet the preset verification conditions, the target heat transfer coefficient distribution is determined as the continuous heat transfer coefficient distribution of the cylinder head fire surface and output.

Images