Combustion chamber outlet temperature modeling prediction method

Abstract

The invention discloses a combustion chamber outlet temperature modeling prediction method, and relates to the technical field of data processing, and the method comprises the steps: obtaining the sample outlet temperature data of a combustion chamber outlet, and the sample condition variable data influencing the sample outlet temperature data; training a plurality of machine learning models based on the sample outlet temperature data and the sample condition variable data; reserving the machine learning models of which the decision coefficients are greater than a preset threshold in the trained machine learning models to form a joint prediction model; wherein the predicted temperature output by the combined prediction model is a comprehensive value of the output temperature of each machine learning model included in the combined prediction model, so that the accuracy of predicting the outlet temperature of the combustion chamber is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular, to a combustion chamber outlet temperature modeling and prediction method. BACKGROUND

[0002] The combustion chamber is a core thermal component of an aero-engine and a gas turbine, and the outlet temperature thereof is a key parameter for measuring performance, affecting component life, and determining thermal efficiency. The traditional calculation method of the outlet temperature is mainly through a simple fitting formula or chemical equilibrium calculation.

[0003] The simple fitting formula is usually an empirical formula obtained through linear or nonlinear regression statistical methods based on a large amount of experimental data, and establishes the relationship between the combustion chamber outlet temperature and the main parameters such as fuel flow and air flow. This formula is relatively simple to calculate and can estimate the outlet temperature to a certain extent. However, since it is fitted based on specific experimental data, the applicability and accuracy of the formula are often greatly limited under different combustion chamber structures, fuel types, etc., and its generalization ability is poor, and the precision is low when facing complex working conditions. This method is only fitted based on specific experimental data, lacks targeted analysis of complex working conditions, and is difficult to comprehensively cover all kinds of detailed differences and technical variables, resulting in great limitations in applicability and accuracy when facing different combustion chamber structures, fuel types, and complex and variable operating conditions. SUMMARY

[0004] The present application aims to at least partially solve one of the technical problems in the related art. To this end, one object of the present application is to provide a combustion chamber outlet temperature modeling and prediction method to improve the accuracy of the prediction of the combustion chamber outlet temperature.

[0005] According to a first aspect of an embodiment of the present application, a combustion chamber outlet temperature modeling and prediction method is provided, the method comprising: obtaining sample outlet temperature data of a combustion chamber outlet and sample condition variable data affecting the sample outlet temperature data; training a plurality of machine learning models based on the sample outlet temperature data and the sample condition variable data; retaining machine learning models with a determination coefficient greater than a preset threshold value in the trained machine learning models to constitute a joint prediction model; wherein the predicted temperature output by the joint prediction model is a comprehensive value of the output temperatures of each machine learning model included in the joint prediction model; inputting the condition variables of the combustion chamber outlet temperature into the joint prediction model to obtain the predicted temperature output by the joint prediction model.

[0006] Optionally, the sample condition variable data is obtained in the following manner: obtaining original condition variable data of the combustion chamber outlet; performing Gaussian filtering on the original condition variable data to obtain first condition variable data; calculating correlations between sample condition variables in the first condition variable data, and if there are a first sample condition variable and a second sample condition variable with a correlation greater than a preset correlation threshold, removing the sample condition variable with a smaller correlation with the sample outlet temperature data from the first sample condition variable and the second sample condition variable; obtaining the sample condition variable data based on the sample condition variables retained in the first condition variable data.

[0007] Optionally, the method further comprises: if the first sample condition variable is fuel flow, and if there is a second sample condition variable with a correlation greater than a preset correlation threshold with the fuel flow, removing the second sample condition variable.

[0008] Optionally, the obtaining of the sample outlet temperature data of the combustion chamber outlet and the sample condition variable data affecting the sample outlet temperature data comprises: respectively adopting time interval sampling, sliding time window sampling and change rate sampling to obtain sample outlet temperatures and sample condition variables of each sampling point in a preset time period; obtaining the sample outlet temperature data and the sample condition variable data based on the normalization results of the obtained sample outlet temperatures and sample condition variables.

[0009] Optionally, the plurality of machine learning models comprises a linear regression model, a ridge regression model, a lasso regression model, an elastic net model, a K-nearest neighbor regression model, a support vector regression model, a decision tree model, a random forest model, a gradient boosting tree model, an XGBoost model, an extreme random tree model and a CatBoost model.

[0010] Optionally, the comprehensive value is obtained in the following manner: after obtaining output temperatures of each machine learning model included in the joint prediction model, calculating a mean value and a standard deviation based on the output temperatures, and removing output temperatures with abnormal values according to the mean value and the standard deviation; calculating an average value of the remaining output temperatures as the comprehensive value.

[0011] Optionally, for each machine learning model, the determination coefficient is represented by a proportion value of a residual sum of squares to a total sum of squares, and the preset threshold is 0.95.

[0012] It can be seen from the above that, according to the scheme provided by the embodiment of the present application, in the case of training multiple machine learning models using sample outlet temperature data and sample condition variable data, the machine learning model with a determination coefficient greater than a preset threshold, that is, the machine learning model with better prediction ability, is reserved, and a joint prediction model is formed, so that the joint prediction model can improve the accuracy of the prediction of the outlet temperature of the combustion chamber. Moreover, the joint prediction model can include multiple machine learning models, can adapt to temperature prediction in multiple scenarios by exerting the different advantages of various types of models, and improves the generalization ability of the model.

[0013] Additional aspects and advantages of the present application will be set forth in part in the description that follows, and in part will become apparent to those skilled in the art upon examination of the following description and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 is a flowchart of a first combustion chamber outlet temperature modeling and prediction method provided by an embodiment of the present application; Figure 2 is a flowchart of a second combustion chamber outlet temperature modeling and prediction method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0015] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0016] In an embodiment of the present application, referring to Figure 1 , a combustion chamber outlet temperature modeling and prediction method is provided, referring to Figure 1 , the method comprises the following steps S101-S104.

[0017] S101: Obtain sample outlet temperature data of the combustion chamber outlet and sample condition variable data affecting the sample outlet temperature data; S102: Train multiple machine learning models based on the sample outlet temperature data and the sample condition variable data; S103: Reserve the machine learning model with a determination coefficient greater than a preset threshold among the trained machine learning models, and form a joint prediction model; wherein the predicted temperature output by the joint prediction model is a comprehensive value of the output temperatures of each machine learning model included in the joint prediction model; S104: Input the condition variables of the combustion chamber outlet temperature into the joint prediction model, and obtain the predicted temperature output by the joint prediction model.

[0018] The sample outlet temperature data records the combustion chamber outlet temperature obtained at different sampling times, thus obtaining full-scale time-series temperature data to form a dataset for training machine learning. The sample condition variable data consists of data formed by condition variables affecting the outlet temperature at the corresponding sampling times, specifically as follows: Figure 2 As shown in step one, the sample condition variables can include air temperature, fuel temperature, air pressure, fuel flow rate, etc. The sample outlet temperature data records the outlet temperature.

[0019] In the solution provided by this invention, time series sampling processing is performed on the data to solve the data similarity problem. This is because the time interval between data collection is too short, and the fluctuation of the obtained data is small, so the similarity of the data is too high, that is, the difference between two adjacent data is too small. When building a machine learning model, overfitting may occur due to the data being too similar, and the influence of each feature variable on the target variable, outlet temperature, cannot be well analyzed. Therefore, the data needs to be processed before building the model to reduce the similarity between the data and improve the rationality of the data.

[0020] In step S102, the sample condition variable data is input into a preset number of machine learning models to obtain the predicted outlet temperature output by the machine learning; the preset machine learning model is trained based on the difference between the predicted outlet temperature and the sample outlet temperature to obtain a model for temperature prediction.

[0021] Specifically, various learning models can include 12 types, such as Linear Regression, Ridge Regression, Lasso Regression, Elastic Network (EN), K-Nearest Neighbor (KNN), Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), XGBoost (XGB), Extremely Randomized Trees (ERT), and CatBoost (CAT), all of which use regression algorithms for model training.

[0022] During training, machine learning models can learn complex mapping relationships between parameters from large amounts of data, thereby enabling high-precision predictions. Furthermore, building machine learning models is very rapid; they can be trained in minutes, and the trained models can be saved after training. When needed, the trained models can be directly called for real-time predictions.

[0023] Taking the linear regression model as an example, during the training process, the sample condition variable data is input into the linear regression model, and the best fitting line is found to minimize the error between the output temperature predicted by the model and the actual temperature in the sample outlet temperature data.

[0024] After training, in one embodiment, for each machine learning model, the coefficient of determination is the ratio of the sum of squared residuals to the sum of squared total squares. This indicates that the preset threshold is 0.95. In other words, it retains... The machine learning model is used as a component of the joint prediction model. The threshold is a manually set value, and in other embodiments, it may be set to 0.9, 0.85, etc.

[0025] In step 103, the above-mentioned comprehensive value can be directly calculated as the average value of the output temperature of each machine learning model included in the joint prediction model.

[0026] Alternatively, in one embodiment, the composite value is obtained as follows: After obtaining the output temperatures of each machine learning model included in the joint prediction model, the mean and standard deviation of each output temperature are calculated, and the output temperatures with outliers are removed based on the mean and standard deviation. The average of the remaining output temperatures is calculated as the composite value.

[0027] In this embodiment, specifically, outlier predicted values ​​that deviate from the confidence interval can be removed using the 3σ principle. After calculating the mean (μ) and standard deviation (σ) of the data, if the output temperature is less than or equal to μ-3σ or greater than or equal to μ+3σ, it is determined to be an outlier. After removing the outlier predictions, the average of the remaining output temperatures is used to obtain a comprehensive value, which is then used as the final predicted value for the outlet temperature.

[0028] In addition to the average, statistical values ​​such as the median and mode can also be used as a composite value, and this embodiment of the invention does not limit this.

[0029] The condition variables in step S104 are the same type of condition variables as the sample condition variables in the aforementioned embodiments, the difference being the collection time; the former is used in the application phase of the model, while the latter is used in the training phase. Following this approach, after using the trained machine learning model, only the condition variables need to be input to obtain the predicted temperature output by the joint prediction model. Compared to related technologies that use chemical equilibrium calculations and solve chemical equilibrium equations to determine the composition and temperature of combustion products, this method is more efficient and reduces the time required when dealing with complex operating conditions.

[0030] This embodiment corresponds to Figure 2 In step three, the application deployment phase, data collection acquires conditional variables. Data processing can be similar to the aforementioned embodiment, performing Gaussian filtering. Model invocation involves inputting the conditional variables into the joint prediction model for processing, obtaining the predicted temperature output by the joint prediction model as the output result.

[0031] As can be seen from the above, according to the solution provided in the embodiments of the present invention, when training multiple machine learning models using sample outlet temperature data and sample condition variable data, by retaining machine learning models with a determination coefficient greater than a preset threshold, i.e., machine learning models with better predictive ability, and constructing a joint prediction model, the joint prediction model can improve the accuracy of predicting the combustion chamber outlet temperature. Furthermore, the joint prediction model can include multiple machine learning models, which can leverage the different advantages of various types of models to adapt to temperature prediction in various scenarios, thus improving the model's generalization ability.

[0032] Therefore, compared to simply using fitting formulas to obtain the relationship between conditional variables and outlet temperature, the above-mentioned joint prediction model can comprehensively cover various detailed differences and conditional variables, improving its applicability and accuracy when facing different combustion chamber structures, fuel types, and complex and ever-changing operating conditions.

[0033] In one embodiment of the present invention, the sample condition variable data is obtained in the following manner: Obtain the raw condition variable data at the combustion chamber outlet; The original condition variable data is processed by Gaussian filtering to obtain the first condition variable data; Calculate the correlation between each sample condition variable in the first condition variable data. If there are first sample condition variables and second sample condition variables with a correlation greater than the preset correlation threshold, then remove the sample condition variables from the first sample condition variables and second sample condition variables that have a low correlation with the sample outlet temperature data. Sample condition variable data are obtained based on the sample condition variables retained in the first condition variable data.

[0034] In this embodiment, the conditional variables measured directly at the combustion chamber outlet are the original conditional variables, such as the real-time measured air temperature and fuel flow rate, which constitute the original conditional variable data.

[0035] Gaussian filtering can reduce noise in raw conditional variable data. For any variable, such as air temperature, the Gaussian kernel G(x) is calculated as follows:

[0036] Where G(x) is the Gaussian kernel at position The weight at each sampling point is x, which is the distance between the air temperature and the average air temperature at any sampling point. It is the standard deviation of the Gaussian kernel.

[0037] After obtaining the Gaussian kernel, the weighted average of the temperature values ​​in the neighborhood of each sampling point is calculated and used to replace the original temperature, thus achieving noise reduction. This process is repeated for each original condition variable to obtain the first condition variable data.

[0038] The correlation can be obtained by calculating the Pearson correlation coefficient between each pair of sample condition variables in the first condition variable data. If the correlation between two sample condition variables is greater than 0.9, that is, the absolute value of the Pearson correlation coefficient is greater than 0.9, the feature with a higher correlation to the outlet temperature is retained when building the model for that period.

[0039] In other words, if the correlation between two feature variables is greater than 0.9 at a certain moment, one is designated as the first sample condition variable and the other as the second sample condition variable. The Pearson correlation coefficients of the first and second sample condition variables and the sample outlet temperature data at that moment are calculated separately to obtain their correlation with the sample outlet temperature. Condition variables with higher correlations are retained. For example, if the absolute value of the Pearson correlation coefficient between the first sample condition variable and the sample outlet temperature is 0.8, and the absolute value of the Pearson correlation coefficient between the second sample condition variable and the sample outlet temperature is 0.7, then the second sample condition variable is removed.

[0040] The final retained sample condition variables constitute the sample condition variable data. This effectively reduces noise in the dataset after Gaussian filtering, and the condition variables are filtered through correlation calculations, effectively removing redundant features. The resulting machine learning model predicts the coefficient of determination. The generalization ability of the model is improved by fusing high-precision models, with a success rate greater than 95%.

[0041] In one scenario, if the first sample condition variable is fuel flow rate, and there exists a second sample condition variable whose correlation with fuel flow rate is greater than a preset correlation threshold, then the second sample condition variable is removed.

[0042] In other words, fuel flow rate is always retained when a conditional variable that is highly correlated with fuel flow rate exists. Fuel flow rate is an important characteristic of combustion in the combustion chamber and has a significant correlation with outlet temperature; therefore, using it as a sample conditional variable for model training is more accurate.

[0043] like Figure 2 As shown, the data processing in step one includes data integration, which involves matching and integrating the conditional variables of each sample according to the time series. Specifically, this involves checking whether the timestamps of each conditional variable are consistent to prevent time series disorder. At the same time, it involves checking whether there are missing values. If so, linear interpolation is used to fill in the missing values. Gaussian noise reduction is performed by Gaussian filtering. Correlation analysis and feature screening are performed by removing conditional variables that are less correlated with the sample outlet temperature data through correlation calculation, while retaining conditional variables with high correlation.

[0044] In one embodiment, see Figure 2 In the second step of the data sampling process, time interval sampling, sliding time window sampling, and rate of change sampling are used to obtain the sample outlet temperature and sample condition variables of each sampling point within a preset time period. Sample outlet temperature data and sample condition variable data are obtained based on the normalization results of the acquired sample outlet temperature and sample condition variable.

[0045] Time interval sampling refers to sampling the condition variable and the sample outlet temperature at fixed time intervals; sliding time window sampling involves setting a predetermined sliding time window and performing sliding sampling according to a set step size; rate of change sampling determines whether to sample based on the degree of change between adjacent data points. Specifically, during sampling, the first point is placed in the sampling pool, and then the Euclidean distance from the next point to the last point in the sampling pool is calculated. If the Euclidean distance is greater than the sampling threshold, the point is placed in the sampling pool; otherwise, if the Euclidean distance is less than the threshold, the point is not sampled, and the calculation of the distance from the next point to the last point in the sampling pool continues.

[0046] Euclidean distance is obtained based on the degree of change of sample condition variables between different sampling points, such as the temperature difference of air temperature and the flow rate difference of fuel at different sampling points. The result is the square root of the sum of the squares of the differences of each sample condition variable.

[0047] Three sampling methods transform the raw data from the sampling period into datasets using three different sampling methods. The sampled datasets are then normalized to obtain datasets containing sample outlet temperature data and sample condition variable data. During model training, the dataset can be randomly divided into training and test sets in an 8:2 ratio. The training set is used to train a machine learning model predicting the combustion chamber outlet temperature, and the test set is used for testing to identify the model that performs best under different sampling methods. Machine learning models with a value greater than 0.95 are saved, and finally, the high-precision models are merged to obtain a more scientific joint prediction model with stronger generalization ability.

[0048] Figure 2 In step two, machine learning construction involves building the aforementioned joint prediction model. Here, post-sampling modeling corresponds to the model training process described in the previous embodiments, while comparing multiple algorithms and evaluating the model is the selection process. Machine learning models with a value greater than 0.95 are considered high-precision models and are saved.

[0049] In step three, the model fusion part involves calling the high-precision model and removing abnormal results, i.e., deleting abnormal predicted values ​​that deviate from the confidence interval using the 3σ principle; integrating the prediction results of each model to obtain the final result, i.e., the average remaining output temperature after deleting abnormal prediction results to obtain the comprehensive value.

[0050] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0051] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0052] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0053] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0054] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0055] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0056] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0057] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A combustion chamber outlet temperature modeling and prediction method, characterized by, The method comprises: obtaining sample outlet temperature data of a combustion chamber outlet and sample condition variable data affecting the sample outlet temperature data; training multiple machine learning models based on the sample outlet temperature data and the sample condition variable data; retaining machine learning models with a coefficient of determination greater than a preset threshold value among the trained machine learning models to form a joint prediction model; wherein a predicted temperature output by the joint prediction model is a comprehensive value of output temperatures of each machine learning model included in the joint prediction model; inputting condition variables of the combustion chamber outlet temperature into the joint prediction model to obtain a predicted temperature output by the joint prediction model.

2. The method of claim 1, wherein, The sample condition variable data is obtained in the following manner: obtaining original condition variable data of the combustion chamber outlet; performing Gaussian filtering processing on the original condition variable data to obtain first condition variable data; calculating the correlation between each sample condition variable in the first condition variable data, and if there are first sample condition variables and second sample condition variables with a correlation greater than a preset correlation threshold value, removing the sample condition variable with a smaller correlation with the sample outlet temperature data among the first sample condition variable and the second sample condition variable; obtaining the sample condition variable data based on the retained sample condition variables in the first condition variable data.

3. The method of claim 2, wherein, The method further comprises: if the first sample condition variable is fuel flow, and if there is a second sample condition variable with a correlation greater than a preset correlation threshold value with the fuel flow, the second sample condition variable is removed.

4. The method of claim 1, wherein, The method of obtaining sample outlet temperature data of a combustion chamber outlet and sample condition variable data affecting the sample outlet temperature data comprises: respectively using time interval sampling, sliding time window sampling and change rate sampling methods to obtain sample outlet temperature and sample condition variable of each sampling point in a preset time period; obtaining the sample outlet temperature data and the sample condition variable data based on the normalization results of the obtained sample outlet temperature and sample condition variable.

5. The method of claim 1, wherein, The multiple machine learning models comprise linear regression model, ridge regression model, lasso regression model, elastic net model, K nearest neighbor regression model, support vector regression model, decision tree model, random forest model, gradient boosting tree model, XGBoost model, extreme random tree model and CatBoost model.

6. The method of claim 1, wherein, The comprehensive value is obtained in the following manner: after obtaining the output temperatures of each machine learning model included in the joint prediction model, calculating the mean and standard deviation based on each output temperature, and removing the output temperature with an abnormal value according to the mean and standard deviation; calculating the average value of the remaining output temperatures as the comprehensive value.

7. The method of claim 1, wherein For each machine learning model, the coefficient of determination is represented by the ratio of residual sum of squares to total sum of squares, and the preset threshold value is 0.95.

Images