Method for improving heat exchange efficiency of heat exchanger based on machine learning

Abstract

The invention provides a method for improving heat exchange efficiency of a heat exchanger based on machine learning, which comprises the following steps: S1, collecting historical data of the heat exchanger to form a corresponding sample set; s2, preprocessing the data; s3, data correlation analysis and feature selection; s4, modeling, and screening an optimal model as an output model; and S5, deploying the model, connecting a model interface with a heat exchanger data monitoring system, and adjusting operation parameters of the heat exchanger according to a model output result. According to the method, a large number of historical data parameters are utilized, computing power analysis is carried out with the help of data science to find an optimal solution, an environment based on an industrial internet platform is constructed, a heat exchange efficiency optimization model is conveniently and rapidly deployed, optimization of operation parameters in the production process is achieved, energy consumption is reduced, and economic benefits are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical product production, and in particular to a method for improving the heat exchange efficiency of a heat exchanger based on machine learning. Background Art

[0002] With the continuous advancement of industrial automation and informatization, intelligent information technology is becoming a key tool for improving production efficiency, reducing costs, and enhancing equipment operational stability. Heat exchangers are core equipment in chemical production processes, and precise control of their heat transfer efficiency has a decisive impact on production energy consumption and product quality. Traditionally, heat exchange management relies primarily on manual adjustments, which not only leads to delayed response times but also makes dynamic optimization difficult, easily resulting in energy waste and reduced equipment performance.

[0003] The patent with publication number CN119227267A discloses a method for optimizing the thermal efficiency of a small end difference of a series plate heat exchanger, which is applied to a heat pump device and relates to the field of heat pump technology. The method includes: collecting the initial temperature information of the heating medium and the heat source, obtaining the heat exchange control parameter space, randomly generating multiple control parameter combinations for heat exchange prediction, obtaining the head-end difference, tail-end difference and corresponding medium and heat source temperature information respectively, and performing thermal efficiency analysis. By calculating the thermal efficiency adaptability and heat exchange adaptability, the control parameters are optimized and adjusted to obtain the optimal head heat exchange control parameters and tail heat exchange control parameters. The present invention solves the technical problem in the prior art that the heat exchanger of the traditional heat pump device has a large end difference and low heat exchange efficiency during operation, which affects the energy efficiency of the device. It achieves the technical effect of improving the thermal efficiency and reducing the end difference by optimizing the control parameters of the series plate heat exchanger in the heat pump device, thereby improving the energy efficiency of the device. This patented method requires randomly generating multiple control parameter combinations for heat exchange prediction, and then optimizing the parameters by calculating fitness. When the parameter space is large, the amount of calculation will increase significantly and the calculation efficiency will be low, resulting in a delay in the heat exchanger response time and difficulty in achieving dynamic optimization. Summary of the Invention

[0004] In view of this, the present invention aims to propose a calculation method based on machine learning to improve the heat transfer efficiency of the heat exchanger, so as to improve the heat transfer efficiency of the heat exchanger. The core purpose of this method is to reduce the dependence on human historical experience and manual operation, thereby reducing response time, improving control accuracy and efficiency, and reducing energy consumption and maintenance costs.

[0005] The present invention discloses a method for improving the heat exchange efficiency of a heat exchanger based on machine learning, comprising:

[0006] Step S1: Collect historical data of the heat exchanger to form a corresponding sample set;

[0007] Step S2: preprocessing the data;

[0008] Step S3: data correlation analysis and feature selection;

[0009] Step S4: Modeling and selecting the optimal model as the output model;

[0010] Step S5: Deploy the model, connect the model interface to the heat exchanger data monitoring system, and adjust the operating parameters of the heat exchanger according to the model output results.

[0011] Furthermore, the historical data of the heat exchanger in step S1 at least includes: mass flow, specific heat capacity, inlet temperature, and outlet temperature.

[0012] Furthermore, the data preprocessing in step S2 includes defect processing, feature setting and data standardization.

[0013] Furthermore, the feature setting includes: defining the mass flow rate, specific heat capacity and inlet temperature as independent variable features, and defining the outlet temperature as a dependent variable feature.

[0014] Furthermore, the data correlation analysis and feature selection in step S3 include:

[0015] Step S31: performing correlation analysis on the independent variable features and the dependent variable features in pairs;

[0016] Step S32: Eliminate independent variable features that are weakly correlated with dependent variable features and retain core features.

[0017] Furthermore, when performing correlation analysis and feature selection, the following steps are also performed:

[0018] Step S321: Determine whether any two features in the independent variable features are strongly correlated, and perform feature engineering processing on the features that are strongly correlated.

[0019] Furthermore, step S4 includes:

[0020] Step S41: using multiple algorithms to perform regression simulation on the processed data and establish corresponding models;

[0021] Step S42: Evaluate the models obtained by different algorithms and use the optimal model as the output model.

[0022] Furthermore, the multiple algorithms in step S41 include one or more of nonlinear SVM regression, linear regression, random forest regression, and BP neural network regression.

[0023] Furthermore, in step S42 , based on the mean square error of each model, the closer the mean square error is to 0, the healthier the model is, and the model with the mean square error closest to 0 is determined as the optimal model.

[0024] Furthermore, the output model is an outlet temperature prediction model obtained by regression simulation using a random forest regression algorithm.

[0025] Compared with the existing technology, the method of improving the heat exchange efficiency of a heat exchanger based on machine learning described in the present invention has the following advantages:

[0026] (1) Utilize a large amount of historical data parameters and use data science to perform computational analysis to find the optimal solution, thereby guiding on-site production;

[0027] (2) Combining rich machine learning algorithms with the professional experience of process personnel, layout design is performed through drag-and-drop operations, and static properties and interactive operation properties are configured to achieve rapid construction of component pages;

[0028] (3) Build an environment based on the Industrial Internet platform to facilitate the rapid deployment of heat exchange efficiency optimization models, optimize production process operating parameters, reduce energy consumption, and improve economic benefits;

[0029] (4) Promote enterprise informatization construction and effectively integrate production and informatization. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0031] Figure 1 This is a flow chart of a method for improving the heat exchange efficiency of a heat exchanger based on machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the technical means, objectives and effects of the present invention easier to understand, the embodiments of the present invention are described in detail below.

[0033] It should be noted that all terms used in the present invention to indicate direction and position, such as "up", "down", "left", "right", "front", "back", "vertical", "horizontal", "inside", "outside", "top", "low", "lateral", "longitudinal", "center", etc., are only used to explain the relative positional relationship and connection status between various components in a specific state. They are only for the convenience of describing the present invention, and do not require that the present invention must be constructed and operated in a specific orientation. Therefore, they cannot be understood as limiting the present invention. In addition, the descriptions of "first", "second", etc. in the present invention are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features.

[0034] In the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood broadly. For example, they may refer to fixed, detachable, or integral connections; mechanical connections; direct connections or indirect connections through an intermediary; and internal communication between two components. Those skilled in the art will understand the specific meanings of these terms in the present invention based on the specific circumstances.

[0035] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative uses 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 any one or more embodiments or examples.

[0036] Example 1

[0037] The present invention discloses a method for improving the heat exchange efficiency of a heat exchanger based on machine learning, comprising:

[0038] Step S1: Collect historical data of the heat exchanger to form a corresponding sample set;

[0039] Step S2: preprocessing the data;

[0040] Step S3: data correlation analysis and feature selection;

[0041] Step S4: Modeling and selecting the optimal model as the output model;

[0042] Step S5: Deploy the model, connect the model interface to the heat exchanger data monitoring system, and adjust the operating parameters of the heat exchanger according to the model output results.

[0043] In the prior art, key parameters of the heat exchanger are usually adjusted manually based on the operator's historical experience. This adjustment method will cause response time delays and make dynamic optimization difficult to achieve. It will also easily lead to large fluctuations in the standard deviation of the heat exchanger outlet temperature, resulting in energy waste or reduced equipment performance, which is not conducive to the safety and service life of the heat exchanger. In this example, through the setting of the above method, it is possible to perform computing power analysis based on historical data with the help of data science to obtain the optimal solution, and then understand on-site production, realize dynamic optimization of parameters in the production process, reduce heat exchanger energy consumption, and improve heat exchange efficiency and economic benefits.

[0044] In this example, the historical data of the heat exchanger in step S1 includes at least: mass flow rate, specific heat capacity, inlet temperature, and outlet temperature, and each of the above data sets is recorded as a feature. The above mass flow rate, inlet temperature, outlet temperature and pressure can be detected and collected by multiple sensors in the existing technology. The detection frequency can refer to the existing technology and will not be described in detail here. The specific heat capacity is determined according to the type of fluid in the heat exchanger. After the above data is collected, it is stored in the industrial database to facilitate later processing and analysis. The industrial database includes a relational database and a time series database. Preferably, the storage tool is the time series database InfluxDB, which is synchronously backed up to the relational database MySQL. Specifically in this example, the data selected is data within the normal fluctuation range (0-1300) as the data source of the machine learning algorithm. Optionally, the pressure data of the heat exchanger can also be collected as subordinate data during data collection, which will not be described in detail here. Recording each of the above data sets as a feature means that the mass flow rate, specific heat capacity, inlet temperature, outlet temperature

[0045] Specifically, the data preprocessing in step S2 includes defect handling, feature setting, and data standardization. Data preprocessing, including defect handling and data design, can improve data quality. Preprocessing abnormal data can reduce its impact on prediction results, making the data more accurate, reliable, and complete, facilitating subsequent analysis and processing.

[0046] Data preprocessing can employ methods used in existing technologies, such as filtering or normalization, and these are not limited here. Specifically, defect handling includes deleting blank rows, removing outliers, and filling missing values. Missing value handling involves replacing variables with missing values ​​with the maximum, minimum, mean, mode, or a custom value. For blank rows, the entire row of data with missing values ​​is directly deleted. For clearly outliers, data can be deleted or replaced with the maximum, minimum, mean, mode, or a custom value. It should be noted that for outlier identification, a corresponding threshold range or graphical range can be set. Values ​​outside the corresponding range are considered outliers and can be deleted or replaced. Data design refers to the structural design and feature engineering of a dataset before model construction to make the data suitable for model training. Specifically, during data preprocessing, attributes to be analyzed can be selected and variable roles defined for these attributes to accurately obtain the desired data features and their relationships.

[0047] As an example, the feature setting includes defining mass flow rate, specific heat capacity, and inlet temperature as independent variable features, and outlet temperature as a dependent variable feature, to facilitate subsequent modeling operations. Other features, such as pressure, are also defined as independent variable features.

[0048] Specifically, taking the normalization method as an example, the data standardization processing includes: normalizing the data of each feature (mass flow rate, specific heat capacity, inlet temperature, outlet temperature) according to formula (1):

[0049] (1)

[0050] Among them, x is a value to be processed in any feature, x S is the normalized value of x, min is the minimum value of the feature corresponding to x, x max is the maximum value of the feature corresponding to x.

[0051] Through the above settings, the eigenvalues ​​of each feature are all scaled to the range of [0, 1], so that features of different dimensions have a more balanced impact on the model.

[0052] As an example of the present invention, the data correlation analysis and feature selection in step S3 include:

[0053] Step S31: performing correlation analysis on the independent variable features and the dependent variable features in pairs;

[0054] Step S32: Eliminate independent variable features that are weakly correlated with dependent variable features and retain core features.

[0055] Specifically, in step S31, the correlation coefficient between the two features is calculated using formula (2):

[0056] (2)

[0057] Among them, r is the correlation coefficient, n is the total number of data points involved in the calculation, X i is the value of the i-th sample in the first feature, is the arithmetic mean of all values ​​in the first feature, Y i is the number of the i-th sample in the second feature, is the arithmetic mean of all values ​​in the first feature.

[0058] Specifically, equation (2) can be used to determine the correlation between two features. For example, by comparing the calculated absolute value of r with a first preset correlation threshold r1, when |r| < r1, it can be determined that the two are weakly correlated. When the independent variable and the dependent variable are weakly correlated, the independent variable can be eliminated to reduce the amount of model learning data, thereby reducing the learning difficulty and improving prediction accuracy. As one optional example, r1 is 0.1-0.3, preferably 0.2.

[0059] It should be noted that when conducting correlation analysis, the Pearson correlation coefficient and Spearman rank correlation can be combined to perform correlation analysis to improve the accuracy of the correlation analysis and effectively screen out the core features.

[0060] In addition, when performing correlation analysis and feature selection, the following steps are required:

[0061] Step S321: Determine whether any two features in the independent variable features are strongly correlated, and perform feature engineering processing on the features that are strongly correlated.

[0062] It should be noted that when two features are strongly correlated between the independent variables, it is easy to overestimate the importance of certain features when the subsequent model is built, so that small data fluctuations cause significant changes in the predicted value, affecting the stability of the model and reducing the accuracy of the model. Specifically, when it is found that there are strongly correlated independent variables, one of the features can be deleted through feature engineering or two strongly correlated features can be created as a combined feature, which helps to optimize the model later and improve the accuracy of the model. Specifically, when judging whether the two independent variables are strongly correlated, the absolute value of r between the two can be calculated and compared with the second preset correlation threshold r2. When |r| ≥ r2, it can be determined that there is a strong correlation between the two. As one of the optional examples, r2 takes a value of 0.7~0.9, preferably 0.8. Through the setting of step S3, the data after preprocessing and feature selection can be integrated into a structured data set, which is convenient for subsequent model construction.

[0063] In this example, step S4 includes:

[0064] Step S41: using multiple algorithms to perform regression simulation on the processed data and establish corresponding models;

[0065] Step S42: Evaluate the models obtained by different algorithms and use the optimal model as the output model.

[0066] Optionally, the multiple algorithms in step S41 include one or more of nonlinear SVM regression, linear regression, random forest regression, and BP neural network regression. In the above algorithms, the key input parameters include several independent variable parameters such as mass flow rate, specific heat capacity, and inlet temperature, and the outlet temperature is used as a dependent variable parameter. The above algorithms perform error evaluation after processing and self-prediction of prefabricated data sets. For data sets with the same independent variables and dependent variables, the model performance of a regression algorithm under one or more sets of parameter combinations or the performance gap between multiple regression algorithms are compared to test the reliability of the regression model. Finally, the best quality regression model is obtained based on some evaluation indicators (such as relative error, mean absolute error, root mean square error, etc.) or graphical presentation.

[0067] Specifically, in step S42, the model with the closest mean square error to 0 is determined as the optimal model based on the mean square error of each model. This approach effectively reduces measurement errors during operation, provides better guidance for precise control of the heat exchanger, optimizes heat transfer efficiency, and reduces energy consumption.

[0068] As one of the preferred embodiments, the output model is an outlet temperature prediction model obtained through regression simulation using a random forest regression algorithm. Specifically, during testing, historical data (a total of 17,281 groups) is first exported for preprocessing to set the quantity, minimum sample number, and maximum sample number, while blank lines and abnormal data are deleted to retain normal values. The independent variables are set as mass flow rate, specific heat capacity, and inlet temperature, and the dependent variable is outlet temperature. Through comparison of multiple algorithms, the outlet temperature prediction model obtained using the forest regression method has the closest mean square error to 0, the highest healthiness, and relatively better prediction accuracy than other models. Therefore, the outlet temperature prediction model obtained after simulation using the random forest regression algorithm is deployed as the preferred model. Specifically, the relevant parameters of the heat exchanger are controlled based on the predicted outlet temperature to achieve automatic control of the heat exchanger.

[0069] Specifically, the random forest regression algorithm is processed in the following way:

[0070] The impurity of the parent node is calculated by formula (3):

[0071] (3)

[0072] Among them, MSE P is the mean square error of the parent node, which is used to measure the impurity of the parent node, n is the number of samples of the parent node, y i is the value of the dependent variable at the i-th node, y n is the average value of n dependent variables at the parent node.

[0073] The weighted sum of the sub-node MSE is calculated using formula (4):

[0074] (4)

[0075] Among them, MSE C is the weighted sum of the mean squared error of the child nodes, MSE L is the mean square error of the left child node, MSE R is the mean square error of the right child node, n L is the number of samples of the left child node, n R is the number of samples of the right child node. L 、MSE R The calculation principle and MSE P The same is no longer limited here.

[0076] Among them, the splitting rule of random forest regression is as follows:

[0077] Traverse all features and all possible thresholds, find the splitting method that maximizes the value of formula (5), and complete the node splitting:

[0078] (5)

[0079] Among them, ΔMSE is the mean square error MSE of the parent node P and the weighted sum MSE of the child nodes MSE C The difference.

[0080] It should be noted that when a child node meets any of the following conditions, it stops splitting and becomes a leaf node:

[0081] 1. The number of checkpoint samples is less than the preset node sample threshold. The node sample threshold is a preset value. Optionally, the node sample threshold can be 10 or other values.

[0082] 2. When the split depth reaches the preset depth, the pre-trial depth is the preset value. Optionally, the preset depth is 5 or other values;

[0083] 3. When the purity reaches the preset purity threshold, the preset purity threshold is a preset value. Optionally, the preset depth is 0.5 or other values. The purity is the mean square error corresponding to the node, which can be calculated according to the principle of formula (2);

[0084] 4. When ΔMSE reaches a preset threshold, optionally, the preset threshold of ΔMSE is 0.1 or another value. In this case, it indicates that the improvement of the split is insufficient and the improvement of subsequent splits is not significant.

[0085] In addition, the random forest algorithm also requires the preset number of decision trees m. Each decision tree uses the autonomous sampling method to extract samples from the collected data set and is independently constructed. The output value of the final model is calculated according to formula (6):

[0086] (6)

[0087] Among them, y is the predicted value output by the random forest regression model, m is the preset number of decision trees, and y j The predicted value output by the jth decision tree. m is preset before the model is built. Optionally, m is 10 or another preset value.

[0088] In order to verify the effect of the prediction model provided by this example, 17281 groups of data before the model deployment and 17281 groups after the model deployment were taken for comparison. The results showed that before using the prediction model deployed in this application, the maximum absolute value of the difference between the outlet temperature and the average temperature of the heat exchanger during operation could reach 37.3°C. When the average temperature was 294.9°C, the maximum outlet temperature of the heat exchanger during operation was 331.3°C, the minimum temperature was 257.6°C, and the standard deviation of the outlet temperature was 8.587°C. After using the prediction model deployed in this application, the maximum absolute value of the difference between the outlet temperature and the average temperature of the heat exchanger during operation was 331.3°C, the minimum temperature was 257.6°C, and the standard deviation of the outlet temperature was 8.587°C. The outlet temperature of the heat exchanger is 15.7°C. When the average temperature is 287.6°C, the maximum outlet temperature of the operating heat exchanger is 303.3°C, the minimum temperature is 272.9°C, and the outlet temperature standard deviation is 2.35°C. The outlet temperature standard deviation of the heat exchanger is a key indicator of the heat exchange efficiency of the heat exchanger. The smaller the outlet temperature standard deviation, the higher the heat exchange efficiency. The higher the outlet temperature standard deviation, the lower the heat exchange efficiency. In this example, through the deployment of the above model, the outlet temperature standard deviation is reduced from 8.587°C to 2.35°C, a decrease of 72.63%, which significantly improves the heat exchange efficiency of the heat exchanger and reduces the energy efficiency of the heat exchanger.

[0089] In practice, users can build models through industrial platforms (such as DCS systems) based on the platform's data science module. This includes: selecting data sample sets: collecting and importing historical key parameter data (mass flow, specific heat capacity, inlet temperature, outlet temperature), and generating corresponding sample sets for data modeling and analysis. Platform creation applications include: data input and output, data preprocessing, signal processing, feature engineering, machine learning algorithms, customization, and model evaluation. The model building design is in a componentized form. Through the visual components provided by the platform, users can drag and drop components to any location on the page. In the editing panel of the corresponding component, samples can be input, and algorithms can be optimized.

[0090] During the data processing stage, the platform's built-in algorithms can perform statistical analysis on each field of the sample set data, including attributes, types, number of missing values, minimum, maximum, standard deviation and other related processing and analysis.

[0091] In data correlation analysis, the built-in algorithm is debugged to establish correlations between corresponding parameters and find the optimal solution. The insight results display a correlation coefficient matrix chart. You can select the field name in the top row to view the correlation coefficient between each field. In this component's insight interface, you can also view the field data obtained by the current component and analyze and compile information for each field.

[0092] In addition, the industrial platform can also publish the models that have been built and publish them to the experimental template, which can be called and used by the platform.

[0093] Specifically, after the deployed model performs prediction calculations based on the input values, the system generates a data error graph. In addition to displaying the data set, the regression model evaluation interface also provides a running error evaluation to analyze the error, relative error, and absolute error of the entire model. At the same time, the error graph also displays the error curve of each regression model and the true value; this evaluation will also be displayed independently for each model, analyzing and drawing the model's error evaluation, error graph, residual scatter plot, etc.

[0094] For published models, relevant personnel can view and retrieve them through the relevant human-computer interaction interface.

[0095] In this example, historical data can be collected into a form, and the relevant parameters of the heat exchanger can be adjusted and optimized with reference to the historical data, effectively improving the heat exchange efficiency and the control level of the production process, and providing strong technical support for the optimization and adjustment of chemical production processes, especially sponge titanium production processes.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for improving heat exchange efficiency of a heat exchanger based on machine learning, characterized in that: include: Step S1: Collect historical data of the heat exchanger to form a corresponding sample set; Step S2: preprocessing the data; Step S3: data correlation analysis and feature selection; Step S4: Modeling and selecting the optimal model as the output model; Step S5: Deploy the model, connect the model interface to the heat exchanger data monitoring system, and adjust the operating parameters of the heat exchanger according to the model output results.

2. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 1, characterized in that: The historical data of the heat exchanger in step S1 at least includes: mass flow, specific heat capacity, inlet temperature, and outlet temperature.

3. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 2, characterized in that: The data preprocessing in step S2 includes defect processing, feature setting and data standardization.

4. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 3, characterized in that: The feature setting includes: defining the mass flow rate, specific heat capacity and inlet temperature as independent variable features, and defining the outlet temperature as a dependent variable feature.

5. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 1, characterized in that: The data correlation analysis and feature selection in step S3 include: Step S31: performing correlation analysis on the independent variable features and the dependent variable features in pairs; Step S32: Eliminate independent variable features that are weakly correlated with dependent variable features and retain core features.

6. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 5, characterized in that: When performing correlation analysis and feature selection, the following steps are also performed: Step S321: Determine whether any two features in the independent variable features are strongly correlated, and perform feature engineering processing on the features that are strongly correlated.

7. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 1, characterized in that: Step S4 includes: Step S41: using multiple algorithms to perform regression simulation on the processed data and establish corresponding models; Step S42: Evaluate the models obtained by different algorithms and use the optimal model as the output model.

8. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 7, characterized in that: The multiple algorithms in step S41 include one or more of nonlinear SVM regression, linear regression, random forest regression, and BP neural network regression.

9. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 7, characterized in that: In step S42 , based on the mean square error of each model, the closer the mean square error is to 0, the healthier the model is, and the model with the mean square error closest to 0 is determined as the optimal model.

10. The method for improving heat exchange efficiency of a heat exchanger based on machine learning according to claim 8, characterized in that: The output model is the outlet temperature prediction model obtained by regression simulation using the random forest regression algorithm.

Images