Credible gasoline octane number loss prediction method, system, equipment and medium

Abstract

The invention provides a credible gasoline octane number loss prediction method, system, equipment and medium, and the method comprises the following steps: obtaining original data of a refinery plant production process and preprocessing the original data, the original data comprising a plurality of process parameters influencing octane number loss; constructing and training a sparse auto-encoder, and performing key feature extraction on the preprocessed original data; constructing a stacked ensemble learning model which comprises a plurality of base learners and a meta learner, performing regression prediction on the depth features by the plurality of base learners to output a plurality of octane number loss prediction intermediate results, performing quadratic regression prediction by using a least square support vector machine as the meta learner, and outputting a loss prediction value of the octane number; and carrying out interpretability analysis on the prediction process of the stacked ensemble learning model by adopting a feature importance analysis and partial dependency graph so as to solve the problem of low prediction precision caused by difficulty in extracting key features from high-dimensional noise-containing industrial data in an octane number loss prediction method.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and chemical process optimization technology, and in particular to a method, system, device and medium for predicting the loss of reliable gasoline octane number. Background Technology

[0002] With the advancement of industrialization, the petrochemical industry occupies an important position in the global economy, especially in the optimization of gasoline refining processes, which directly affects energy efficiency, environmental protection, and economic benefits. In gasoline refining, octane number (RON) is a crucial indicator of gasoline quality, and octane number loss impacts gasoline quality and energy utilization efficiency. Traditional gasoline refining processes involve complex chemical reactions and physical processes, and the energy consumption and emissions issues in these processes urgently need to be addressed. To achieve energy conservation, emission reduction, and improved production efficiency, optimizing the prediction and control of octane number loss has become a key technical challenge for the petrochemical industry.

[0003] In traditional gasoline refining processes, the prediction of octane number loss relies on empirical methods and simplified mathematical models. These methods typically ignore the nonlinear relationships between high-dimensional data and have poor ability to handle noisy data. With the development of artificial intelligence technology, machine learning-based prediction models are gradually being applied to industrial optimization. However, some existing machine learning methods still face problems such as difficulty in feature selection, significant impact from data noise, and insufficient prediction accuracy when dealing with complex industrial data.

[0004] In existing technologies, the main methods for predicting octane number loss are as follows: (1) Traditional regression analysis In early research and applications, multiple linear regression or nonlinear regression models were often used to predict octane number loss. These methods are simple and easy to use, but they are poor at handling high-dimensional and complex data and cannot effectively capture the nonlinear relationships between variables. For example, classic regression models often rely on manually selected features and perform poorly on noisy data, resulting in low prediction accuracy.

[0005] (2) Machine learning-based prediction methods In recent years, with the rise of machine learning, methods such as Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB) have been widely used in octane number loss prediction. For example, Random Forest improves prediction accuracy by ensembling multiple decision trees and can effectively handle high-dimensional data and nonlinear relationships. However, these methods often face problems such as overfitting and insufficient feature selection, and lack good interpretability.

[0006] (3) Deep learning methods To further improve prediction accuracy, deep learning methods (such as neural networks) have also been applied to octane number loss prediction. Sparse autoencoders (SAEs), an unsupervised machine learning algorithm, automatically extract key features from data and perform dimensionality reduction, and have been applied in multiple fields. However, this method has drawbacks: traditional neural network models are prone to overfitting when trained on high-dimensional data, and the training process is relatively complex, making it difficult to quickly adapt to industrial data.

[0007] (4) Stacking EL Stacking is a method to improve model performance by combining multiple base learners. Existing stacking ensemble methods typically combine multiple machine learning models (such as decision trees, SVMs, random forests, etc.) to achieve stronger predictive capabilities. However, stacking ensemble methods still have limitations in feature extraction and data processing, failing to effectively solve the challenge of feature extraction from high-dimensional data.

[0008] Therefore, there is an urgent need for a reliable method for predicting gasoline octane number loss, in order to solve the technical problem that existing octane number loss prediction methods are unable to extract key features from high-dimensional and noisy industrial data, resulting in low prediction accuracy. Summary of the Invention

[0009] To overcome the problems existing in the related technologies, this disclosure provides a reliable gasoline octane number loss prediction method, system, device and medium to solve the technical problem that existing octane number loss prediction methods in the related technologies have difficulty in extracting key features from high-dimensional and noisy industrial data, resulting in low prediction accuracy.

[0010] This specification provides one or more embodiments of a method for predicting the loss of reliable gasoline octane number, including the following steps: The raw data of the refinery's production process is acquired and preprocessed, including multiple process parameters that affect octane number loss. A sparse autoencoder is constructed and trained. The sparse autoencoder extracts key features from the preprocessed raw data and removes data redundancy and noise to obtain deep features. A stacked ensemble learning model is constructed, which includes multiple base learners and a meta learner. The multiple base learners perform regression prediction on the deep features and output multiple intermediate octane number loss prediction results. The least squares support vector machine is used as the meta learner to perform secondary regression prediction and output the octane number loss prediction value. Feature importance analysis and partial dependency graphs are used to perform interpretability analysis on the prediction process of the stacked ensemble learning model.

[0011] Preferably, the method further includes the following steps: The model performance was evaluated and optimized using a comprehensive evaluation index of coefficient of determination, mean square error, and median absolute error.

[0012] Preferably, the method further includes the following steps: Based on the interpretability analysis, the process parameters that have the greatest impact on octane number loss are identified, and optimized operation suggestions for the gasoline refining process are output.

[0013] Preferably, the plurality of base learners include Lasso regression, ridge regression, support vector regression, elastic net regression, naive Bayes, logistic regression, random forest, gradient boosting regression, extreme random tree, and extreme gradient boosting.

[0014] Preferably, the sparse autoencoder extracts key features from the preprocessed raw data, removes data redundancy and noise to obtain deep features, specifically including the following steps: The original data includes 367 variables; The 367 variables were trained using a sparse autoencoder model. By adjusting the number of neurons in the hidden layer and the sparsity constraint parameters, the 367 variables were transformed into 31 deep features. The features of the 367 variables were nonlinearly mapped to extract key features. After removing noise and redundant information, the 31 deep features were obtained.

[0015] This specification provides one or more embodiments of a loss prediction system for reliable gasoline octane number, including a data acquisition module, a deep feature extraction module, a prediction module, and an analysis module; The data acquisition module is used to acquire and preprocess raw data of the refinery's production process. The raw data includes multiple process parameters that affect octane number loss. The deep feature extraction module is used to construct and train a sparse autoencoder. The sparse autoencoder extracts key features from the preprocessed raw data and removes data redundancy and noise to obtain deep features. The prediction module is used to construct a stacked ensemble learning model, which includes multiple base learners and a meta-learner. The multiple base learners perform regression prediction on the deep features and output multiple intermediate octane number loss prediction results. The least squares support vector machine is used as the meta-learner to perform secondary regression prediction and output the octane number loss prediction value. The analysis module is used to evaluate and optimize the model's performance using a comprehensive evaluation index of the coefficient of determination, mean square error, and median absolute error.

[0016] Preferably, it also includes an optimization module for evaluating and optimizing the model's performance using a comprehensive evaluation index of the coefficient of determination, mean square error, and median absolute error.

[0017] Preferably, the analysis module is further configured to determine the process parameters that have the greatest impact on octane number loss based on the interpretability analysis, and output optimized operation suggestions for the gasoline refining process.

[0018] This specification provides one or more embodiments of a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the loss prediction method for reliable gasoline octane number as described above.

[0019] This specification provides one or more embodiments of a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for predicting the loss of reliable gasoline octane number as described above.

[0020] This disclosure provides a reliable method, system, device, and medium for predicting gasoline octane number loss. Its advantages lie in acquiring and preprocessing raw data from the refinery production process, ensuring data integrity and availability, and providing high-quality input for subsequent feature extraction and model training, thus mitigating prediction bias risks caused by data quality issues from the outset. Furthermore, it constructs and trains a sparse autoencoder, which extracts key features from the preprocessed raw data, removes data redundancy and noise to obtain deep features, and performs feature optimization to generate deep features that better reflect the inherent laws of octane number loss. This not only improves the upper limit of accuracy of the subsequent prediction model but also reduces model training complexity through dimensionality reduction and redundancy removal, thus providing a more reliable prediction path. This paper provides "efficient and highly correlated" feature inputs; it constructs a stacked ensemble learning model, which includes multiple base learners and one meta-learner. The multiple base learners perform regression prediction on the deep features, outputting multiple intermediate octane number loss prediction results. A least squares support vector machine is used as the meta-learner for secondary regression prediction, outputting the predicted octane number loss value. This approach covers all operating conditions through multiple base learners and improves prediction accuracy through the meta-learner, while enhancing the model's adaptability to changes in operating conditions. The output quantitative prediction value provides a direct basis for refinery production decisions. Through feature importance analysis and partial dependency graphs, the originally complex "black box" model prediction process becomes transparent and interpretable. It can quantify the influence weight of various process parameters on octane number loss, helping engineers quickly identify key control variables. Simultaneously, it can intuitively display the nonlinear relationship between single or two parameters and octane number loss, revealing their inherent influence patterns. This makes the model's predictions no longer isolated numerical values, but a decision-making basis containing causal logic, which greatly enhances the model's credibility, acceptability, and practical value in industrial practice, and provides clear guidance for refineries to carry out targeted process optimization. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a method for predicting the loss of reliable gasoline octane number provided in one or more embodiments of this specification; Figure 2 This is a schematic diagram of a quadratic regression prediction process provided for one or more embodiments of this specification. Figure 3 Feature importance analysis in interpretability analysis diagrams provided for one or more embodiments of this specification; Figure 4 Partial dependency graph analysis in the interpretability analysis diagram provided for one or more embodiments of this specification; Figure 5 This is a schematic diagram illustrating how a sparse autoencoder, provided in one or more embodiments of this specification, reduces the dimensionality of variables affecting octane number loss. Figure 6 A schematic diagram of a loss prediction system for reliable gasoline octane number provided for one or more embodiments of this specification; Figure 7 This is a schematic diagram of the structure of a computer device provided for one or more embodiments of this specification. Detailed Implementation

[0023] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this invention.

[0024] The present invention will now be described in detail with reference to specific embodiments and accompanying drawings.

[0025] Method Implementation Examples According to embodiments of the present invention, a method for predicting the loss of reliable gasoline octane number is provided, such as... Figure 1The diagram shown is a flowchart illustrating the method for predicting the loss of reliable gasoline octane number provided in this embodiment. The method for predicting the loss of reliable gasoline octane number according to this embodiment includes the following steps: S110. Obtain and preprocess the raw data of the refinery's production process. The raw data includes multiple process parameters that affect octane number loss, such as temperature, pressure, and reactant concentration.

[0026] S120. Construct and train a sparse autoencoder. The sparse autoencoder extracts key features from the preprocessed raw data, removes data redundancy and noise to obtain deep features. The sparse autoencoder is an unsupervised learning method that compresses data by automatically encoding and decoding, and removes redundant features while maintaining important information.

[0027] S130. Construct a stacked ensemble learning model. Stacked ensemble learning is an ensemble learning method that improves model performance by combining multiple base learners, integrating the advantages of multiple machine learning models and reducing the limitations of a single model. The stacked ensemble learning model includes multiple base learners and a meta-learner for regression prediction. The base learners include Lasso regression, Ridge regression, Support Vector Regression, Elastic Net Regression, Naive Bayes, Logistic Regression, Random Forest, Gradient Boosting Regression, Extreme Random Tree, and Extreme Gradient Boosting. Each base learner performs regression prediction on deep features, outputting multiple intermediate octane number loss predictions. Through a stacked ensemble strategy, the outputs of these base learners are then input into the meta-learner. To aggregate the outputs of the base learners, a least squares support vector machine is used as the meta-learner for secondary regression prediction, outputting the predicted octane number loss value, such as... Figure 2 The diagram shown illustrates the quadratic regression prediction process provided in this embodiment. Due to the robustness of LSSVM in handling large-scale, high-dimensional, and nonlinear datasets, it has significant advantages in optimizing overall prediction performance. Compared to traditional regression models, stacked ensemble learning effectively reduces overfitting by integrating the advantages of multiple models and can handle complex nonlinear data relationships.

[0028] S140. Feature importance analysis and partial dependency graphs are used to perform interpretability analysis on the prediction process of the stacked ensemble learning model. This not only helps to analyze which process parameters have the greatest impact on octane number loss, but also provides specific operational suggestions for refinery process optimization and promotes the implementation of green production processes.

[0029] The method provided in this embodiment acquires and preprocesses raw data from the refinery production process to ensure data integrity and usability, providing high-quality input for subsequent feature extraction and model training, thus mitigating the risk of prediction bias caused by data quality issues from the outset. It constructs and trains a sparse autoencoder, which extracts key features from the preprocessed raw data, removing data redundancy and noise to obtain deep features. This allows for feature optimization, generating deep features that better reflect the inherent patterns of octane number loss. This not only improves the upper limit of accuracy of the subsequent prediction model but also reduces model training complexity through dimensionality reduction and redundancy removal, providing "efficient and highly correlated" features for the prediction process. The model employs a stacked ensemble learning model, which includes multiple base learners and a meta-learner. The base learners perform regression predictions on deep features, outputting multiple intermediate octane number loss predictions. A least-squares support vector machine is used as the meta-learner for secondary regression prediction, outputting the predicted octane number loss value. This approach covers all operating conditions through multiple base learners and improves prediction accuracy through the meta-learner, while also enhancing the model's adaptability to changes in operating conditions. The quantitative prediction value provides direct guidance for refinery production decisions. Through feature importance analysis and partial dependency graphs, the previously complex "black box" model prediction process becomes transparent and interpretable. It can quantify the influence weight of various process parameters on octane number loss, helping engineers quickly identify key control variables. Simultaneously, it can intuitively display the nonlinear relationship between single or two parameters and octane number loss, revealing their inherent influence patterns. This transforms the model's prediction results from isolated numerical values ​​into a decision-making basis containing causal logic, greatly enhancing the model's credibility, acceptability, and practical value in industrial practice, providing clear guidance for refineries to conduct targeted process optimization.

[0030] In one embodiment, the method further includes the following steps: To ensure the effectiveness and reliability of the model, the present invention rigorously trains and evaluates the model. Real chemical plant data is used for training and testing, and the following evaluation metrics are used to measure model performance (Table 1): A comprehensive evaluation metric of coefficient of determination, mean squared error, and median absolute error is used to evaluate and optimize the model. Specifically, this includes evaluating a single model, a stacked ensemble learning model, and a sparse autoencoder + stacked ensemble learning model separately, and conducting comparative analysis. The coefficient of determination is used to measure the degree of fit between the model's predictions and the actual values. A higher MSE indicates better model prediction accuracy; the mean squared error (MSE) is used to evaluate the sum of squared prediction errors, and a lower MSE indicates that the prediction result is closer to the actual value; the median absolute error (MSE) is used to evaluate the median of the model prediction error, which can effectively reflect the model's stability and robustness. Through the comprehensive evaluation of these indicators, the SAE+StackingEL model demonstrates superior performance compared to the traditional random forest (RF) model. It increased by about 10%, while MSE decreased by about 68%.

[0031] Table 1 Model Evaluation Results

[0032] In one embodiment, the following steps are also included: Based on interpretability analysis, the process parameters with the greatest impact on octane number loss were identified, and optimized operation suggestions for the gasoline refining process were output to promote the implementation of green production processes. Figure 3 As shown, this is a feature importance analysis diagram of the interpretability analysis provided in this embodiment, such as... Figure 4 The diagram shown illustrates partial dependency graph analysis, a schematic diagram of interpretability analysis provided in this embodiment.

[0033] in, Figure 3 The data shows that characteristics 9 (octane number of spent adsorbent) and characteristic 35 (sulfur content of refined gasoline unit products) have the highest PI values, exceeding 0.06 and 0.05, respectively. This is because the RON of spent adsorbent directly reflects the octane quality of the feedstock and is a key determinant of the final gasoline performance, while sulfur content significantly affects combustion characteristics and catalyst deactivation, thus having a significant impact on RON stability. In terms of variable categories, the most influential factor is the characteristics of the spent adsorbent, especially the octane number, as it is a direct indicator of gasoline quality. In addition, operating variables such as filter differential pressure and furnace inlet pressure are also important factors, indicating that process conditions and equipment performance have a significant impact on RON prediction and refining optimization. Figure 4 The following illustrates the nonlinear impact of top-ranked features on RON loss: Feature 9 (RON of spent adsorbent): Its effect is small between -3 and -0.8; between -0.8 and 1.2, RON loss decreases with fluctuations; above 1.2, RON loss decreases sharply. This indicates that higher adsorbent RON represents less deactivation and better sulfur adsorption selectivity. Feature 35 (Sulfur content of treated gasoline): RON loss increases with increasing sulfur content, but the growth trend gradually levels off when the feature value exceeds -1, indicating a saturation effect; further increases in sulfur content do not proportionally exacerbate catalyst poisoning. Feature 265 (Differential pressure of ME-105 filter): RON loss does not change significantly between -2 and -1.3; between -1.3 and 1.2, RON loss increases moderately; above 1.2, RON loss increases sharply. This indicates that once the pressure drop exceeds the threshold, uneven feed distribution will severely affect process performance. Feature 77 (Main Burner Gas Inlet Pressure): Between -2 and -0.6, RON loss initially increases and then decreases slightly; after -0.55, RON loss steadily decreases. This indicates that maintaining a sufficient and stable inlet pressure ensures uniform heating and optimal temperature control, thereby reducing octane degradation.

[0034] In summary, the combined PI and PDP analyses reveal the importance of global characteristics and local nonlinear mechanisms: adsorbent properties are the most decisive factor, while sulfur content and filter / furnace operation play important supporting roles. These insights provide clear optimization directions for refining practices—namely, maintaining adsorbent activity, controlling sulfur migration, minimizing filter resistance, and stabilizing furnace operation to effectively suppress RON loss.

[0035] The method provided in this embodiment ensures the reliability of predictions through high-quality data support, optimized feature extraction, high-precision integrated models, and comprehensive evaluation. It then identifies key influencing process parameters through interpretable analysis and finally outputs optimized operation suggestions for gasoline refining, helping refineries reduce octane number loss and improve production efficiency.

[0036] This embodiment not only focuses on technical performance but also considers its actual contribution to environmental protection and economic benefits. By predicting and optimizing octane number loss, the energy efficiency of refineries can be significantly improved, and pollution emissions reduced. The specific assessment is as follows: Energy optimization: Reducing octane number loss by 0.5 units can reduce energy waste and save significant energy consumption in a refinery with an annual output of 1 million tons.

[0037] Economic benefits: According to the assessment results, reducing RON loss by 0.5 units can bring economic benefits of approximately RMB 75 million, which provides a feasible technical route for refinery production optimization and energy conservation and emission reduction.

[0038] Environmental protection: Optimizing octane number loss helps reduce pollutant emissions, drive green industrial transformation, and promote sustainable development in the petrochemical industry.

[0039] In one embodiment, a sparse autoencoder extracts key features from the preprocessed raw data, removes data redundancy and noise to obtain deep features, specifically including the following steps: The raw data obtained from the refinery included 367 variables. A sparse autoencoder model was used to train these 367 variables. Leveraging the network's compression capabilities, the number of hidden layer neurons and sparsity constraint parameters were adjusted to transform the 367 variables into 31 deep features, such as... Figure 5 The diagram shown illustrates how the sparse autoencoder provided in this embodiment reduces the dimensionality of variables affecting octane number loss. It describes how nonlinear mapping is performed on the features of 367 variables to extract key features. After removing noise and redundant information from the 31 deep features obtained through training, 31 deep features are obtained, which can retain features that significantly affect prediction, thus improving the training efficiency and prediction accuracy of subsequent models.

[0040] System Implementation Examples According to embodiments of the present invention, a reliable gasoline octane number loss prediction system is provided, such as... Figure 6 The diagram shown is a structural schematic of the reliable gasoline octane number loss prediction system provided in this embodiment. The reliable gasoline octane number loss prediction system according to this embodiment includes a data acquisition module 61, a deep feature extraction module 62, a prediction module 63, and an analysis module 64.

[0041] The data acquisition module 61 is used to acquire and preprocess the raw data of the refinery's production process. The raw data includes multiple process parameters that affect the loss of octane number.

[0042] The deep feature extraction module 62 is used to construct and train a sparse autoencoder. The sparse autoencoder extracts key features from the preprocessed raw data and removes data redundancy and noise to obtain deep features.

[0043] Prediction module 63 is used to construct a stacked ensemble learning model. The stacked ensemble learning model includes multiple base learners and a meta learner. The multiple base learners perform regression prediction on deep features and output multiple intermediate octane number loss prediction results. The least squares support vector machine is used as the meta learner to perform secondary regression prediction and output the octane number loss prediction value.

[0044] Analysis module 64 is used to perform interpretability analysis on the prediction process of the stacked ensemble learning model using feature importance analysis and partial dependency graphs.

[0045] The system provided in this embodiment includes a data acquisition module 61 that acquires and preprocesses raw data from the refinery's production process to ensure data integrity and usability, providing high-quality input for subsequent feature extraction and model training, and mitigating prediction bias risks caused by data quality issues from the outset. The deep feature extraction module 62 constructs and trains a sparse autoencoder, which extracts key features from the preprocessed raw data, removing data redundancy and noise to obtain deep features. This allows for feature optimization, generating deep features that better reflect the inherent patterns of octane number loss, thereby improving the upper limit of the accuracy of the subsequent prediction model and reducing model training complexity through dimensionality reduction and redundancy removal, providing "highly efficient and highly correlated" features for the prediction process. The model takes the following inputs: The prediction module 63 constructs a stacked ensemble learning model, which includes multiple base learners and a meta-learner. The base learners perform regression predictions on deep features, outputting multiple intermediate octane number loss predictions. The meta-learner uses a least-squares support vector machine for secondary regression prediction, outputting the predicted octane number loss value. This approach covers all operating conditions through multiple base learners and improves prediction accuracy through the meta-learner, while also enhancing the model's adaptability to changes in operating conditions. The output quantified prediction value provides direct guidance for refinery production decisions. The analysis module 64 uses feature importance analysis and partial dependency graphs to make the originally complex "black box" model prediction process transparent and interpretable. It can quantify the influence weight of each process parameter on octane number loss, helping engineers quickly identify key control variables. Simultaneously, it can intuitively display the nonlinear relationship between one or two parameters and octane number loss, revealing its inherent influence patterns. This makes the model's prediction results no longer isolated numerical values, but a decision-making basis containing causal logic, greatly improving the model's credibility, acceptability, and practical value in industrial practice, providing clear guidance for refineries to conduct targeted process optimization.

[0046] In one embodiment, the system further includes an optimization module for evaluating and optimizing the model's performance using a comprehensive evaluation metric that combines the coefficient of determination, mean square error, and median absolute error.

[0047] In one embodiment, the analysis module 64 is also configured to determine the process parameters that have the greatest impact on octane number loss based on interpretability analysis, and output optimized operation recommendations for the gasoline refining process.

[0048] The system provided in this embodiment ensures prediction reliability through high-quality data support, optimized feature extraction, high-precision integrated models, and comprehensive evaluation. It then identifies key influencing process parameters through interpretable analysis and finally outputs optimized operation suggestions for gasoline refining, helping refineries reduce octane number loss and improve production efficiency.

[0049] The embodiments of the present invention are system embodiments corresponding to the above method embodiments. The specific operations of each module processing step can be understood by referring to the description of the method embodiments, and will not be repeated here.

[0050] like Figure 7 As shown, the present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements the method for predicting the loss of reliable gasoline octane number in the above embodiments, or when the computer program is executed by a processor, it implements the method for predicting the loss of reliable gasoline octane number in the above embodiments.

[0051] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0052] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and the contents not described in detail in the specification of the present invention are known to those skilled in the art.

Claims

1. A method for predicting the loss of reliable gasoline octane number, characterized in that, Includes the following steps: The raw data of the refinery's production process is acquired and preprocessed, including multiple process parameters that affect octane number loss. A sparse autoencoder is constructed and trained. The sparse autoencoder extracts key features from the preprocessed raw data and removes data redundancy and noise to obtain deep features. A stacked ensemble learning model is constructed, which includes multiple base learners and a meta learner. The multiple base learners perform regression prediction on the deep features and output multiple intermediate octane number loss prediction results. The least squares support vector machine is used as the meta learner to perform secondary regression prediction and output the octane number loss prediction value. Feature importance analysis and partial dependency graphs are used to perform interpretability analysis on the prediction process of the stacked ensemble learning model.

2. The method for predicting the loss of reliable gasoline octane number as described in claim 1, characterized in that, It also includes the following steps: The model performance was evaluated and optimized using a comprehensive evaluation index of coefficient of determination, mean square error, and median absolute error.

3. The method for predicting the loss of reliable gasoline octane number as described in claim 1, characterized in that, It also includes the following steps: Based on the interpretability analysis, the process parameters that have the greatest impact on octane number loss are identified, and optimized operation suggestions for the gasoline refining process are output.

4. The method for predicting the loss of reliable gasoline octane number as described in claim 1, characterized in that, The base learners include Lasso regression, Ridge regression, Support Vector Regression, Elastic Net Regression, Naive Bayes, Logistic Regression, Random Forest, Gradient Boosting Regression, Extreme Random Tree, and Extreme Gradient Boosting.

5. The method for predicting the loss of reliable gasoline octane number as described in claim 1, characterized in that, The sparse autoencoder extracts key features from the preprocessed raw data, removes data redundancy and noise to obtain deep features, specifically including the following steps: The original data includes 367 variables; The 367 variables were trained using a sparse autoencoder model. By adjusting the number of neurons in the hidden layer and the sparsity constraint parameters, the 367 variables were transformed into 31 deep features. The features of the 367 variables were nonlinearly mapped to extract key features. After removing noise and redundant information, the 31 deep features were obtained.

6. A reliable gasoline octane number loss prediction system, characterized in that, It includes a data acquisition module, a deep feature extraction module, a prediction module, and an analysis module; The data acquisition module is used to acquire and preprocess raw data of the refinery's production process. The raw data includes multiple process parameters that affect octane number loss. The deep feature extraction module is used to construct and train a sparse autoencoder. The sparse autoencoder extracts key features from the preprocessed raw data and removes data redundancy and noise to obtain deep features. The prediction module is used to construct a stacked ensemble learning model, which includes multiple base learners and a meta-learner. The multiple base learners perform regression prediction on the deep features and output multiple intermediate octane number loss prediction results. The least squares support vector machine is used as the meta-learner to perform secondary regression prediction and output the octane number loss prediction value. The analysis module is used to perform interpretability analysis on the prediction process of the stacked ensemble learning model using feature importance analysis and partial dependency graphs.

7. The reliable gasoline octane number loss prediction system as described in claim 6, characterized in that, It also includes an optimization module, which is used to evaluate and optimize the model's performance using a comprehensive evaluation index of the coefficient of determination, mean square error, and median absolute error.

8. The reliable gasoline octane number loss prediction system as described in claim 6, characterized in that, The analysis module is also configured to determine the process parameters that have the greatest impact on octane number loss based on the interpretability analysis, and output optimized operation suggestions for the gasoline refining process.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting the loss of reliable gasoline octane number as described in any one of claims 1 to 5.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method for predicting the loss of reliable gasoline octane number as described in any one of claims 1 to 5.

Images