Method and system for predicting abnormal working condition of furnace tube of ethylene cracking furnace
By pre-processing and model training of the parameter data of the ethylene cracking furnace furnace tube, combined with iterative training and depth-first search optimization technology, the problems of insufficient data samples and unbalanced samples in the furnace tube working condition prediction are solved, and the accuracy and efficiency of prediction are improved.
Patent Information
- Application Number
- CN202411672683.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-21
- Publication Date
- 2025-03-18
AI Technical Summary
In the working condition prediction of ethylene cracking furnace tubes, there are problems such as insufficient data samples and imbalance of coking operating conditions, which leads to low prediction accuracy of the model under small sample conditions and is easy to overfit.
A method for predicting abnormal working conditions of ethylene cracking furnace furnace tubes is proposed. By acquiring and preprocessing furnace tube parameter data, the prediction model is trained using classification results, and the model is optimized through iterative training and depth-first search, reducing the dependence on a large number of labeled data and improving the model's performance under small sample data.
It improves the accuracy and efficiency of abnormal working conditions prediction of ethylene cracking furnace tubes, reduces the uncertainty brought about by small sample data, and enhances the robustness and generalization ability of the model.
Smart Images
Figure CN119647659A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the fields of petrochemical industry and computer, and more specifically to a method and system for predicting abnormal operating conditions of furnace tubes of an ethylene cracking furnace. Background Art
[0002] Ethylene is the most basic raw material in the petrochemical industry. In the process of producing ethylene by thermal cracking, the production capacity and technology of the ethylene cracking furnace directly determine the production scale, output and product quality of the entire ethylene plant. The research on its production operation optimization is of great significance in improving the production level and economic benefits of ethylene plants. Since carburization and coking are always inevitable in the cracking furnace tubes, if these abnormal conditions are not discovered and handled in time, they may cause damage to the ethylene cracking furnace tubes and affect the safe and stable operation of the device. Therefore, predicting abnormal conditions and taking corresponding measures can improve the production efficiency of the ethylene cracking furnace and achieve safe and reliable production.
[0003] At present, the deep learning-based prediction technology for ethylene cracking furnace tube operating conditions has developed rapidly. However, ethylene cracking furnace tubes have the characteristics of few operating condition data samples in each section between tubes and unbalanced coking condition samples. The coking condition of the tube accounts for a small proportion of the normal condition samples, and under small sample conditions, the expression of traditional ethylene cracking furnace tube operating condition characteristics has strong uncertainty. Summary of the invention
[0004] The present invention aims to overcome at least one defect (shortcoming) of the above-mentioned prior art and provide a method and system for predicting abnormal operating conditions of ethylene cracking furnace tubes, so as to achieve the effect of predicting abnormal operating conditions of ethylene cracking furnace tubes.
[0005] According to a first aspect of the present application, a method for predicting abnormal operating conditions of a furnace tube of an ethylene cracking furnace is provided, comprising:
[0006] Obtaining the parameters of the ethylene cracking furnace tube as sample data;
[0007] Preprocessing the sample data of the furnace tube parameters of the ethylene cracking furnace to obtain preprocessed sample data;
[0008] Classifying the preprocessed sample data to obtain classification results, wherein the classification results include a labeled data set and an unlabeled data set;
[0009] Using the classification results to train the prediction model to obtain a trained prediction model;
[0010] Obtain real-time ethylene cracking furnace tube parameter data;
[0011] The real-time ethylene cracking furnace tube parameter data is preprocessed and then input into the trained prediction model for prediction to obtain a prediction result.
[0012] The method for predicting abnormal operating conditions of ethylene cracking furnace tubes predicts real-time operating conditions of ethylene cracking furnace tubes using known ethylene cracking furnace tube parameters according to a preset prediction model. By training the preset prediction model and optimizing the prediction model, the prediction accuracy of the prediction model can be improved, thereby improving the prediction efficiency of the prediction method, and achieving accurate prediction of real-time operating conditions of ethylene cracking furnace tubes. The method for predicting abnormal operating conditions of ethylene cracking furnace tubes is conducive to timely predicting abnormal operating conditions of ethylene cracking furnace tubes during the production process, and taking corresponding measures for abnormal operating conditions of ethylene cracking furnace tubes to avoid losses caused by abnormal operating conditions of ethylene cracking furnace tubes.
[0013] Optionally, the using the classification result to train the prediction model to obtain a trained prediction model includes:
[0014] Dividing the labeled data set into a training set and a test set;
[0015] Perform the following iterative training:
[0016] Using the training set to train the prediction model to obtain a trained prediction model;
[0017] Inputting the unlabeled data set into the trained prediction model for prediction, and selecting part of the sample data from the unlabeled data set for annotation based on the prediction to obtain the labeled data;
[0018] The labeled data is counted into the training set, and the sample data corresponding to the labeled data is deleted from the unlabeled data set to obtain an updated unlabeled data set;
[0019] The iterative training is cyclically performed using the training set that includes the labeled data and the updated unlabeled data set until an iteration stopping condition is met, thereby obtaining a trained prediction model;
[0020] The trained prediction model is tested using the test set to obtain a final prediction model.
[0021] The prediction model is trained. The trained prediction model can learn deeper feature representation by training on known ethylene cracking furnace tube parameter sample data, and has a certain feature basis, so that it can perform better on specific tasks and reduce dependence on a large amount of labeled data. This allows the model to perform well even when there is less data, reduces the consumption of computing resources and time, and makes the prediction process more efficient.
[0022] Optionally, the iteration stopping condition is that the number of iterative training of the prediction model reaches a preset number of iterations or the number of training sets reaches a preset number.
[0023] The iteration condition plays a vital role in the entire prediction method process. The iteration condition determines when the training loop ends and ensures that the loop body is executed as expected. Appropriate iteration conditions can prevent the training from falling into an infinite loop, reduce unnecessary calculations and memory usage, and thus improve the performance of the program. Through iteration conditions, access to sample resources can be controlled to prevent resource exhaustion. At the same time, iteration conditions can be used to implement complex conditional logic, so that the program can perform different operations according to different conditions. Iteration conditions can be used to detect and handle errors to ensure that the program can respond correctly when encountering an exception. It can also be used to track and update the status of the program to ensure the correctness of the program logic. In summary, iteration conditions are one of the key factors to ensure that the program logic is correct and runs efficiently. The correct use of iteration conditions can make the program more robust, easy to maintain and expand, and play a role in standardizing the prediction model.
[0024] Optionally, the selecting of part of the sample data from the unlabeled data set is performed by adopting a “high uncertainty” sampling strategy.
[0025] The core idea of the "high uncertainty" sampling strategy is that the samples near the decision boundary of the model usually have higher uncertainty, so by labeling these samples, the model can learn the key features and decision boundaries of the data more quickly. Specifically, by evaluating the uncertainty of unlabeled data, selecting the most valuable samples for model retraining, and helping the model to learn quickly, the convergence of the model is accelerated, which helps the model to better understand the data distribution and enhance the generalization ability of the model; selecting samples with high uncertainty for labeling can reduce the amount of labeled data required and reduce the cost of data labeling; focusing on samples with high uncertainty can better handle noise and outliers in the data and improve the robustness of the model itself; at the same time, the high uncertainty sampling strategy can dynamically adjust the sampling strategy according to the current state of the model, so that the model can adapt to the changing data environment. In active learning, a "high uncertainty" sampling strategy is adopted, that is, low-confidence sample data is selected for labeling, or in a statistical framework, samples with the smallest difference between the maximum posterior probability and the second maximum posterior probability are selected. In summary, the high uncertainty sampling strategy can improve the learning efficiency and performance of the model by focusing on samples with high uncertainty, while reducing the cost of data labeling. It is very valuable in active learning and other scenarios that require efficient use of labeled data.
[0026] Optionally, the method for predicting abnormal operating conditions of ethylene cracking furnace tubes further includes:
[0027] During the iterative training, the prediction model is optimized using depth-first search.
[0028] The Depth-First Search (DFS) is a search algorithm widely used in graphs and tree structures. Compared with Breadth-First Search (BFS), the Depth-First Search DFS does not need to store the adjacency list of all nodes during the search process, so its space complexity is usually lower, especially when processing large-scale graphs, and can process infinite graphs; the prediction model is optimized by using the Depth-First Search, and DFS can find all possible solutions, not just one solution. The algorithm logic of DFS is relatively simple and easy to program. DFS can search deeply, which is very useful for finding solutions with greater depth or exploring deep structures. DFS naturally implements path backtracking during the search process, which is very useful when solving problems that require backtracking (such as solving certain types of puzzles). DFS is suitable for scenarios that require depth-first search. By performing a depth-first search on the grid structure of the prediction model, a grid structure with higher performance can be found, thereby optimizing the prediction model and improving the overall prediction level.
[0029] Optionally, optimizing the prediction model by using depth-first search also includes optimizing the network structure of the prediction model, specifically including:
[0030] Setting a search space, a search strategy and an optimization time for the prediction model;
[0031] Optimizing the prediction model based on the search space, search strategy, optimization time and training set, and taking all generated model structures, parameters learned by the model structures and model loss values as optimization history values during the optimization process;
[0032] Based on the optimization history value, the optimal model structure is selected as the trained prediction model.
[0033] The network structure of the prediction model is optimized, and the optimized grid structure can reduce the amount of calculation of the model during training and prediction, thereby improving the overall calculation efficiency. By optimizing the grid structure, the model can better adapt to different data distributions, improve its prediction performance on unknown real-time data, help the model learn more general feature representations on limited data, reduce the risk of overfitting, and make the model more stable in the face of slight changes in input data, reduce the fluctuation of prediction results, and provide better interpretability through the attention mechanism, helping to understand the decision-making process of the model can adapt to different data sets and application scenarios, automatically extract key features in time series data, capture long-term dependencies and pay attention to important historical information, thereby improving the accuracy and stability of predictions, reducing unnecessary parameters, reducing the computational complexity of the model, and improving the performance and applicability of the model in many aspects, making it more suitable for actual prediction tasks.
[0034] Optionally, the generated model structure acquisition method includes:
[0035] generating a new model structure from the search space using an optimized acquisition function algorithm;
[0036] And / or, the parameters learned by the model structure include hyperparameters, and optimizing the prediction model using depth-first search also includes optimizing the hyperparameters.
[0037] generating a new model structure from the search space using the optimized acquisition function algorithm,
[0038] Optionally, the optimizing the hyperparameters specifically includes:
[0039] Performing a hyperparameter analysis on the generated model structure according to the optimization time to obtain a hyperparameter value range of the generated model structure;
[0040] The hyperparameter value of the generated model structure is selected according to the prediction result accuracy value corresponding to the hyperparameter value in the hyperparameter value range.
[0041] Optimizing the hyperparameters can make the model achieve better performance on the training set and the validation set. The optimized hyperparameters can help the model learn generalized features on the training set, avoid overfitting on the training set, and speed up the training of the model, so that the model converges faster. When the model faces different data distributions and noise, it has better robustness. The optimized hyperparameters can reduce the training time and computing resource consumption of the model. Some hyperparameters (such as regularization coefficients) will affect the interpretability of the model. By optimizing these hyperparameters, the prediction results of the model can be more easily explained. By optimizing the hyperparameters, the model can perform better on unknown data and improve the generalization ability of the model. Different data sets may require different hyperparameter settings. Optimizing the hyperparameters can help the model better adapt to different data sets and have better flexibility and stability. Optimizing the hyperparameters is a very important step in the training process of the prediction model. It can help the model better adapt to the data and improve the performance and robustness of the model.
[0042] Optionally, the prediction model includes an input layer, a CNN layer, an Informer layer, a GRU layer, an Attention layer, and a furnace pipe abnormal operating condition output layer connected in sequence, wherein the Informer layer includes an encoder connected to the CNN layer and a decoder connected to the encoder, the output of the decoder is connected to the GRU layer, the encoder is composed of a multi-head probabilistic sparse self-attention mechanism and a self-attention distillation mechanism, and the decoder is composed of an occluded multi-head probabilistic sparse self-attention mechanism and a multi-head self-attention mechanism.
[0043] The prediction model improves the prediction efficiency through the attention mechanism. The encoder in the Informer layer introduces a combination of a multi-head probabilistic sparse self-attention mechanism and a self-attention distillation mechanism. The multi-head probabilistic sparse self-attention mechanism focuses on the important parts of the sequence and improves the ability to extract key information. The self-attention distillation mechanism further optimizes the parameters of the model, making the model more lightweight while maintaining performance. The encoder in the Informer layer introduces a multi-head self-attention mechanism and a probabilistic sparse self-attention mechanism. The multi-head self-attention mechanism runs multiple self-attention layers in parallel, so that the model can simultaneously focus on multiple different features and relationships in the input sequence, thereby enhancing the expression ability of the model and accelerating the prediction process through parallel computing. The probabilistic sparse self-attention mechanism reduces the amount of calculation by selectively focusing on the key parts of the sequence, thereby reducing the complexity of the model. The prediction model combines CNN, RNN and Attention mechanisms, and is suitable for processing complex spatiotemporal data. At the same time, the attention mechanism can provide interpretability for the model and help understand the decision-making process of the model. In addition, compared with the model that only uses CNN or GRU network, the CNN-GRU model combined with attention mechanism has achieved better results in multivariate time series regression prediction tasks. The introduction of attention mechanism can help the model better understand and utilize input data, thereby improving the accuracy and stability of prediction. The prediction model can automatically extract key features in time series data, capture long-term dependencies and pay attention to important historical information, thereby improving the accuracy and stability of prediction.
[0044] According to a second aspect of the present application, a system for predicting abnormal operating conditions of a furnace tube of an ethylene cracking furnace is provided, comprising:
[0045] A first acquisition module is used to acquire the parameters of the ethylene cracking furnace tube as sample data;
[0046] A preprocessing module, used for preprocessing the ethylene cracking furnace tube sample data to obtain preprocessed sample data;
[0047] A classification module, used to classify the preprocessed sample data to obtain a classification result, wherein the classification result includes a labeled data set and an unlabeled data set;
[0048] A training module, used to train the prediction model using the classification results to obtain a trained prediction model;
[0049] The second acquisition module is used to acquire real-time ethylene cracking furnace tube parameter data;
[0050] The prediction module is used to input the real-time ethylene cracking furnace tube parameter data into the trained prediction model after preprocessing to obtain the prediction result.
[0051] The modules in the prediction system cooperate with each other to form a unified whole. The first acquisition module acquires the parameters of the ethylene cracking furnace tube as sample data, and the sample data of the ethylene cracking furnace tube is preprocessed by the preprocessing module to obtain the preprocessed sample data. The classification module classifies the preprocessed sample data to obtain a classification result, and the classification result includes a labeled data set and an unlabeled data set; the prediction model is trained by using the classification module to obtain a trained prediction model so that it has the conditions to process data; the real-time ethylene cracking furnace tube parameter data is acquired by the second acquisition module, and then the real-time ethylene cracking furnace tube parameter data is preprocessed by the prediction module to obtain a prediction result.
[0052] According to a third aspect of the present application, an electronic device is provided, including:
[0053] a memory for storing one or more computer programs;
[0054] The processor implements the method for predicting abnormal operating conditions of furnace tubes of an ethylene cracking furnace as described in the first aspect above when the one or more computer programs are executed by the processor.
[0055] The electronic device provides a complete set of equipment for predicting abnormal operating conditions of the ethylene cracking furnace tubes. The memory stores data, the processor processes the data, and the processed data is then stored in the memory.
[0056] According to a fourth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement a method for predicting abnormal operating conditions of an ethylene cracking furnace tube as described in the first aspect above when executed.
[0057] Based on any of the above aspects, an ethylene cracking furnace tube abnormal condition prediction method and system, electronic device, and computer storage medium provided by the embodiment of the present application, the prediction model in the prediction method extracts features from time series data through convolution operation, and can efficiently extract spatial features of data. At the same time, the GRU layer in the prediction model, as a variant of the RNN layer, solves the problem of gradient vanishing or gradient explosion that is easy to occur in traditional RNN when processing long sequences, and is suitable for modeling time series data. The attention mechanism is also introduced in the prediction model, which can automatically focus on the important parts of the input data, improve the model's ability to capture key information, and enable the model to automatically identify the most important features for the prediction task, and perform weighted fusion on them, thereby improving the interpretability of the model. By combining the advantages of CNN, GRU, and attention mechanism, the model automatically extracts key features in time series data, captures long-term dependencies, and pays attention to important historical information. The prediction model is optimized through depth-first search, and the accuracy of the model's prediction of abnormal conditions of ethylene cracking furnace tubes is improved. At the same time, under the condition of small sample data, the uncertainty generated by small sample data is reduced, and the prediction accuracy and prediction efficiency of the prediction method are improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0059] Figure 1 A flow chart of a method for predicting abnormal operating conditions of a furnace tube of an ethylene cracking furnace provided in this embodiment.
[0060] Figure 2 A schematic diagram of the prediction model structure in a method for predicting abnormal operating conditions of a furnace tube of an ethylene cracking furnace provided in this embodiment.
[0061] Figure 3 A schematic diagram of the functional modules of a system for predicting abnormal operating conditions of furnace tubes of an ethylene cracking furnace provided in this embodiment.
[0062] Figure 4 This is a schematic diagram of the structure of an electronic device provided in this embodiment. DETAILED DESCRIPTION
[0063] The drawings of this application are only used for illustrative purposes and should not be construed as limiting the present application. In order to better illustrate the following embodiments, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; it is understandable to those skilled in the art that some well-known structures and their descriptions in the drawings may be omitted.
[0064] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.
[0065] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0066] At present, the deep learning-based prediction technology for the working condition of ethylene cracking furnace tubes has been rapidly developed. However, the working condition data samples of each section between the tubes of ethylene cracking furnaces are small and the samples of coking working condition are unbalanced. The proportion of coking working condition of the tubes to the normal working condition samples is small, which means that the amount of data available for training the model is small, which limits the ability of the model to learn data distribution and features. It is difficult to extract sufficiently rich features to represent the complexity of the working condition of the tubes, resulting in the inability of the model to accurately predict. The model is also prone to overfitting on the training data, and what is learned is the noise in the training data rather than the potential data distribution, which will lead to a decrease in the generalization ability of the model on new data. The working condition of the tubes of ethylene cracking furnaces may be affected by many factors, such as raw material properties, operating conditions, etc. Changes in these factors may increase the complexity of the working condition of the tubes, and small sample data are difficult to cover all possible working condition changes. In the case of small sample data, the model may not be able to accurately estimate the distribution of the data, resulting in greater uncertainty in the prediction results.
[0067] like Figure 1 As shown, this embodiment provides a method for predicting abnormal operating conditions of a furnace tube of an ethylene cracking furnace, which may include the following steps:
[0068] S110, obtaining parameters of ethylene cracking furnace tubes as sample data;
[0069] In this embodiment, the ethylene cracking furnace tube parameters may include temperature parameters, specifically furnace tube outlet temperature, furnace temperature, furnace tube surface temperature, etc., and may also include pressure parameters, flow parameters, material composition, furnace tube metal distribution operating data, etc.
[0070] In an optional implementation, parameters of an ethylene cracking furnace tube are obtained as sample data, wherein the parameters of the tube include oil raw material parameters of the ethylene cracking furnace cracking raw material, cracking product parameters, cracking furnace related process parameters and operating condition parameters of different tubes. The abnormal operating condition of the ethylene cracking furnace tube is trained using a prediction model based on the parameters, and the obtained parameter categories include but are not limited to these.
[0071] S120, preprocessing the sample data of the ethylene cracking furnace to obtain preprocessed sample data;
[0072] In this embodiment, the sample data of the ethylene cracking furnace tube parameters are preprocessed to obtain preprocessed sample data. The preprocessing can be data cleaning, data conversion, etc. The data cleaning includes processing missing values, processing abnormal values, and processing repeated values; the data conversion includes standardization / normalization, logarithmic transformation, unique hot encoding, data discretization, etc., and the standardization / normalization also includes minimum-maximum normalization, Z-score normalization (standardization) and decimal calibration normalization, etc.
[0073] In an optional implementation, the sample data of the ethylene cracking furnace is preprocessed for data conversion, specifically, the sample data of the tube parameters of the ethylene cracking furnace is normalized. The normalization scales the sample data with different characteristics to a common scale so that the sample data with different characteristics have a balanced impact on the model, which helps the prediction model work more efficiently. The implementation of the preprocessing includes but is not limited to this.
[0074] S130, classifying the preprocessed sample data to obtain a classification result, wherein the classification result includes a labeled data set and an unlabeled data set;
[0075] In this embodiment, the preprocessed sample data is classified to obtain a classification result, and the classification can be based on data type, data distribution characteristics, data dimension, data quality, classification purpose, model type, data stability, etc. Specifically, the data type includes category or label-based data and numerical value-based data, and the numerical value-based data can also be divided into discrete and continuous data; the classification characteristics of the data include normal distribution and skewed distribution; the data dimension includes the number of features and sample size; the data quality includes missing values and outliers; the classification purpose includes predictive classification and descriptive classification; the model type includes supervised learning and unsupervised learning; the data stability includes whether the data changes over time, etc.
[0076] In an optional implementation, the processed sample data is classified according to the data type, and the processed sample data is divided into a labeled data set and an unlabeled data set, and the classification method is not specifically limited.
[0077] S140, training the prediction model using the classification result to obtain a trained prediction model;
[0078] In this embodiment, the prediction model is trained using the classification results. The purpose of training the prediction model is to enable the prediction model to learn the patterns and rules in the data set and improve the accuracy of the prediction. The trained prediction model can adapt to new data sets and environments, so that it can show good prediction performance under different data sets. It can also automate the decision-making process, reduce manual intervention, and improve efficiency. At the same time, the trained prediction model can be expanded to larger data sets or more complex tasks without starting from scratch, reducing the cost of the corresponding operation.
[0079] The prediction model can be trained by supervised learning, that is, using labeled data to train the model; it can also be unsupervised learning, that is, discovering patterns on unlabeled data, such as clustering; it can also be semi-supervised learning, that is, combining a small amount of labeled data with a large amount of unlabeled data or reinforcement learning, that is, learning how to make decisions through interaction with the environment. During the training process, sample data is collected, and sampling strategies include information gain sampling, core set sampling, diversity sampling, representative sampling, etc.
[0080] In an optional implementation, the prediction model is a CNN-Informer-GRU-Attention model, which includes an input layer, a CNN layer, an Informer layer, a GRU layer, an Attention layer, and a furnace pipe abnormal operating condition output layer, and the input layer, CNN layer, Informer layer, GRU layer, Attention layer, and furnace pipe abnormal operating condition output layer are connected in sequence, wherein the Informer layer includes an encoder connected to the CNN layer and a decoder connected to the encoder, the output of the decoder is connected to the GRU layer, the encoder is composed of a multi-head probabilistic sparse self-attention mechanism and a self-attention distillation mechanism, and the decoder is composed of an occluded multi-head probabilistic sparse self-attention mechanism and a multi-head self-attention mechanism.
[0081] In an optional implementation, illustratively, Figure 2 A schematic diagram of a grid structure of a prediction model 210 provided in an embodiment of the present application, wherein the grid structure of the prediction model 210 may include an input layer 211, a CNN layer 212, an Informer layer 213, a GRU layer 214, an Attention layer 215, and a furnace tube abnormal operating condition output layer 216.
[0082] In an optional implementation, the classification result is used to train the prediction model to obtain a trained prediction model, and the classification result includes a labeled data set and an unlabeled data set. Selecting the labeled data set to train the model is supervised learning, which is specifically implemented by dividing the acquired sample data into a labeled data set and an unlabeled data set, and then dividing the labeled data set into a training set and a test set. The training set is first used to train the CNN-Informer-GRU-Attention model to obtain a trained CNN-Informer-GRU-Attention model, and then the unlabeled data set is input into the trained CNN-Informer-GRU-Attention model for prediction. The trained CNN-Informer-GRU-Attention model is trained. The U-Attention model is combined with an active learning algorithm, and the data is manually labeled according to the "high uncertainty" sampling strategy. In the specific implementation process, a confidence level is pre-set, and the confidence level is not specifically limited and is set according to the actual prediction situation; the confidence level corresponding to each sample in the unlabeled data set is obtained; samples in the unlabeled data set whose confidence levels are lower than the preset confidence levels are selected for labeling, and the labeled low-confidence sample data are included in the training set. The labeled sample data will be used as the training set to establish an initial model, and the CNN-Informer-GRU-Attention model will be retrained until the specified number of iterations is reached or the training set reaches the specified number, and finally the trained prediction model is tested using the test set to obtain the final prediction model. The training method includes but is not limited to this.
[0083] In a preferred implementation, during the iterative training, the prediction model is optimized using a depth-first search, specifically including optimizing the network structure of the prediction model: setting a search space, a search strategy, and an optimization time for the prediction model, optimizing the CNN-Informer-GRU-Attention model according to the search space, the search strategy, the optimization time, and the training set, using the training data as the input for the optimization of the CNN-Informer-GRU-Attention model structure, taking all generated model structures, parameters learned by the model structures, and model loss values as optimization history values during the optimization process, generating the next network structure to be observed in the search space by optimizing the acquisition function algorithm, outputting a new model structure and the required network morphism operation, and transforming the existing architecture into a new architecture, such as architecture f i Transformed into a new architecture f i+1(i represents the i-th CNN-Informer-GRU-Attention network structure searched by the optimization algorithm during execution), the new architecture generated is evaluated by the evaluation metric function and its diagnostic performance is obtained to obtain the model loss value, based on all the CNN-BiL STM network structures that have been optimized. The above process is repeated continuously within the determined optimization time, and finally the optimal CNN network structure in the optimization history value is automatically marked, and the optimal CNN network structure belongs to the limited range of the search space.
[0084] In a preferred implementation, the prediction model is optimized using depth-first search, which also includes optimizing the hyperparameters, performing a hyperparameter analysis on the generated model structure according to the optimization time to obtain a hyperparameter value range for the generated model structure; and selecting a hyperparameter value for the generated model structure according to the prediction result accuracy value corresponding to the hyperparameter value in the hyperparameter value range. The optimization method is not specifically limited.
[0085] In a more preferred implementation, the search space sets the depth interval of each layer in the grid structure according to the CNN-Informer-GRU-Attention model, such as the CNN layer is set to [1,3], the Informer layer is set to [1,4], and so on, and the depth interval range of each layer is optimized from top to bottom; the search strategy is to explore the optimal depth from a top CNN layer, start to visit the minimum value of the depth interval of the CNN layer, and then start from the maximum value of the depth interval of the CNN layer that has not been visited, and perform depth-first traversal until all layers in the CNN-Informer-GRU-Attention model are visited. If there are other network layers in the CNN-Informer-GRU-Attention model that have not been visited at this time, then select another network layer in the CNN-Informer-GRU-Attention model that has not been visited as the starting point, repeat the above process, and the whole process is repeated until all layers are visited. The root mean square error (RMSE) is calculated for the corresponding prediction models of all generated grid structures, and the network structure with the smallest root mean square error (RMSE) is selected as the final model structure. The setting of the optimization time is not specifically limited, and the search space and the search strategy include but are not limited to this.
[0086] S150, obtaining real-time ethylene cracking furnace tube parameter data;
[0087] In this embodiment, the real-time ethylene cracking furnace tube parameters may include temperature parameters, specifically tube outlet temperature, furnace temperature, tube surface temperature, etc., and may also be pressure parameters, flow parameters, material composition, tube metal distribution operating data, etc.
[0088] In an optional implementation, real-time parameters of ethylene cracking furnace tubes are obtained as sample data, and the parameters of the tubes include oil raw material parameters of the ethylene cracking furnace cracking raw materials, cracking product parameters, cracking furnace related process parameters and operating parameters of different tubes. The abnormal operating conditions of the ethylene cracking furnace tubes are predicted using a prediction model based on the parameters, and the obtained parameter categories include but are not limited to these.
[0089] S160, preprocessing the real-time ethylene cracking furnace tube parameter data and inputting the preprocessed data into the trained prediction model to obtain a prediction result.
[0090] In this embodiment, the prediction model can be a linear regression model, a logistic regression model, a neural network model, a decision tree model, a random forest model, etc.
[0091] In a specific implementation, the prediction model is used to predict the real-time ethylene cracking furnace tube parameter data, and the real-time ethylene cracking furnace tube parameter data is input into the prediction model. The input layer in the prediction model inputs the original data into the network. The real-time ethylene cracking furnace tube parameter data will be normalized in the preprocessing to obtain data in a unified dimension, so that the model can learn and process better. The CNN layer extracts the features of the normalized data through convolution operations to capture local patterns and trends in time series data. The neurons of the convolution layer are only connected to a part of the input data to achieve local perception and parameter sharing, reduce model parameters and prevent overfitting. The Informer layer processes sequence data through a self-attention mechanism to capture long-distance dependencies. Effectively process and analyze the long-term operation data of the furnace tube, identify the patterns that may cause anomalies, the GRU layer helps the model understand the time series changes of the furnace tube operation status, the Attention layer enhances the model's attention to key information by assigning different weights to different parts of the sequence, so that the model pays more attention to those features and patterns that are more important for predicting the abnormal working conditions of the furnace tube, and finally the abnormal working condition output layer of the furnace tube converts the output of the previous layer into the final prediction result, such as the output layer usually uses one or more neurons to represent different abnormal categories or states, and outputs the probability of each category through an activation function (such as soft max), and the prediction process of the prediction model includes but is not limited to this.
[0092] The following describes a technical solution of a method for predicting abnormal operating conditions of a furnace tube of an ethylene cracking furnace provided by the present application in conjunction with a specific embodiment.
[0093] like Figure 3 As shown, the embodiment of the present application also provides an ethylene cracking furnace tube abnormal operating condition prediction system. Optionally, the ethylene cracking furnace tube abnormal operating condition prediction system may include:
[0094] The first acquisition module 311 is used to acquire the parameters of the ethylene cracking furnace tube as sample data;
[0095] In this embodiment, the first acquisition module 311 can be used to execute Figure 1 As shown in step S110, for a detailed description of the first acquisition module 311, reference may be made to the description of step S110.
[0096] A preprocessing module 312 is used to preprocess the sample data of the ethylene cracking furnace tube operating conditions to obtain preprocessed sample data;
[0097] In this embodiment, the preprocessing module 312 can be used to perform Figure 1 As shown in step S120, for the detailed description of the preprocessing module 312, reference may be made to the description of step S120.
[0098] A classification module 313, used to classify the preprocessed sample data to obtain a classification result, wherein the classification result includes a labeled data set and an unlabeled data set;
[0099] In this embodiment, the classification module 313 can be used to perform Figure 1 As shown in step S130, for the detailed description of the classification module 313, reference may be made to the description of step S130.
[0100] A training module 314, configured to train the prediction model using the classification result to obtain a trained prediction model;
[0101] In this embodiment, the training module 314 can be used to perform Figure 1 As shown in step S140, for the specific description of the training module 314, please refer to the description of step S140.
[0102] The second acquisition module 315 is used to acquire real-time ethylene cracking furnace tube parameter data;
[0103] In this embodiment, the second acquisition module 315 can be used to perform Figure 1As shown in step S150, for the specific description of the second acquisition module 315, reference may be made to the description of the step S150.
[0104] The prediction module 316 is used to input the real-time ethylene cracking furnace tube parameter data into the trained prediction model after preprocessing to obtain a prediction result.
[0105] In this embodiment, the prediction module 316 may be used to perform Figure 1 As shown in step S160, for the detailed description of the prediction module 316, reference may be made to the description of step S160.
[0106] It can be understood that the above-mentioned device embodiment and the above-mentioned method embodiment can correspond to each other, and similar descriptions of the device embodiment can refer to the method embodiment. To avoid repetition, it will not be repeated here. An ethylene cracking furnace tube abnormal operating condition prediction system provided in the embodiment of the present application can execute an ethylene cracking furnace tube abnormal operating condition prediction method provided in any embodiment of the present application, and has the corresponding functional modules and beneficial effects of the execution method. The functional modules of the ethylene cracking furnace tube abnormal operating condition prediction system can be implemented in the form of hardware, can be implemented by instructions in the form of software, and can also be implemented by a combination of hardware and software modules.
[0107] Specifically, each step of the method embodiment of the present application can be completed by the hardware integrated logic circuit and / or software instructions in the processor. The steps of the abnormal operating condition prediction method of the ethylene cracking furnace tube in the embodiment of the present application can be directly reflected as a hardware coding processor for execution, or can be executed by a combination of hardware and software modules in the coding processor.
[0108] Optionally, the software module may be located in a random access memory, a read-only memory, a programmable read-only memory, a flash memory, an electrically erasable programmable memory, a register, or other storage media. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps in the above method embodiment in combination with its hardware.
[0109] The present application embodiment provides an electronic device 410, whose structure is as follows: Figure 4 As shown, the electronic device 410 includes a memory 411, a processor 412, a communication module 413 and an input / output interface 414, etc. Optionally, the memory 411, the processor 412, the communication module 413 and the input / output interface 414 can be connected and communicated through a bus 415.
[0110] The memory 411 is used to store one or more computer programs and transmit the code of the computer program to the processor 412; when the one or more computer programs are executed by the processor 412, a method for predicting abnormal operating conditions of an ethylene cracking furnace tube in an embodiment of the present application is implemented.
[0111] Optionally, the electronic device 410 can be connected to a network via a communication module 413 to communicate with other devices, such as a terminal or a server, through the network to achieve data interaction. The electronic device 410 can be various forms of digital computers, such as desktop computers, servers, workstations, mainframe computers or other types of computers. The electronic device 410 can also be various forms of mobile terminals, such as smart phones, tablet computers, wearable devices (such as helmets, glasses, watches, etc.) and other similar mobile terminals.
[0112] Optionally, the electronic device 410 can be connected to the required input / output devices, such as a keyboard, a display device, etc., through the input / output interface 414. The electronic device 410 itself can have a display device, and can also be connected to other display devices through the input / output interface 414. Optionally, a storage device, such as a hard disk, can also be connected through the input / output interface 414, so that data in the electronic device 410 can be stored in the storage device, or data in the storage device can be read, and data in the storage device can also be stored in the memory 411. It can be understood that the input / output interface 414 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 414 can be a component of the electronic device 410, or it can be an external device connected to the electronic device 410 when needed.
[0113] Optionally, the memory 411 may be a volatile memory and / or a non-volatile memory, the volatile memory may be a random access memory, etc., and the non-volatile memory may be a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory or a flash memory, etc.
[0114] Optionally, the computer program stored in the processor 412 may be divided into one or more modules, which are stored in the memory 411 and executed by the processor 412 to complete the method provided in the embodiment. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the computer program instruction segments are used to describe the execution process of the computer program in the electronic device 410.
[0115] Optionally, the processor 412 may be various general and / or special processing components with processing and computing capabilities. Some examples of the processor 412 include, but are not limited to, a central processing unit, a graphics processing unit, a digital signal processor, various special artificial intelligence computing chips, various processors running machine learning model algorithms, and may also be any appropriate controller, microcontroller, processor, etc. The processor 412 executes various methods and processes of this embodiment, illustratively, such as a method for predicting abnormal operating conditions of an ethylene cracking furnace tube in an embodiment of the present application.
[0116] Optionally, the bus 415 may include a path for transmitting information. The bus 415 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. According to different functions, the bus 415 may be divided into an address bus, a data bus, a control bus, etc.
[0117] In an optional implementation, the embodiment of the present application further provides a computer storage medium on which a computer program is stored, and when the computer program is executed by a computer, the computer can perform the method of the above method embodiment. Part or all of the computer program can be loaded and / or installed on the memory 411 of the electronic device 410. When the computer program is executed by the processor 412, one or more steps of the method for predicting abnormal operating conditions of an ethylene cracking furnace tube in the embodiment of the present application can be performed.
[0118] Optionally, the computer-readable storage medium may be a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or the like.
[0119] Obviously, the above embodiments of the present application are merely examples for clearly illustrating the technical solution of the present application, and are not intended to limit the specific implementation methods of the present application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of the present application shall be included in the protection scope of the claims of the present application.
Claims
1. A method for predicting abnormal operating conditions of ethylene cracking furnace tubes, characterized in that: The method comprises: Obtaining the parameters of the ethylene cracking furnace tube as sample data; Preprocessing the sample data of the furnace tube parameters of the ethylene cracking furnace to obtain preprocessed sample data; Classifying the preprocessed sample data to obtain classification results, wherein the classification results include a labeled data set and an unlabeled data set; Using the classification results to train the prediction model to obtain a trained prediction model; Obtain real-time ethylene cracking furnace tube parameter data; The real-time ethylene cracking furnace tube parameter data is preprocessed and then input into the trained prediction model for prediction to obtain a prediction result.
2. The method for predicting abnormal operating conditions of an ethylene cracking furnace tube according to claim 1, characterized in that: The step of training the prediction model using the classification result to obtain the trained prediction model includes: Dividing the labeled data set into a training set and a test set; Perform the following iterative training: Using the training set to train the prediction model to obtain a trained prediction model; Inputting the unlabeled data set into the trained prediction model for prediction, and selecting part of the sample data from the unlabeled data set for annotation based on the prediction to obtain the labeled data; The labeled data is counted into the training set, and the sample data corresponding to the labeled data is deleted from the unlabeled data set to obtain an updated unlabeled data set; The iterative training is cyclically performed using the training set that includes the labeled data and the updated unlabeled data set until an iteration stopping condition is met, thereby obtaining a trained prediction model; The trained prediction model is tested using the test set to obtain a final prediction model.
3. The method for predicting abnormal operating conditions of an ethylene cracking furnace tube according to claim 2, characterized in that: The iteration stopping condition is that the number of iterative training of the prediction model reaches a preset number of iterations or the number of training sets reaches a preset number.
4. The method for predicting abnormal operating conditions of an ethylene cracking furnace tube according to claim 2, characterized in that: The selection of part of the sample data from the unlabeled data set is performed using a "high uncertainty" sampling strategy.
5. The method for predicting abnormal operating conditions of an ethylene cracking furnace tube according to claim 2, characterized in that: The method further comprises: During the iterative training, the prediction model is optimized using depth-first search.
6. A method for predicting abnormal operating conditions of ethylene cracking furnace tubes according to claim 5, characterized in that: Optimizing the prediction model using depth-first search also includes optimizing the network structure of the prediction model, specifically including: Setting a search space, a search strategy and an optimization time for the prediction model; Optimizing the prediction model based on the search space, search strategy, optimization time and training set, and taking all generated model structures, parameters learned by the model structures and model loss values as optimization history values during the optimization process; Based on the optimization history value, the optimal model structure is selected as the trained prediction model.
7. A method for predicting abnormal operating conditions of ethylene cracking furnace tubes according to claim 6, characterized in that: The generated model structure acquisition method comprises: generating a new model structure from the search space using an optimized acquisition function algorithm; And / or, the parameters learned by the model structure include hyperparameters, and optimizing the prediction model using depth-first search also includes optimizing the hyperparameters.
8. A method for predicting abnormal operating conditions of ethylene cracking furnace tubes according to claim 7, characterized in that: The optimizing of the hyperparameters specifically includes: Performing a hyperparameter analysis on the generated model structure according to the optimization time to obtain a hyperparameter value range of the generated model structure; The hyperparameter value of the generated model structure is selected according to the prediction result accuracy value corresponding to the hyperparameter value in the hyperparameter value range.
9. A method for predicting abnormal operating conditions of an ethylene cracking furnace tube according to any one of claims 1 to 7, characterized in that: The prediction model includes an input layer, a CNN layer, an Informer layer, a GRU layer, an Attention layer, and a furnace tube abnormal operating condition output layer connected in sequence, wherein the Informer layer includes an encoder connected to the CNN layer and a decoder connected to the encoder, the output of the decoder is connected to the GRU layer, the encoder is composed of a multi-head probabilistic sparse self-attention mechanism and a self-attention distillation mechanism, and the decoder is composed of an occluded multi-head probabilistic sparse self-attention mechanism and a multi-head self-attention mechanism.
10. An ethylene cracking furnace tube abnormal operating condition prediction system, characterized in that: include: A first acquisition module is used to acquire the parameters of the ethylene cracking furnace tube as sample data; A preprocessing module, used for preprocessing the ethylene cracking furnace tube sample data to obtain preprocessed sample data; A classification module, used to classify the preprocessed sample data to obtain a classification result, wherein the classification result includes a labeled data set and an unlabeled data set; A training module, used to train the prediction model using the classification results to obtain a trained prediction model; The second acquisition module is used to acquire real-time ethylene cracking furnace tube parameter data; The prediction module is used to input the real-time ethylene cracking furnace tube parameter data into the trained prediction model after preprocessing to obtain the prediction result.