A method and system for optimizing refining and chemical processing units based on big data.

Abstract

This invention relates to a method and system for process optimization of refining and chemical plants based on big data, comprising the following steps: acquiring the current operating conditions and process reference number data of the refining and chemical plant; based on the current operating conditions and process reference number data, selecting appropriate excellent parameter points from a pre-established excellent parameter recommendation table for different operating conditions and outputting them to optimize the process parameters of the refining and chemical plant under the current operating conditions. The excellent parameter recommendation table is obtained by analyzing a large amount of historical data from the refining and chemical plant based on K-means clustering and SVDD methods, and is continuously updated using the latest dataset. This invention has advantages such as strong practicality and wide applicability, and has excellent application prospects and commercial value, and can be widely applied in the field of refining and chemical plant process optimization technology.

Description

Technical Field

[0001] This invention relates to the field of refining and chemical plant process optimization technology, and in particular to a method and system for refining and chemical plant process optimization based on big data. Background Technology

[0002] With the development of computer technology and the continuous expansion of its application areas, a large amount of historical data has gradually accumulated in enterprise information systems, covering all aspects of the manufacturing process. Currently, the informatization of most domestic enterprises is still mainly at the management level, focusing only on data collection, storage, and retrieval, and cannot meet the needs of data analysis and processing.

[0003] Data clustering is a common and efficient method for processing large datasets. Its purpose is to determine a finite set of categories to describe the dataset based on the similarity between data objects. As an analytical process that divides a set of objects into multiple classes composed of similar objects based on their characteristics, clustering is an unsupervised learning method that applies a certain standard or the inherent properties and patterns of the data to a sample dataset. It is highly effective in handling the complex and numerous process parameters in production processes.

[0004] As a powerful analytical tool in data mining, cluster analysis generally has two uses:

[0005] (1) As an independent data mining tool, it discovers the distribution characteristics of data;

[0006] (2) As a data preprocessing step for other data analysis methods, it provides other methods with data that has been grouped based on a certain pattern, so that other methods can further perform professional analysis on the corresponding data partitioning results.

[0007] Currently, cluster analysis has been successfully applied in many fields, including image processing, pattern recognition, business, biology, geography, web services, and information retrieval. Through data cluster analysis, information hidden within a large amount of seemingly chaotic data can be concentrated, extracted, and refined to identify the inherent patterns of the research object and uncover potential patterns. It can also help businesses adjust market policies, reduce risks, rationally face the market, and make correct decisions. Furthermore, it can assist governments in adjusting future management policies and economic structures, and proactively responding to ecological development.

[0008] Support Vector Description (SVDD), as an emerging intelligent algorithm, works by finding the most suitable hypersphere that contains all data points from a set of sample data distributed in the feature space. This hypersphere serves as the decision boundary to classify target and anomalous samples, thereby enabling abnormal behavior detection. SVDD is an extension of the traditional SVM algorithm and has been widely applied in fault detection, early warning, intrusion detection, and process optimization.

[0009] However, high sample size and data dimensionality require significant computational resources, which is detrimental to process optimization and anomaly detection in refining and chemical plants. By combining K-means clustering with the SVDD method, the K-means algorithm can perform preliminary processing of large amounts of data. Then, SVDD hypersphere training is performed on data from different categories. By extracting the data segments with optimal operating conditions, the optimal process parameter values ​​are obtained. This combination of methods reduces the complexity of the algorithm within the hypersphere and also lowers the classification error rate, thus providing a guarantee for process optimization in refining and chemical plants.

[0010] In actual production, the manufacturing process of chemical products is relatively complex, and the accuracy of various process parameters directly affects the quality of the finished product. Currently, refining and chemical plants face the following problems in improving product quality:

[0011] (1) There is a lot of historical data in the refining and chemical equipment of various enterprises, but this data has not been used reasonably to improve the product quality of the refining and chemical equipment and improve the economic benefits of the enterprises.

[0012] (2) Each refining and chemical unit of the enterprise has different production needs under different production plans. In the actual production process, there are multiple working conditions. It is difficult for humans to determine the optimal values ​​of key parameters under different working conditions in the refining and chemical unit.

[0013] (3) How to analyze the key process parameters of refining and chemical equipment through a large amount of historical data and obtain the optimal industrial parameter values;

[0014] (4) How to ensure that enterprises can acquire and accumulate the optimal values ​​of key parameters under different working conditions in the long term, and ensure the long-term effectiveness of training results. Summary of the Invention

[0015] To address the aforementioned problems, the purpose of this invention is to provide a method and system for optimizing refining and chemical equipment processes based on big data. This method utilizes a large amount of historical data from refining and chemical equipment within an enterprise to optimize the refining and chemical equipment processes, providing a guarantee for process optimization and a reliable basis for factories to pursue higher product quality and economic benefits.

[0016] To achieve the above objectives, the present invention adopts the following technical solution:

[0017] In a first aspect, the present invention provides a method for optimizing the process of a refining and chemical plant based on big data, comprising the following steps:

[0018] Obtain the current operating status and process tag data of the refining and chemical unit from the upper-level cloud platform or DCS;

[0019] Based on the current operating conditions and process tag number data, appropriate excellent parameter points are selected from the pre-established excellent parameter recommendation table for different operating conditions and output to optimize the process parameters of the refining and chemical unit under the current operating conditions.

[0020] Furthermore, the recommended optimal parameters for different operating conditions are obtained by training K-means and SVDD hyperspheres using historical data from the enterprise's refining and chemical units, including:

[0021] 1.1) Collect historical data from the enterprise's refining and chemical units and preprocess it, then segment the data matrix to form an input data matrix;

[0022] 1.2) For each data segment's input data matrix, after classification using the improved K-means clustering method, the SVDD method is used for hypersphere training, and abnormal hyperspheres are discarded in conjunction with visualization analysis to obtain different hyperspheres and typical process parameter points with timestamps.

[0023] 1.3) Determine whether the number of typical process parameters selected by segmentation meets the preset conditions. If it does, it means that the number of segments is reasonable. After training all segments, proceed to step 1.4). Otherwise, reselect the number of segments and return to step 1.1) until the preset conditions are met.

[0024] 1.4) Based on the obtained typical process parameter points, artificial mechanism fusion and screening are carried out;

[0025] 1.5) Output the required quality indicators, corresponding operating data, and corresponding excellent process reference points, and store them in the excellent parameter recommendation table.

[0026] Furthermore, in step 1.1), historical data from the enterprise's refining and chemical processing units is collected and preprocessed, and the data is segmented to form an input data matrix, including:

[0027] Historical data were collected based on key process variables related to product quality, as determined by on-site experience.

[0028] Select the time period during which all key process variables are effectively sampled;

[0029] Determine the initial number of segments and segment the data accordingly;

[0030] For each data segment, the output measurement values ​​of each key process variable at the same time are combined into a vector to form an input data matrix.

[0031] Further, in step 1.2), for each segment of the input data matrix, after classification using the improved K-means clustering method, hypersphere training is performed using the SVDD method, and abnormal hyperspheres are discarded in conjunction with visualization analysis, resulting in different hyperspheres and typical process parameter points with timestamps, including:

[0032] An improved K-means clustering method is used to select the optimal number of clusters and to classify and label the input data matrix.

[0033] The SVDD method is used to train a hypersphere on each category of data to extract data features under different working conditions;

[0034] Visualize and analyze the hyperspheres in each data segment, discarding hyperspheres with less than the preset value of internal data or those with abnormal visualization, and retaining the remaining normal hyperspheres.

[0035] All preprocessed data were re-substituted into the retained hypersphere, and the median method was used to extract representative optimal parameters for different operating conditions, which were then used as typical process parameter points.

[0036] Furthermore, in step 1.4), the artificial mechanism fusion and screening based on the obtained typical process parameter points refers to selecting process parameter points that meet preset quality requirements by matching the extracted typical process points with timestamps, operating conditions, and quality data, and storing them in an excellent parameter recommendation table, including:

[0037] Product quality indicators and tag data anomaly screening: If the product quality fluctuates more than the preset value near the timestamp of the process point, the typical process point will be discarded; if the tag data of the process point is abnormal, that is, 0 or negative, the typical process point will also be discarded.

[0038] Typical process point similarity comparison: The tag data of typical process points are compared pairwise. If the similarity is greater than the preset value, their quality indicators are compared. If there is a difference in the quality indicators, the fluctuation of the tag data near the timestamp of the two typical process points is observed to see if it is greater than the preset value. If the fluctuation of the tag data of one process point is greater than the preset value, the typical process point is discarded. If the fluctuation of the tag data of both process points is greater than the preset value or both are less than the other preset value, both are discarded.

[0039] Furthermore, based on the current operating conditions and process tag number data, the process parameters of the refining and chemical unit are optimized by selecting appropriate optimal parameter points from a pre-established optimal parameter recommendation table for different operating conditions, including:

[0040] The current DCS process tag number data is obtained through the interface, and the current operating condition is identified.

[0041] The optimal process parameter points that can be selected under the current operating conditions are obtained from the recommended optimal parameter table.

[0042] Calculate the distance between the currently running parameters and all selectable optimal process parameter points under its operating conditions, and extract the optimal process parameter point with the smallest distance for output hot backup.

[0043] Furthermore, the method includes: acquiring a dataset of operating conditions and process reference numbers within a preset time period, expanding and updating the hypersphere cluster based on the acquired dataset, and obtaining an updated recommended table of excellent parameters for different operating conditions.

[0044] Furthermore, the process involves acquiring a dataset of operating conditions and process reference numbers within a preset time period, expanding and updating the hypersphere cluster based on the acquired dataset, and obtaining an updated recommended table of excellent parameters for different operating conditions, including:

[0045] 3.1) Obtain the DCS tag number dataset within a preset time period and preprocess it to obtain the latest dataset;

[0046] 3.2) Check whether the latest dataset falls into an existing hypersphere. If it does, proceed to step 3.3); otherwise, proceed to step 3.4.

[0047] 3.3) Determine which existing hypersphere the latest dataset falls into, and determine whether the data volume of the hypersphere is less than the set threshold. If so, add the latest dataset into the hypersphere and retrain the hypersphere to obtain new hypersphere features and extract new typical process points; otherwise, do not update.

[0048] 3.4) Record the latest dataset and form a new input data matrix, perform hypersphere cluster training, and store the extracted process parameter points in the excellent parameter recommendation table.

[0049] Secondly, the present invention provides a refining and chemical plant process optimization system based on big data, comprising:

[0050] The data acquisition module is used to acquire the current operating conditions and process tag data of the refining and chemical unit;

[0051] The parameter optimization module is used to select appropriate excellent parameter points from a pre-established table of recommended excellent parameters for different operating conditions based on the current operating conditions and process tag number data, and to output them to optimize the process parameters of the refining and chemical unit under the current operating conditions.

[0052] Thirdly, the present invention provides a computer-readable storage medium for storing one or more programs, said one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any of the methods.

[0053] Fourthly, the present invention provides a computing device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods.

[0054] The present invention has the following advantages due to the adoption of the above technical solutions:

[0055] (1) This invention is based on the K-means clustering-SVDD method to analyze a large amount of historical data in the refining and chemical plant, and effectively utilizes the enterprise's historical data storage.

[0056] (2) An improved K-means algorithm is used to classify and label the massive data under different working conditions. By plotting the trend of the number of clusters and the loss function, the problem of inaccurate manual setting of the number of clusters is avoided, the manual workload is reduced, and the accuracy of the clustering results is improved.

[0057] (3) The SVDD hypersphere is trained one by one using the clustering classification results. The combination of the two methods reduces the complexity of the algorithm on the one hand and improves the accuracy of data feature extraction on the other.

[0058] (4) By adopting the method of algorithm training and mechanism fusion and screening, and data segmentation, it is possible to extract the working conditions with good product quality in greater detail, and extract the optimal parameters of the best working conditions from them, so as to provide reliable guidance for enterprise production.

[0059] (5) Hypersphere visualization analysis provides a clearer display of data characteristics for actual production, provides conditions for in-depth data mining, and provides a clearer human-computer interaction mode.

[0060] (6) Online updates and optimizations ensure the effective expansion and updates of the hypersphere cluster, maintain the comprehensiveness and integration of the hypersphere cluster, and thus ensure that the hypersphere cluster can cover a variety of working conditions and various optimization points, ultimately ensuring the long-term effectiveness of the method.

[0061] Therefore, this invention has the advantages of strong practicality and wide applicability, and has excellent application prospects and commercial value. It can be widely applied in the field of process optimization technology for refining and chemical equipment. Attached Figure Description

[0062] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings:

[0063] Figure 1 This is a flowchart of a big data-based process optimization method for refining and chemical equipment provided in an embodiment of the present invention;

[0064] Figure 2 This is a flowchart of the offline training process for process optimization provided in an embodiment of the present invention;

[0065] Figure 3 This is a schematic diagram of the data acquisition and preprocessing method provided in an embodiment of the present invention;

[0066] Figure 4 This is a flowchart of the algorithm training part provided in an embodiment of the present invention;

[0067] Figure 5 This is a structural diagram of the SVDD algorithm provided in an embodiment of the present invention, wherein triangles represent abnormal data, circles represent support vectors, and squares represent normal data;

[0068] Figure 6 This is a schematic diagram of the optimal parameter extraction method provided in an embodiment of the present invention;

[0069] Figure 7 This is a flowchart of the online process optimization recommendation function provided in the embodiments of the present invention;

[0070] Figure 8 This is a flowchart of the online process optimization update function provided in the embodiments of the present invention;

[0071] Figure 9 This is a diagram showing the preliminary training K value and loss function selection provided in this embodiment of the invention;

[0072] Figure 10 This is a hypersphere visualization distribution map of the data in segment 5_2 provided in this embodiment of the invention;

[0073] Figure 11 This is a hypersphere visualization distribution map of the data in segment 3_2 provided in this embodiment of the invention;

[0074] Figure 12This is a hypersphere visualization distribution map of the data in segment 3_3 provided in the embodiments of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0076] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0077] In some embodiments of the present invention, a method for optimizing refining and chemical plant processes based on big data is provided, mainly divided into two parts: offline training and online recommendation and update optimization. In the offline training part, a large amount of historical data from the enterprise's refining and chemical plant is used and reasonably segmented. K-means clustering is first performed on each segment, followed by SVDD hypersphere training for each data category. Then, visual analysis is performed to discard outliers and hyperspheres with a small number of instances. Typical process points are then extracted from the retained hyperspheres. High-quality process parameter points are selected through manual mechanism fusion and screening and stored in an excellent parameter recommendation table. In the online recommendation and update optimization part, using the acquired current operating conditions and process reference number data, suitable excellent parameter points are selected from the excellent parameter recommendation table for output. This provides a basis for on-site personnel operations and guides production. Simultaneously, the excellent parameter recommendation table is updated using data from a preset time period, ensuring the effective expansion and updating of the hypersphere cluster, maintaining the comprehensiveness and integration of the hypersphere cluster, and providing a reliable basis for the plant to pursue higher product quality and economic benefits.

[0078] Correspondingly, in other embodiments of the present invention, a refining and chemical plant process optimization system, equipment, and storage medium based on big data are provided.

[0079] Example 1

[0080] like Figure 1 As shown, this embodiment provides a method for optimizing the refining and chemical processing equipment based on big data, which includes the following steps:

[0081] 1) Offline training: Data segmentation is performed using a large amount of historical data from the refining and chemical units in the enterprise, and K-means and SVDD hypersphere training is carried out. After hypersphere screening through visualization analysis, hypersphere clusters and their data characteristics are obtained. Then, combined with artificial mechanism fusion and screening, a recommended table of excellent parameters for different working conditions is obtained.

[0082] 2) Online recommendation: Based on the acquired current operating conditions and process tag number data, select appropriate excellent parameter points from the excellent parameter recommendation table for different operating conditions obtained in step 1) and output them to optimize the process parameters of the refining and chemical unit under the current operating conditions.

[0083] In some implementations, the method further includes the following steps:

[0084] 3) Online update: Obtain the dataset of working conditions and process reference numbers within a preset time period, expand and update the hypersphere cluster in step 1) based on the obtained dataset, and obtain the updated recommended table of excellent parameters for different working conditions.

[0085] In some implementations, such as Figure 2 As shown, step 1) above includes the following steps:

[0086] 1.1) Collect historical data from the enterprise's refining and chemical units, preprocess the data, and then segment the data matrix appropriately to form an input data matrix;

[0087] 1.2) For each data segment, the input data matrix is ​​used for algorithm training. That is, after classification using the improved K-means clustering method, the hypersphere is trained using the support vector description (SVDD) method. Abnormal hyperspheres are discarded by combining visualization analysis to obtain different hyperspheres and typical process parameter points with timestamps.

[0088] 1.3) Determine whether the number of typical process parameters selected by segmentation meets the preset condition, that is, not less than the number of quality index distribution intervals of the segment. If it meets the condition, it means that the number of segments is reasonable. After training all segments, proceed to step 1.4). Otherwise, reselect the number of segments and return to step 1.1) until the preset condition is met.

[0089] 1.4) Based on the obtained typical process parameter points, artificial mechanism fusion and screening are carried out;

[0090] 1.5) Output the required quality indicators, corresponding operating data, and corresponding excellent process reference points, and store them in the excellent parameter recommendation table.

[0091] In some implementations, in step 1.1) above, offline training should select long-term historical data of key process parameters in the refining and chemical unit, and the collected data should be in a steady state to avoid affecting the training results. Since different production plans have different production needs in the actual production process, and the actual production process includes multiple operating conditions, it is possible to analyze a large amount of historical data—including data from different production plans and operating conditions—to find multiple possibilities for optimal parameters.

[0092] like Figure 3 As shown, specifically, it includes the following steps:

[0093] 1.1.1) For refining and chemical units, historical data on key process variables related to product quality are collected based on on-site experience.

[0094] In this embodiment, the key process variables identified mainly include the following three categories:

[0095] ① Long-term historical data of key process variables, including output measurements such as flow rate, liquid level, pressure, and temperature;

[0096] ② Technological indicators for determining product quality, such as color, selectivity, and conversion rate;

[0097] ③ Indicators that determine process load and operating conditions, such as hydrogen peroxide and catalyst concentrations;

[0098] 1.1.2) Determine the initial number of segments and divide the data into appropriate segments;

[0099] 1.1.3) For each data segment, select the time period in which all key process variables are effectively sampled, and combine the output measurement values ​​of each key process variable at the same time into a vector to form the input data matrix.

[0100] In some implementations, in step 1.2) above, such as Figure 4 As shown, it includes the following steps:

[0101] 1.2.1) An improved K-means clustering method is adopted to select the optimal number of clusters and to classify and label the input data matrix.

[0102] 1.2.2) The SVDD method is used to train a hypersphere for each category of data and extract the data features of each hypersphere cluster, i.e., different working conditions.

[0103] 1.2.3) Visualize and analyze the hyperspheres in each data segment, discarding hyperspheres with small internal data or abnormal visualization, and retaining the remaining normal hyperspheres;

[0104] 1.2.4) Substitute all the data back into the retained hypersphere and use the median method to extract the representative optimal parameters for different operating conditions as typical process parameter points.

[0105] In some implementations, in step 1.2.1) above, since the amount of data in the refining and chemical equipment is huge and the dimensionality is high, it is impossible to manually set a suitable number of clusters. Therefore, this embodiment first determines the optimal number of clusters K by using an improved K-means clustering algorithm, and then performs clustering based on the optimal number of clusters K.

[0106] The specific method for selecting the optimal number of clusters K is as follows:

[0107] ① Determine the set of possible values ​​for K and the maximum number of iterations t. max ;

[0108] ② For each value of K, iterate and update the centroid position of the cluster. The loss function that terminates the iteration is denoted as J. minn ;

[0109] ③ Compare the loss functions for all K values ​​in the same coordinate system;

[0110] ④ Observe the image. When the loss function value shows a clear inflection point, the corresponding K value is the optimal number of clusters.

[0111] The specific implementation steps of the K-means algorithm are as follows:

[0112] ① Determine the optimal number of clusters K using the methods described above;

[0113] ② Select K objects as initial cluster centers;

[0114] ③ Calculate the distance from each cluster object to the cluster center;

[0115] ④ Update the cluster centers until the cluster centers no longer change or the maximum number of iterations is reached, at which point the algorithm terminates;

[0116] ⑤ Save and label the data contained in each category.

[0117] In some implementations, the K-means algorithm defines the loss function as the sum of squared errors J between each sample and its cluster centroid. If the difference J between two consecutive iterations is less than a certain threshold, the iteration terminates, yielding the final clustering result:

[0118]

[0119] Where t is the number of iterations and ε is a set threshold. The iteration and termination methods are as follows:

[0120] Begin

[0121]

[0122] Calculate and update cluster centers

[0123] Calculate the distance from each cluster object to the cluster center.

[0124] The cluster centers no longer change or the maximum number of iterations is reached.

[0125] End

[0126] The model parameters contained in each cluster after K-means clustering are used to build the SVDD hypersphere model.

[0127] In some implementations, the basic idea of ​​SVDD in step 1.2.2) above is to construct a hyperboundary to form a hypersphere that contains as many positive samples as possible from the training samples, thereby maximizing the separation between positive and negative samples. The core problem of SVDD is to find the optimal boundary to achieve the best detection performance.

[0128] SVDD maps functions from the original space to the characteristic space through a nonlinear transformation function Φ:x→F, and then searches for a hypersphere with the smallest volume in the characteristic space.

[0129] Commonly used kernel functions include: linear kernel function, polynomial kernel function, Gaussian kernel function, and sigmoid kernel function.

[0130] like Figure 5 As shown, this invention uses the Gaussian kernel function to optimize SVDD, mapping the training data to a higher-dimensional space for hypersphere computation. The Gaussian kernel function is as follows:

[0131] K(x i ,x j )=exp(-(x i -x j ) 2 / s 2 )

[0132] In some implementations, the Lagrange coefficients are set to satisfy 0 < α in all normal training samples. i The samples <C are called support vectors, α i It is x i The corresponding Lagrange coefficients, where C is a constant. The center and radius R of the hypersphere are calculated, and the distance d from the test sample to the center of the hypersphere is also calculated.

[0133] If d≤kR (k≥1), it means that the test sample is on or inside the hypersphere and is a normal sample; otherwise, it is an abnormal sample.

[0134] In actual training, a small number of negative samples should be added to the positive sample training set to prevent overfitting. Assume the labels of the positive and negative samples in the training set are as follows:

[0135]

[0136] The calculation methods for the center and radius of the hypersphere, as well as the distance from the test sample to the center of the hypersphere, will change.

[0137] The formulas for calculating the center and radius of a hypersphere are:

[0138]

[0139] Test sample x t The distance to the center of the hypersphere is:

[0140]

[0141] In some implementations, after the hypersphere is trained, its center, radius, and other features are obtained. At this point, the median method is used, which involves substituting all the preprocessed data back into the hypersphere, calculating the distance between each data point and the center, sorting them, and extracting the process parameters at the timestamp of the median as typical process points representing the characteristics of the hypersphere.

[0142] Preferably, step 1.2.3 above includes the following steps:

[0143] 1.2.3.1) Draw a visualization of the hypersphere distribution for each data segment;

[0144] 1.2.3.2) Determine whether the hypersphere visualization graph contains hyperspheres that need to be discarded. There are two possible reasons for discarding: ① The amount of data inside the hypersphere is too small; ② The hypersphere visualization distribution is abnormal, that is, the distribution is relatively scattered or most parameters are 0.

[0145] 1.2.3.3) Record the retained hypersphere.

[0146] Preferably, in step 1.2.4) above, as follows: Figure 6 As shown, it includes the following steps:

[0147] 1.2.4.1) Calculate the median of the training data inside the hypersphere;

[0148] 1.2.4.2) Substitute the test set data into the hypersphere and calculate the distance between the test data and the median of the hypersphere.

[0149] 1.2.4.3) Determine the data closest to the median;

[0150] 1.2.4.4) Extract typical process parameter points, output quality indicators, corresponding operating condition data, and corresponding ideal process tag points.

[0151] Preferably, in step 1.4) above, artificial mechanism fusion and screening are performed, which involves matching the extracted typical process points with timestamps, operating conditions, and quality data, and selecting the process parameter points with better quality and storing them in the excellent parameter recommendation table. The artificial mechanism fusion mainly includes two steps to screen typical process points:

[0152] ① Screening for anomalies in product quality indicators and tag data: If the product quality fluctuates significantly around the timestamp of a process point, the typical process point will be discarded; if the tag data of the process point is abnormal, i.e., 0 or negative, the typical process point will also be discarded.

[0153] ② Comparison of similarity between typical process points: If the tag data of typical process points are compared pairwise, if the similarity is large, then the quality indicators are further observed. If there are differences in the quality indicators, then the tag data near the timestamp of the two process points are observed to see if there are large fluctuations. If the tag data of one process point fluctuates greatly, then the process point with large fluctuations is discarded. If the two tag data fluctuate greatly or not much, then both are discarded.

[0154] Finally, the retained data is used to remove typical process points that result in poor product quality, forming the final recommended table of excellent parameters.

[0155] Preferably, such as Figure 7 As shown, step 2) above includes the following steps:

[0156] 2.1) Obtain the current DCS process tag number data through the interface and identify the current operating condition;

[0157] 2.2) Obtain the optimal process parameter points that can be selected under the current operating conditions from the optimal parameter recommendation table;

[0158] 2.3) Calculate the distance between the currently running parameters and all the selectable excellent process parameter points under its operating conditions, and extract the excellent process parameter point with the smallest distance for output hot backup.

[0159] Preferably, such as Figure 8 As shown, step 3) above includes the following steps:

[0160] 3.1) Obtain the DCS tag number dataset within a preset time period (e.g., the most recent month) and preprocess it to obtain the latest dataset;

[0161] 3.2) Check whether the latest dataset falls into an existing hypersphere. If it does, proceed to step 3.3); otherwise, proceed to step 3.4.

[0162] 3.3) Determine which existing hypersphere the latest dataset falls into, and determine whether the data volume of the hypersphere is less than the set threshold. If so, add the latest dataset into the hypersphere and retrain the hypersphere to obtain new hypersphere features and extract new typical process points; otherwise, do not update.

[0163] 3.4) Record the latest dataset and form a new input data matrix. Use the same method as in step 1) to train the hypersphere cluster and store the extracted high-quality process parameter points in the excellent parameter recommendation table.

[0164] This invention improves product quality by optimizing key process parameters, provides a basis for parameter selection for different production plans and operating conditions, and makes rational use of massive and difficult-to-analyze data. This ensures the effective expansion and updating of the hypersphere cluster, maintains its comprehensiveness and integration, and guarantees that the hypersphere cluster can cover various operating conditions and optimization points, ultimately ensuring the long-term effectiveness of the method. It has advantages such as strong practicality and wide applicability, and possesses excellent application prospects and commercial value.

[0165] Example 2

[0166] As the most complex production device in chemical processes, the reactor is used as an example in this invention to illustrate the implementation steps of the method. During production operation, product quality is affected by multiple loops and various materials. In actual production, adjustments to production plans and changes in operating conditions make it impractical to manually set parameter values ​​for various process variables under different production plans and operating conditions, making it difficult to guarantee product quality. Simultaneously, regular sampling and testing of the reactor solution are necessary on-site, and data collection of key process variable parameters in critical loops is required to provide data support for big data-driven process optimization. This allows refining and chemical enterprises to better ensure product quality through a data-driven approach.

[0167] In a certain refining reactor, the effluent from an ammonia oxime reaction reactor is tested for indicators such as the purity, selectivity, and conversion rate of the reactants as the basis for judging product quality. The concentrations of hydrogen peroxide and catalyst are used as the basis for judging process load and operating conditions. The reactor contains 12 key process variables, including output measurements from loops such as flow rate, liquid level, and temperature. For offline training, data were collected for these 12 key variables from March to May, with a sampling frequency of 60 seconds.

[0168] After data sampling is completed, data preprocessing is required. The effective sample length of the 12 key variables is 123769. The parameters of the 12 key variables at the same time are combined into a data vector to form the initial data matrix for big data process optimization.

[0169] The offline training part of big data process optimization is mainly divided into two steps: algorithm training and mechanism fusion and screening.

[0170] First, the data is divided into reasonable segments. After determining the number of segments, K-means clustering becomes the first stage of algorithm training. The K-means clustering algorithm needs to select the optimal number of clusters to achieve the best clustering effect, ensuring the algorithm's effectiveness and accuracy. If the number of data samples is small, the number of clusters can be selected manually based on experience; if the number of data samples is huge, the optimal clustering value needs to be selected using the "inflection point method": that is, calculating the sum of squared errors of each class under different numbers of clusters. As the number of clusters increases, the number of points in each class decreases, and the sum of squared errors decreases accordingly. Observe the slope of the curve; the number of clusters corresponding to the obvious inflection point is the optimal clustering value. Taking the 12-dimensional data matrix formed in the above example as an example, the data is divided into 5 segments. Taking the second segment as an example, the optimal image is selected as follows: Figure 9 As shown in the figure, the vertical axis represents the sum of squared errors within a cluster, and the horizontal axis represents the number of clusters.

[0171] As shown in the graph, a clear inflection point occurs when the number of clusters is between 4 and 5, indicating that the optimal number of clusters for the control loop is 5. With the optimal number of clusters K=5 manually input, the first stage of algorithm training is complete. The second stage of algorithm training then begins—training the SVDD hypersphere model.

[0172] After clustering the data using the K-means clustering algorithm, the model parameters of different categories are labeled and sorted according to the number of model parameters in each category from largest to smallest to establish SVDD hypersphere models. Since all data in the above data matrix are clustered into 5 categories, 5 SVDD hypersphere models need to be established. First, the SVDD hypersphere model parameters are determined, including: selecting a Gaussian kernel function as the SVDD kernel function; a hypersphere kernel width of 12; and a hypersphere threshold of 1.5. Second, during the SVDD hypersphere modeling process, a small number of negative class samples need to be added to the positive class samples to prevent overfitting. Therefore, two samples are taken from each hypersphere except the Kth hypersphere as negative class samples, so that the Kth hypersphere integrates a combined training set consisting of positive and negative class samples.

[0173] After completing the hypersphere training, it needs to be visualized and analyzed. The visualization distribution diagram of the hypersphere in segment 5_2 is shown below. Figure 10 As shown, the hyperspheres in this segment are distributed compactly and reasonably, and all of them can be preserved.

[0174] Furthermore, using a visualization of some training results as an example, we demonstrate which visualizations need to be discarded. For instance... Figure 11As shown, taking the third hypersphere in segment 3_2 as an example, its training exhibits two distributions, which are quite scattered, so it is discarded. Hyperspheres 5 and 6 clearly contain fewer data samples, so they are also discarded, while the other hyperspheres can be retained. Figure 12 As shown, the visualization of the second hypersphere in paragraph 3_3 is obviously abnormal, containing many cases where the bit data is 0 and is quite scattered. Therefore, this hypersphere is discarded, while the other hyperspheres can be retained.

[0175] At this point, taking the data in segment 5_2 as an example, based on the distance between the center of each retained SVDD hypersphere and the test data, the test set is re-substituted into the hypersphere for analysis—the median of the test data vector contained in each hypersphere is solved, and the test data vector closest to the obtained median is found by traversal. The time, location, and specific value of the most representative data vector can be extracted. The results are shown in Table 1.

[0176] Table 1, Section 5_2: Example of Algorithm Training Results

[0177]

[0178] At this point, the algorithm training is complete, and the next step is to proceed to the mechanism fusion and selection section.

[0179] It is important to note that in actual refining and chemical plants, the effects of adjusting key process variables are not immediately reflected in the output quality indicators; there is a certain time delay. This delay varies depending on the reactor, with most reactors experiencing a delay of approximately 4-48 hours. Therefore, during the artificial mechanism fusion and screening process, it is necessary to average the test results of quality indicators and operating condition indicators within the delay period following the timestamp of the process points extracted during algorithm training. This average is then used to observe whether the conditions for excellent parameter points are met to determine whether to store them in the recommendation table. In this example, the actual delay may reach 24-48 hours; therefore, the average of the test results of quality indicators and operating condition indicators within 48 hours after the timestamp is selected to determine its effectiveness.

[0180] Specifically, the product quality indicators—color, selectivity, and conversion rate—as well as the operating condition indicators—hydrogen peroxide and catalyst concentration—of the typical data sampling points obtained during the algorithm training process of the hypersphere retained in segment 5_2 were extracted separately, as shown in Table 2.

[0181] Table 2, Section 5_2, shows the corresponding product quality indicators and operating conditions.

[0182]

[0183] At this point, we can proceed with the analysis of mechanism fusion and screening. First, we observe the quality indicators near all timestamps. Among them, the quality indicators near the typical process points extracted from the No. 4 hypersphere fluctuate greatly, so the quality indicators of the extracted parameters are difficult to guarantee, and therefore they are directly discarded. Next, we check whether the remaining 4 typical process points have zero tag data. In the example of this data segment, this situation did not occur, so we continue to retain the above 4 parameter points.

[0184] Next, a similarity comparison was performed. It was found that the reference data and quality indicators of hyperspheres 2 and 3 were similar, so both were retained. Finally, their quality indicators were observed, and parameter points with poor quality indicators were eliminated. In this example, the quality indicator corresponding to the typical process point extracted from hypersphere 5 was 181.18, which was considered poor (lower chroma and selectivity and conversion rate greater than 99.5 indicate better product quality), so this typical process point was discarded. The final retained typical process points, numbered 1, 2, and 3, were stored in the recommendation table as excellent parameter points. The table shows that parameter point 1 has a quality indicator of 134.86, while points 2 and 3 are both around 150, indicating excellent quality. Furthermore, the fluctuations in their quality indicators and reference data around their respective timestamps are small, ensuring the reliability of the extracted excellent process parameters and further demonstrating the rationality and effectiveness of the method.

[0185] By storing the data of each hypersphere and its internal structure from the offline training section, along with recommended data for different operating conditions, in a database, online updates and optimizations can be performed. This database is then connected to the enterprise's real-time database to identify the current operating condition, extract process reference number data, obtain the optimal process point parameters stored for the current operating condition from the recommendation table, calculate the distance between the current reference number data and each optimal process parameter, and select the optimal parameter point with the smallest distance for output hot backup.

[0186] In addition, during long-term use, the hypersphere data features and recommendation table are updated monthly through the online learning function. The DCS position dataset from the past month is selected, preprocessed, and then sequentially determined which hypersphere the data falls into. Data falling into a hypersphere is stored and updated. The existing hyperspheres are retrained to update the hypersphere features and extract the latest typical process points. Then, through manual mechanism fusion and screening, superior process parameter points are extracted to update and expand the recommendation table. If most of the data from the past month does not fall into an existing hypersphere, the data that falls into a hypersphere is recorded and a new data matrix is ​​formed. This matrix is ​​trained using the same method as offline training: the data matrix is ​​re-clustered to obtain the number of clusters, and then SVDD features are extracted for each category. Recent data not falling into existing hyperspheres are substituted into the median method to extract representative process parameter points. Then, through manual mechanism fusion and screening, timestamps, corresponding operating conditions, and quality data are matched to select superior process parameter points for storage in the recommendation table.

[0187] The big data-based refining and chemical plant process optimization methods and systems have unique advantages in handling multivariate and massive datasets. Although the factors affecting product quality in chemical processes are diverse and cannot be directly analyzed manually, relying instead on experience for adjustments, the K-means clustering method can effectively provide a partitioning scheme through unsupervised iteration. Furthermore, based on the current classification results, the SVDD hypersphere training method is used to include as many positive samples as possible from the training set, maximizing the separation between positive and negative samples and seeking the optimal boundary for the most accurate data partitioning. Notably, during the actual training process, a small number of negative samples are added to the positive sample training set to prevent overfitting, further ensuring the accuracy of data analysis. Simultaneously, visualization analysis is incorporated to more clearly display the distribution of the SVDD hyperspheres, providing a reliable basis for further data mining and hypersphere selection.

[0188] Once the algorithm training process is complete, the manual mechanism fusion and screening process can begin, ensuring the quality of selected typical process parameter points. By rigorously screening the quality indicators and tag data of typical process points from three aspects—data anomalies, data fluctuations, and similarity comparisons—the optimal parameter points are ultimately obtained. This further enhances the effectiveness and rationality of the overall technical solution, laying a solid foundation for guiding actual production.

[0189] In addition, an online update and optimization component has been added, which not only provides real-time optimization solutions but also periodically updates and expands the data to maintain the comprehensiveness and versatility of the hypersphere cluster. This ensures that the hypersphere cluster can cover various working conditions and various optimization points, guaranteeing the long-term effectiveness of the method.

[0190] In summary, given the current situation where most industrial data is piling up, has low utilization rates, and is difficult to use, big data process optimization algorithms can perfectly utilize this data, providing higher quality products for the process and improving economic efficiency.

[0191] Compared with the prior art, the present invention has the following advantages:

[0192] Traditional process optimization involves establishing mechanistic models, deriving mathematical models of the process, and finding optimal operating conditions. While this method offers high accuracy, it is extremely difficult to implement. Real-world processes are complex, with numerous variable factors, making it impossible to provide precise mathematical models, and it can only address relatively limited problems. Furthermore, most common data-driven optimization methods involve one-time calculations, making it difficult to enable enterprises to perform long-term, autonomous optimization in situations where production scheduling plans frequently change. In contrast, the advantages of this invention are summarized as follows:

[0193] (1) This invention does not require an accurate mathematical model and does not need to consider the problem of model mismatch;

[0194] (2) The online update and optimization part of the present invention can ensure coverage of various optimization points under multiple working conditions, and can effectively update and expand, maintain comprehensiveness and integration, and enable enterprises to carry out long-term independent optimization and effectiveness.

[0195] (3) The present invention has strong adaptability in various environments and processes, and has the ability to continuously improve and deeply train and optimize.

[0196] (4) This invention effectively utilizes massive historical time series data to conduct in-depth analysis and mining, which can quickly obtain valuable information and form a process parameter optimization scheme that can be promoted.

[0197] (5) This invention combines visualization analysis methods to present data characteristics more intuitively, providing a reliable basis for further data mining and data screening.

[0198] Example 3

[0199] The above-described embodiment 1 provides a method for optimizing refining and chemical processing units based on big data. Correspondingly, this embodiment provides a system for optimizing refining and chemical processing units based on big data. The system provided in this embodiment can implement the process optimization method for refining and chemical processing units based on big data of embodiment 1. This system can be implemented through software, hardware, or a combination of both. For example, the system may include integrated or separate functional modules or units to execute the corresponding steps in the methods of embodiment 1. Since the system in this embodiment is basically similar to the method embodiment, the description process in this embodiment is relatively simple. Relevant details can be found in the description of parts of embodiment 1. The system embodiment provided in this embodiment is merely illustrative.

[0200] The big data-based refining and chemical plant process optimization system provided in this embodiment includes:

[0201] The data acquisition module is used to acquire the current operating conditions and process tag data of the refining and chemical unit;

[0202] The parameter optimization module is used to select appropriate excellent parameter points from a pre-established table of recommended excellent parameters for different operating conditions based on the current operating conditions and process tag number data, and to output them to optimize the process parameters of the refining and chemical unit under the current operating conditions.

[0203] In some implementations, the system also includes an online update module, which is used to acquire a dataset of operating conditions and process reference numbers within a preset time period, expand and update the hypersphere cluster based on the acquired dataset, and obtain an updated recommended table of excellent parameters for different operating conditions.

[0204] Example 4

[0205] This embodiment provides a processing device corresponding to the big data-based refining and chemical plant process optimization method provided in Embodiment 1. The processing device can be a client-side processing device, such as a mobile phone, laptop, tablet, desktop computer, etc., to execute the method of Embodiment 1.

[0206] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it executes the big data-based refining and chemical plant process optimization method provided in Embodiment 1.

[0207] In some embodiments, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0208] In other embodiments, the processor can be a general-purpose processor of various types, such as a central processing unit (CPU) or a digital signal processor (DSP), and is not limited thereto.

[0209] Example 5

[0210] The big data-based refining and chemical plant process optimization method of this embodiment 1 can be specifically implemented as a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the big data-based refining and chemical plant process optimization method of this embodiment 1 are loaded.

[0211] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0212] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for optimizing refining and chemical processing equipment based on big data, characterized in that... Includes the following steps: Obtain the current operating status and process tag data of the refining and chemical unit from the upper-level cloud platform or DCS; Based on the current operating conditions and process tag number data, select appropriate excellent parameter points from the pre-established excellent parameter recommendation table for different operating conditions and output them to optimize the process parameters of the refining and chemical unit under the current operating conditions. The recommended optimal parameters for different operating conditions are obtained by training K-means and SVDD hyperspheres using historical data from the enterprise's refining and chemical units, and include: 1.1) Collect historical data from the enterprise's refining and chemical units, preprocess the data, and segment the data to form an input data matrix; 1.2) For each data segment, the input data matrix is ​​classified using the improved K-means clustering method, and then the SVDD method is used for hypersphere training. Abnormal hyperspheres are discarded by combining visualization analysis to obtain different hyperspheres and typical process parameter points with timestamps. 1.3) Determine whether the number of typical process parameters selected by segmentation meets the preset conditions. If it does, it means that the number of segments is reasonable. After training all segments, proceed to step 1.4). Otherwise, reselect the number of segments and return to step 1.1) until the preset conditions are met. 1.4) Based on the obtained typical process parameter points, artificial mechanism fusion and screening are carried out. Artificial mechanism fusion and screening refers to matching the extracted typical process points with timestamps, operating conditions and quality data, and selecting process parameter points whose quality meets the preset requirements and storing them in the excellent parameter recommendation table. 1.5) Output the required quality indicators, corresponding operating data, and corresponding excellent process reference points, and store them in the excellent parameter recommendation table; In step 1.2), for each data segment's input data matrix, after classification using an improved K-means clustering method, hypersphere training is performed using the SVDD method, and abnormal hyperspheres are discarded based on visualization analysis, resulting in different hyperspheres and typical process parameter points with timestamps, including: An improved K-means clustering method is used to select the optimal number of clusters and to classify and label the input data matrix. The SVDD method is used to train a hypersphere on each category of data to extract data features under different working conditions; Visualize and analyze the hyperspheres in each data segment, discarding hyperspheres with less than the preset value of internal data or those with abnormal visualization, and retaining the remaining normal hyperspheres. All the preprocessed data were resubmitted into the retained hypersphere, and the median method was used to extract the representative optimal parameters for different working conditions as typical process parameter points. The process parameters of the refining unit under the current operating conditions are optimized based on the current operating conditions and process tag number data, by selecting appropriate optimal parameter points from a pre-established optimal parameter recommendation table for different operating conditions and outputting them. This includes: The current DCS process tag number data is obtained through the interface, and the current operating condition is identified. The optimal process parameter points that can be selected under the current operating conditions are obtained from the recommended optimal parameter table. Calculate the distance between the currently running parameters and all selectable optimal process parameter points under its operating conditions, and extract the optimal process parameter point with the smallest distance for output.

2. The method for optimizing refining and chemical processing equipment based on big data as described in claim 1, characterized in that, In step 1.1), historical data from the enterprise's refining and chemical processing units is collected and preprocessed. The data is then segmented to form an input data matrix, including: Historical data were collected based on key process variables related to product quality, as determined by on-site experience. Select the time period during which all key process variables are effectively sampled; Determine the initial number of segments and segment the data accordingly; For each data segment, the output measurement values ​​of each key process variable at the same time are combined into a vector to form an input data matrix.

3. The method for optimizing refining and chemical processing equipment based on big data as described in claim 1, characterized in that, In step 1.4), based on the obtained typical process parameter points, artificial mechanism fusion and screening are performed, including: Product quality indicators and tag data anomaly screening: If the product quality fluctuates more than the preset value near the timestamp of the process point, the typical process point will be discarded; if the tag data of the process point is abnormal, that is, 0 or negative, the typical process point will also be discarded. Typical process point similarity comparison: The tag data of typical process points are compared pairwise. If the similarity is greater than the preset value, their quality indicators are compared. If there is a difference in the quality indicators, the fluctuation of the tag data near the timestamp of the two typical process points is observed to see if it is greater than the preset value. If the fluctuation of the tag data of one process point is greater than the preset value, the typical process point is discarded. If the fluctuation of the tag data of both process points is greater than the preset value or both are less than the other preset value, both are discarded.

4. The method for optimizing refining and chemical processing equipment based on big data as described in claim 1, characterized in that, This method includes: acquiring a dataset of operating conditions and process reference numbers within a preset time period; expanding and updating the hypersphere cluster based on the acquired dataset; and obtaining an updated recommended table of optimal parameters for different operating conditions; including: 3.1) Obtain the DCS tag number dataset within a preset time period and preprocess it to obtain the latest dataset; 3.2) Check whether the latest dataset falls into an existing hypersphere. If so, proceed to step 3.3; otherwise, proceed to step 3.

4. 3.3) Determine which existing hypersphere the latest dataset falls into, and determine whether the data volume of the hypersphere is less than the set threshold. If so, add the latest dataset into the hypersphere and retrain the hypersphere to obtain new hypersphere features and extract new typical process points; otherwise, do not update. 3.4) Record the latest dataset and form a new input data matrix, perform hypersphere cluster training, and store the extracted process parameter points in the excellent parameter recommendation table.

5. A process optimization system for refining and chemical plants based on big data, characterized in that, include: The data acquisition module is used to acquire the current operating status and process tag data of the refining and chemical unit from the upper-level cloud platform or DCS; The parameter optimization module is used to select appropriate excellent parameter points from a pre-established table of recommended excellent parameters for different operating conditions based on the current operating conditions and process tag number data, and to output them to optimize the process parameters of the refining and chemical unit under the current operating conditions. The recommended optimal parameters for different operating conditions are obtained by training K-means and SVDD hyperspheres using historical data from the enterprise's refining and chemical units, and include: 1.1) Collect historical data from the enterprise's refining and chemical units, preprocess the data, and segment the data to form an input data matrix; 1.2) For each data segment, the input data matrix is ​​classified using the improved K-means clustering method, and then the SVDD method is used for hypersphere training. Abnormal hyperspheres are discarded by combining visualization analysis to obtain different hyperspheres and typical process parameter points with timestamps. 1.3) Determine whether the number of typical process parameters selected by segmentation meets the preset conditions. If it does, it means that the number of segments is reasonable. After training all segments, proceed to step 1.4). Otherwise, reselect the number of segments and return to step 1.1) until the preset conditions are met. 1.4) Based on the obtained typical process parameter points, artificial mechanism fusion and screening are carried out. Artificial mechanism fusion and screening refers to matching the extracted typical process points with timestamps, operating conditions and quality data, and selecting process parameter points whose quality meets the preset requirements and storing them in the excellent parameter recommendation table. 1.5) Output the required quality indicators, corresponding operating data, and corresponding excellent process reference points, and store them in the excellent parameter recommendation table; In step 1.2), for each data segment's input data matrix, after classification using an improved K-means clustering method, hypersphere training is performed using the SVDD method, and abnormal hyperspheres are discarded based on visualization analysis, resulting in different hyperspheres and typical process parameter points with timestamps, including: An improved K-means clustering method is used to select the optimal number of clusters and to classify and label the input data matrix. The SVDD method is used to train a hypersphere on each category of data to extract data features under different working conditions; Visualize and analyze the hyperspheres in each data segment, discarding hyperspheres with less than the preset value of internal data or those with abnormal visualization, and retaining the remaining normal hyperspheres. All the preprocessed data were resubmitted into the retained hypersphere, and the median method was used to extract the representative optimal parameters for different working conditions as typical process parameter points. The process parameters of the refining unit under the current operating conditions are optimized based on the current operating conditions and process tag number data, by selecting appropriate optimal parameter points from a pre-established optimal parameter recommendation table for different operating conditions and outputting them. This includes: The current DCS process tag number data is obtained through the interface, and the current operating condition is identified. The optimal process parameter points that can be selected under the current operating conditions are obtained from the recommended optimal parameter table. Calculate the distance between the currently running parameters and all selectable optimal process parameter points under its operating conditions, and extract the optimal process parameter point with the smallest distance for output.

6. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 4.

7. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 4.

Images