Gasoline dry point online quality prediction method, system and equipment based on big data and medium

Abstract

The invention relates to a gasoline dry point online quality prediction method, system, equipment and medium based on big data, and the method comprises the following steps: determining a key process bit number influencing a gasoline dry point quality index through the analysis of a process flow and a process mechanism, and carrying out the extraction and preprocessing of the long-term historical data of the key process bit number; training a pre-constructed BP neural network; after key process bit number data corresponding to the current prediction moment are extracted and preprocessed based on a set prediction target, the key process bit number data are input into a BP neural network to generate an instantaneous prediction value, and a smooth online quality index prediction value is generated through sliding weighted mean filtering processing. The device can be widely applied to the field of petroleum and petrochemical refining devices.

Description

Technical Field

[0001] This invention belongs to the field of petroleum and petrochemical refining equipment, and in particular relates to a method, system, equipment and medium for online quality prediction of gasoline dry point based on big data. Background Technology

[0002] Gasoline dry point is a key quality indicator for catalytic cracking units in the oil refining process, directly affecting the quality and yield of gasoline products. Infrequent testing of gasoline dry point (only twice a day) prevents production personnel from promptly identifying changes in product quality and making adjustments, leading to problems such as reduced product quality. Therefore, accurate prediction of quality indicators is crucial for reducing product defect rates, improving production efficiency, lowering energy consumption and costs, and ensuring production sustainability. Furthermore, quality indicators reflect equipment operating status and changes in operating conditions, providing data support for optimizing production parameters and improving resource utilization.

[0003] With industrial development, industrial process quality prediction has become a crucial means to improve product quality, optimize production efficiency, and reduce costs. In refining and chemical processes, the changing trends and values ​​of key quality indicators play a decisive role in process operation. Currently, prediction methods for industrial process quality are mainly divided into two categories: model-driven and data-driven. Model-driven methods rely on complex mathematical or statistical models, making it difficult to capture core quality indicators, thus resulting in low prediction efficiency in complex production processes. Data-driven methods, on the other hand, utilize actual production data and employ data feature extraction techniques and machine learning for self-learning and self-adaptation. This allows for more effective mining of key features from historical data, adapting to the prediction of quality indicators in time-series production processes, and has become a current research hotspot.

[0004] However, due to the lack of online analyzers in the field equipment, it is difficult to obtain quality index data in real time, making it difficult to adjust the operating conditions in a timely manner. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a method, system, device, and medium for online quality prediction of gasoline dry point based on big data. By combining feature extraction optimization, BP neural network, and sliding weighted mean filtering techniques, efficient and accurate prediction of gasoline dry point is achieved.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for online quality prediction of gasoline dry point based on big data, comprising the following steps:

[0008] By analyzing the process flow and mechanism, the key process reference numbers affecting the dry point quality index of gasoline were determined. After extracting and preprocessing the long-term historical data of the key process reference numbers, a pre-constructed BP neural network was trained.

[0009] Based on the set prediction target, the key process reference data corresponding to the current prediction time is extracted and preprocessed, then input into a BP neural network to generate instantaneous prediction values, and then processed by sliding weighted average filtering to generate smooth online quality index prediction values.

[0010] Furthermore, the process of determining the key process reference numbers affecting the gasoline dry point quality index through process flow and mechanism analysis, and then training a pre-constructed BP neural network after extracting and preprocessing the long-term historical data of the key process reference numbers, includes:

[0011] The process flow and mechanism were analyzed to identify the key process variables affecting the dry point quality index of gasoline.

[0012] Collect long-term historical time-series data of key process variables and extract typical process points through data mining. The extracted typical process points include process tag number data and the timestamps corresponding to the process tag number data.

[0013] Within 24 hours after the typical process point timestamp, find the test timestamp that is closest to the typical process point timestamp and its corresponding gasoline dry point test value.

[0014] The reverse timestamp is determined based on the test timestamp, and the reverse timestamp and the gasoline dry point test values ​​and process tag data within the preset time period before and after are fused according to the preset fusion rules.

[0015] The data features are obtained by dimensionality reduction of the process tag data within the back-dated timestamp and the preset time period before and after by SVD, and then a prediction summary table is formed together with the median of the process tag data within the back-dated timestamp and the preset time period before and after.

[0016] The pre-built BP neural network is trained based on the prediction master table to obtain the BP neural network prediction model. Further, the data features are obtained by dimensionality reduction of the back-derived timestamps and process tag number data within the preset time periods before and after the timestamps using SVD, and these features are combined with the median of the process tag number data within the preset time periods before and after the timestamps to form the prediction master table, including:

[0017] Standardize and transform the process tag data within the backward timestamp and the preset time periods before and after it;

[0018] Singular value decomposition was performed on the standardized and square-transformed process reference data using SVD decomposition to extract data features.

[0019] The data features, together with the median of the process tag data within the preset time periods before and after the timestamps, are used to form a prediction summary table.

[0020] Furthermore, the step of training a pre-built BP neural network based on the prediction summary table to obtain a BP neural network prediction model includes:

[0021] Construct a BP neural network;

[0022] The BP neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to read the data features and median corresponding to the process reference number and perform normalization processing. The hidden layer contains an activation function, which is used to perform nonlinear transformation on the input features and extract deep feature information. The output layer is used to output the predicted value.

[0023] Determine the loss function and training parameters, and train the constructed BP neural network using the prediction summary table.

[0024] Furthermore, after extracting and preprocessing the key process reference data corresponding to the current prediction time based on the set prediction target, the data is input into a BP neural network to generate instantaneous prediction values. These prediction values ​​are then processed through a sliding weighted average filter to generate smooth online quality index prediction values, including:

[0025] Set forecasting targets based on actual needs;

[0026] Based on the preset target, determine the reverse timestamp;

[0027] Extract and perform tag number operations on the reverse timestamp and the process tag number data within the preset time period before and after it to form a data matrix, and calculate the median of the process tag number data;

[0028] The data features are obtained by extracting features from the data matrix using SVD, and then the data features are concatenated with the median of the process reference number to form a complete input vector.

[0029] The input vector is fed into the BP neural network for prediction to obtain the instantaneous predicted value of the quality index;

[0030] The instantaneous predicted value is filtered by a moving weighted average to obtain the smooth online quality index prediction value at the current prediction time.

[0031] Prepare data and write it back to the real-time database.

[0032] Furthermore, before inputting the input vector into the BP neural network for prediction to obtain the instantaneous predicted value of the quality indicator, the following steps are also included:

[0033] Determine if all data features extracted by SVD are 0. If so, take the previous prediction value as the current instantaneous prediction value; otherwise, perform BP neural network prediction.

[0034] Furthermore, the sliding weighted average filtering of the instantaneous predicted value refers to performing a sliding weighted average filtering on the instantaneous predicted values ​​corresponding to several prediction times before the current prediction time, and using the filtering result as the smooth online quality index prediction value for the current prediction time.

[0035] Secondly, the present invention provides a gasoline dry point online quality prediction system based on big data, comprising:

[0036] The offline training module is used to determine the key process reference numbers that affect the dry point quality index of gasoline through process flow and process mechanism analysis, and to train a pre-constructed BP neural network after extracting and preprocessing the long-term historical data of the key process reference numbers.

[0037] The online prediction module is used to extract and preprocess the key process reference data corresponding to the current prediction time based on the set prediction target, input it into the BP neural network to generate instantaneous prediction values, and then generate smooth online quality index prediction values ​​through sliding weighted average filtering.

[0038] 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 method.

[0039] Fourthly, the present invention provides a computing device comprising: one or more processors and a memory, wherein the memory stores one or more programs and is configured to be executed by the one or more processors, the one or more programs including instructions for performing any method.

[0040] The present invention has the following advantages due to the adoption of the above technical solutions:

[0041] 1. This invention utilizes singular value decomposition (SVD) technology to extract data features. It can not only capture data features more comprehensively, but also effectively avoid deviations caused by fluctuations or pulsations at a certain timestamp process point, thereby further improving the reliability of prediction results.

[0042] 2. This invention utilizes a BP shallow neural network for nonlinear regression, which can efficiently handle nonlinear relationships, adaptively learn features, and has good adaptability.

[0043] 3. This invention effectively suppresses periodic interference, reduces random fluctuations in data, and significantly improves generalization and stability by calculating the weighted average of the predicted instantaneous values ​​of continuous data points.

[0044] Therefore, this invention can be widely applied in the field of quality inspection. Attached Figure Description

[0045] 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:

[0046] Figure 1 This is a schematic diagram of the online quality prediction method for gasoline dry point based on big data provided in an embodiment of the present invention;

[0047] Figure 2 This is a schematic diagram illustrating the predicted effect from March 14th to March 19th, 2025, according to an embodiment of the present invention.

[0048] Figure 3 This is a schematic diagram illustrating the predicted effect from March 9th to March 14th, 2025, according to an embodiment of the present invention.

[0049] Figure 4 This is a schematic diagram of the prediction effect from March 6th to March 11th, 2025, provided by an embodiment of the present invention. Detailed Implementation

[0050] 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.

[0051] 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 scope of exemplary embodiments according to the invention. 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.

[0052] In some embodiments of the present invention, a method for online quality prediction of gasoline dry point based on big data is provided, comprising: analyzing process reference numbers that affect quality indicators through process mechanism analysis; extracting typical process points from long-term historical reference number data and fusing them with quality indicators; extracting data features near historical quality indicators using Singular Value Decomposition (SVD) and fusing them with process points to obtain a prediction summary table; training a BP neural network prediction model; obtaining data features near the current time through SVD online and combining them with the corresponding process points as input to the BP neural network prediction model for quality prediction; calculating a sliding weighted average value through a sliding window as the current prediction value output and writing it back to the real-time database. This invention can be widely applied in refining and chemical plants in the petroleum and petrochemical industry.

[0053] Correspondingly, in other embodiments of the present invention, a gasoline dry point online quality prediction system, device, and medium based on big data are provided.

[0054] Example 1

[0055] like Figure 1 As shown, this invention provides an online quality prediction method for gasoline dry point based on big data, which includes the following steps:

[0056] 1) The key process reference numbers affecting the dry point quality index of gasoline are determined by process flow and process mechanism analysis, and the long-term historical data of the key process reference numbers are extracted and preprocessed to train a pre-constructed BP neural network.

[0057] 2) Based on the set prediction target, the key process reference data corresponding to the current prediction time is extracted and preprocessed, then input into the BP neural network to generate instantaneous prediction values, and then processed by sliding weighted average filtering to generate smooth online quality prediction values.

[0058] Furthermore, in step 1) above, during continuous industrial production, changes in quality indicators are gradually reflected over time and often do not appear immediately; that is, the influence of process parameters on quality indicators has a certain time delay. Considering the time delay in how changes in process parameters are reflected in quality indicators, this invention establishes a method for integrating typical process points with quality indicators, specifically including the following steps:

[0059] 1.1) Analyze the process flow and mechanism to determine the key process variables that affect the dry point quality index of gasoline;

[0060] 1.2) Collect long-term historical time-series data of key process variables and extract typical process points through data mining. The extracted typical process points include process tag number data and the timestamps corresponding to the process tag number data.

[0061] 1.3) Within 24 hours after the typical process point timestamp, find the test timestamp that is closest to the typical process point timestamp and its corresponding gasoline dry point test value;

[0062] 1.4) Determine the reverse timestamp based on the test timestamp, and merge the reverse timestamp and the gasoline dry point test values ​​and process tag data within the preset time period before and after according to the preset fusion rules;

[0063] 1.5) The data features are obtained by dimensionality reduction of the back-dated timestamp and the process tag data within the preset time period before and after the time-stamp using SVD (Singular Value Decomposition), and are combined with the median of the back-dated timestamp and the process tag data within the preset time period before and after the time-stamp to form a prediction summary table.

[0064] 1.6) The pre-built BP neural network is trained based on the prediction summary table to obtain the BP neural network prediction model.

[0065] Furthermore, in step 1.4 above, after finding the gasoline dry point test value, it is necessary to integrate the gasoline dry point test value and its corresponding process tag data. This requires back-calculating based on the sampling situation and the influence time of key process variables on the gasoline dry point quality index to determine the back-calculated timestamp.

[0066] Specifically, the specific time delay is deduced by working backward from the laboratory timestamp, thereby identifying the reverse-dated timestamp corresponding to the process tag number affecting the quality indicators. This process also needs to consider the sampling time in actual on-site production. For example, in this embodiment, the workshop samples one hour earlier in the morning and half an hour earlier in the afternoon. Therefore, different fusion rules need to be set for the morning and afternoon to find the historical process tag number data for the corresponding time periods. Furthermore, if the process tag number data at the reverse-dated timestamp is directly selected, the data at that moment may be skewed. Therefore, it is also necessary to extract the process tag number data matrix near the reverse-dated timestamp. Based on field experience, process tag number data within 15 minutes before and after this time delay will affect the quality indicators. Therefore, after determining the specific time delay by working backward from the laboratory timestamp, it is also necessary to extract the process tag number data within 15 minutes before and after the reverse-dated timestamp.

[0067] Furthermore, step 1.5 above includes the following steps:

[0068] 1.5.1) Standardize and transform the process tag data within the back-derived timestamp and the preset time periods before and after it;

[0069] 1.5.2) Singular value decomposition is performed on the standardized and square-transformed process tag data through SVD decomposition to reduce data dimensionality and extract data features;

[0070] Singular Value Decomposition (SVD) is a commonly used matrix factorization technique, mainly used for data dimensionality reduction and feature extraction. Its core idea is to extract the most important signal components and remove noise interference by decomposing a standardized data matrix. Singular values ​​reflect the importance of data in different directions and can be used to construct low-dimensional feature representations, thereby improving modeling efficiency.

[0071] 1.5.3) Combine the data features with the median of the process tag data within the preset time period before and after the reverse timestamp to form a prediction summary table.

[0072] Furthermore, in step 1.6) above, when training the pre-built BP neural network based on the prediction summary table, the following steps are included:

[0073] 1.6.1) Construct a BP neural network;

[0074] In this embodiment, the constructed BP neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to read the data features and median corresponding to the process reference number and perform normalization processing. The hidden layer contains an activation function, which is used to perform nonlinear transformation on the input features to extract deep feature information. The output layer is used to output the predicted value.

[0075] 1.6.2) Determine the loss function and training parameters, and train the constructed BP neural network using the prediction summary table.

[0076] In this embodiment, the mean squared error (MSE) is used as the loss function when training the BP neural network, and the Adam optimizer is used for parameter updates. After a preset number of training rounds, the BP neural network model is used to make predictions on the validation set and compared with the true values. Finally, the trained BP neural network model and normalized parameters are saved for subsequent use.

[0077] Furthermore, step 2) above includes the following steps:

[0078] 2.1) Set forecasting targets based on actual needs.

[0079] To achieve high-frequency, real-time quality monitoring, this embodiment sets up a quality prediction to be performed every minute, generating a quality prediction value.

[0080] 2.2) Based on the preset target, determine the historical data window used for prediction, that is, determine the reverse timestamp based on the current prediction time.

[0081] Since changes in tag number data do not immediately reflect quality indicators due to a time delay, it is necessary to find the process tag number data corresponding to the current prediction time. For gasoline dry point, the time delay is 1.5 hours. Therefore, this embodiment needs to calculate the time delay backward from the current prediction time to determine the historical data window used for prediction.

[0082] 2.3) Extract and perform tag number operations on the reverse timestamp and the process tag number data within the preset time period before and after it to form a data matrix, and calculate the median of the process tag number data.

[0083] After determining the reverse timestamp, due to potential fluctuations in the production environment, the process variable data at the reverse timestamp may also fluctuate. Therefore, it is necessary to extract the tag number data for the 15 minutes before and after the reverse timestamp. Tag number calculations are then performed based on the obtained process tag number data to form a data matrix, and the median of the tag number data is taken.

[0084] 2.4) The data features are obtained by extracting features from the data matrix through SVD, and the data features are concatenated with the median of the process reference number to form a complete input vector.

[0085] 2.5) Input the input vector into the BP neural network for prediction to obtain the instantaneous predicted value of the quality index.

[0086] The accuracy of test data is affected by the testing methods and techniques used. Nonlinear regression modeling is required. In this embodiment, the input vector is input into a BP neural network to generate a predicted value, which serves as the instantaneous predicted value of the quality indicator, so that the prediction result can reflect the current production status in real time.

[0087] 2.6) Apply a sliding weighted average filter to the instantaneous predicted value to obtain the smoothed online quality index predicted value at the current prediction time.

[0088] Because the current predicted value has a continuous dependency on the predicted values ​​from previous times, a moving weighted average filter is required. The instantaneous predicted values ​​from the previous few minutes are weighted together with the current instantaneous predicted value to generate a smoothed quality index predicted value for the current prediction time.

[0089] 2.7) Prepare data and write back to the real-time database.

[0090] The system stores the smoothed quality index prediction value and instantaneous prediction value at the current prediction time for use in the next prediction; the prediction results are written back to the real-time database for displaying historical prediction curves on the front end, and enables the production system to obtain the latest quality prediction data at any time and optimize process parameters accordingly.

[0091] Furthermore, prior to step 2.5 above, the following steps are also included:

[0092] Determine if all data features extracted by SVD are 0. If so, take the previous prediction value as the current instantaneous prediction value; otherwise, proceed to step 2.5) to perform BP neural network prediction.

[0093] Further, in step 2.6) above, the sliding weighted mean filtering process includes:

[0094] Data is retrieved through an online interface, extracting the process tag data features and median of the process tag data corresponding to the current quality indicator. These are then used as input to a BP neural network model to obtain the predicted value. Since there is a continuous dependency between the predicted values ​​at the current time and those at previous times, the model output value cannot be used as the current predicted value and must be filtered.

[0095] This invention calculates a moving weighted average by combining the output values ​​of several previous BP neural network models, and outputs the filtered value as the current prediction value, which can reduce random fluctuations in data and suppress short-term noise interference.

[0096] This invention improves the effectiveness of model input by integrating quality indicators with corresponding process reference data through the time delay between quality indicators and process parameters. It optimizes the data structure using feature extraction methods, enhances the nonlinear mapping capability in the production process through BP neural network modeling, and improves the robustness of prediction results by smoothing short-term fluctuations through moving weighted mean filtering. The system can write back prediction results in real time and continuously perform data calculations for the next time step, ensuring the efficiency and continuity of quality prediction and providing accurate and reliable decision support for quality control in intelligent manufacturing environments.

[0097] Example 2

[0098] This embodiment further introduces the online quality prediction method for gasoline dry point based on big data proposed in this invention by performing online prediction of the gasoline dry point of a catalytic cracking unit in a domestic refinery. The method of this invention includes two stages: an offline training stage and an online prediction stage, which are described below.

[0099] The offline training phase includes the following steps:

[0100] 1) Process Mechanism Analysis

[0101] By analyzing the process flow and mechanism of the catalytic cracking split unit, the key process reference numbers affecting the quality indicators can be identified as: 8TIC202_PV, 8PI204_PV, 8FIC206_MV, 8FIQ522-8FIC246_PV, 8FIC211_PV, 8FIC211+8FIC213_PV, and 8FIC207_PV.

[0102] 2) Extract typical process points

[0103] Specifically, including:

[0104] First, historical data from the past two years were collected for the identified key process reference numbers, with a sampling frequency of 60 seconds.

[0105] Secondly, the long-term historical data is segmented, and the optimal number of clusters for each segment is determined through cluster analysis.

[0106] Finally, each data segment is classified based on the optimal number of clusters, and typical process points are extracted for each data segment.

[0107] 3) Matching of test values

[0108] Within 24 hours of the timestamp of a typical process point, find the most recent test timestamp and its corresponding test value.

[0109] 4) Indicator Integration

[0110] Based on the time delay, process tag data that affects quality indicators is matched to the test timestamp. Since the workshop samples are taken 1 hour earlier in the morning and 30 minutes earlier in the afternoon, it is necessary to match the data separately for the morning and afternoon sessions, and at the same time obtain process tag data for nearby time periods (e.g., within 15 minutes before and after the timestamp).

[0111] 5) Generation of the forecast summary table

[0112] First, the data features are obtained by dimensionality reduction of the reverse timestamp and the process tag data within the preset time period before and after it through SVD.

[0113] Secondly, the median value of the process tag number is calculated based on the reverse timestamp and the process tag number data within the preset time period before and after it;

[0114] Finally, the data characteristics and the median of the process reference number are combined to form a prediction summary table, and subsequent predictions are made based on the prediction summary table.

[0115] 6) Training the BP neural network

[0116] Specifically, the BP neural network constructs input features and target values ​​by reading and normalizing feature data, and then trains it; it uses mean squared error (MSE) as the loss function and uses the Adam optimizer to update parameters. The optimal BP neural network model is obtained through training.

[0117] The online prediction phase includes the following steps:

[0118] 1) Set prediction targets

[0119] Specifically, this embodiment generates a quality prediction value every minute to achieve high-frequency, real-time quality monitoring. This mechanism ensures that key quality indicators in the production process are always under control, making process adjustments more precise and timely, thereby improving the automation level and quality stability of the production process.

[0120] 2) Determine the historical data window

[0121] Specifically, a historical data window for prediction is determined by extrapolating 1.5 hours backward from the prediction time. The extrapolation time is set to 1.5 hours based on the time it takes for parameters to affect quality. Setting a reasonable time window ensures that the input data contains sufficient historical information, enabling the model to learn the changing trends of the production process and accurately reflect the temporal characteristics of the current production state, thus improving the accuracy and stability of the prediction.

[0122] 3) Data extraction and preprocessing

[0123] Specifically, the process tag number data within the reverse timestamp and the preceding and following 15 minutes is extracted, and tag number calculations are performed to extract the 30 calculated tag numbers and the median tag number data for each tag within 30 minutes. This step ensures the integrity of the input data, while reducing redundant information and improving data quality through data calculations, making subsequent feature extraction and model training more efficient.

[0124] 4) Feature extraction

[0125] Based on the data matrix obtained by processing the reverse timestamp and process tag number data within 15 minutes before and after, SVD is used to extract data features, which, together with the midpoints, form a complete input vector for quality prediction. During feature extraction, there may be instances where the online data is very stable, i.e., the data features are 0. In such cases, to avoid errors in subsequent nonlinear regression using a BP neural network, no prediction is made, and the previous prediction value is used as the current prediction value.

[0126] More specifically, this embodiment utilizes SVD decomposition technology to extract features from online data, specifically for real-time data. Where q is the sampling time of the sample, m is the number of bit numbers, and the corresponding data stream sliding window matrix X q Represented as:

[0127]

[0128] X q Standardized as:

[0129]

[0130] in, and They represent X respectively q The mean and standard deviation. μ q and Σ q To a certain extent, it represents the real-time operating conditions of the process, using the sliding window matrix X. q Standardization reduces the impact of different operating conditions on data flow to some extent.

[0131] Perform a square transformation on the standardized matrix based on the relationship between m and n. When m = n, the matrix needs to be transformed. Perform a square transformation to obtain

[0132] Then Perform singular value decomposition:

[0133]

[0134] Among them, U q ∈R n×n M is a left singular vector. q ∈R n×m Given a positive semi-definite diagonal matrix and V q ∈R m×m It is a right singular vector. Furthermore, and The m singular values ​​are M q The m non-zero elements on the diagonal are represented as

[0135] σ q,j =M q,j,j

[0136] Where j = 1, 2, ..., m. Here, the sample x q Through sliding window matrix Mapped to the corresponding singular values This yields the characteristics of the online data.

[0137] 5) Backpropagation (BP) neural network prediction

[0138] Specifically, the feature data and median position data are concatenated into a complete input vector, which is then input into a BP neural network for nonlinear regression prediction to generate predicted values. The BP neural network has strong self-learning and generalization capabilities, and can accurately capture complex nonlinear relationships in the production process, providing high-precision predictions for key quality indicators.

[0139] 6) Generate instantaneous predicted values

[0140] Specifically, the output value of the BP neural network is used as the instantaneous predicted value of the quality indicator, enabling the prediction result to reflect the current production status in real time. Simultaneously, this result provides a basis for subsequent smoothing processing to improve the stability of the prediction.

[0141] 7) Moving weighted mean filter

[0142] Specifically, due to periodic interference and random fluctuations in data, and the continuous dependence between the current predicted value and the predicted values ​​from previous times, a sliding weighted average filter is required. This involves selecting an appropriate sliding window and calculating the weighted average of the predicted instantaneous values ​​from the previous few minutes and the current predicted instantaneous value to generate a smoothed quality prediction value for the current moment. This method effectively reduces the impact of equipment noise or short-term operating condition fluctuations on the prediction results, making the quality prediction more stable and reliable, and improving the usability of the prediction results.

[0143] More specifically, taking a sliding window of 10, the predicted instantaneous values ​​from the previous 9 minutes and the current instantaneous predicted value are multiplied by their respective weights and then summed to calculate the smoothing quality prediction value for the current moment. The formula is as follows:

[0144]

[0145] Where b is the decay factor; t is the length of the time series; d is the normalization coefficient, used to ensure that the sum of all weights is 1; and sequence is the weight. The above formula is the moving weighted average filter formula, where is the current smooth quality prediction value.

[0146] 8) Update data

[0147] Specifically, the system stores the smoothed value and instantaneous predicted value of this forecast for use in the next forecast. This step ensures the consistency and stability of the forecasting process, enabling the system to continuously perform efficient and accurate online quality forecasts.

[0148] 9) Result storage and real-time write-back

[0149] Specifically, quality prediction points are created in the real-time database, and the prediction results are written back to the real-time database, enabling the production system to obtain the latest quality prediction data at any time and optimize process parameters accordingly. Through this closed-loop control mechanism, not only is intelligent quality management achieved, improving production efficiency, reducing quality fluctuations, and ensuring the stability and consistency of product quality, but the historical time predictions of quality can also be displayed on the management network.

[0150] like Figures 2-4 The figure shows the actual operating curves for a portion of the production process after applying this online prediction method. The blue curve represents the actual value, the green curve represents the online analyzer's analysis value, and the red curve represents the predicted value. The actual operating curves show that the predicted value of the gasoline dry point generally follows the same trend as the actual test value, with a relatively small prediction error. This is significantly better than the analysis value obtained from the online analyzer, demonstrating the effectiveness of the online prediction method used in this invention.

[0151] The pseudocode for the specific online prediction part is as follows:

[0152]

[0153]

[0154]

[0155] Example 3

[0156] The above-described embodiment 1 provides a method for online prediction of gasoline dry point quality based on big data. Correspondingly, this embodiment provides a system for online prediction of gasoline dry point quality based on big data. The system provided in this embodiment can implement the online prediction method for gasoline dry point quality based on big data of embodiment 1. The system can be implemented by 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. For relevant details, please refer to the description of embodiment 1. The system embodiment provided in this embodiment is merely illustrative.

[0157] This embodiment provides a gasoline dry point online quality prediction system based on big data, including:

[0158] The offline training module is used to determine the key process reference numbers that affect the dry point quality index of gasoline through process flow and process mechanism analysis, and to train a pre-constructed BP neural network after extracting and preprocessing the long-term historical data of the key process reference numbers.

[0159] The online prediction module is used to extract and preprocess the key process reference data corresponding to the current prediction time based on the set prediction target, input it into the BP neural network to generate instantaneous prediction values, and then generate smooth online quality index prediction values ​​through sliding weighted average filtering.

[0160] Furthermore, the online prediction module includes:

[0161] The frequency setting module is used to set the prediction target according to actual needs, that is, to configure the generation frequency of quality prediction values. This module can achieve the optimal allocation of prediction resources, ensuring timeliness while avoiding waste of computing resources.

[0162] The data acquisition module is used to extract process data within a historical time window and perform preprocessing to improve data quality and make subsequent feature extraction and model training more efficient.

[0163] The feature reduction module is used to convert the extracted high-dimensional process reference data into low-dimensional feature vectors. This module can effectively extract the information most relevant to quality prediction, remove redundant variables, and optimize the data structure.

[0164] The neural network prediction module is used to perform BP neural network regression calculations, which can accurately capture complex nonlinear relationships in the production process and make high-precision predictions.

[0165] The filtering optimization module is used to perform sliding weighted average processing on the instantaneous predicted values. This module smooths the instantaneous predicted values ​​output by the neural network, eliminates random fluctuations, and improves prediction stability.

[0166] The data storage and feedback module is used to store the prediction results and write them back to the real-time database closed-loop control mechanism, providing historical time predictions for front-end display;

[0167] The webpage front-end display module is used to show real-time forecast values ​​and historical trend curves to production site personnel. Users can select to view the forecast value at a specific moment and the forecast curve for a specific time period, providing a reference for on-site production personnel to make corresponding adjustments.

[0168] Example 4

[0169] This embodiment provides a processing device corresponding to the online quality prediction method for gasoline dry point based on big data provided in Embodiment 1. The processing device can be a processing device for a client, such as a mobile phone, laptop, tablet computer, desktop computer, etc., to execute the method of Embodiment 1.

[0170] 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 online gasoline dry point quality prediction method based on big data provided in Embodiment 1.

[0171] Preferably, 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.

[0172] Preferably, the processor can be any type of general-purpose processor such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation herein.

[0173] Example 5

[0174] The online gasoline dry point quality prediction method based on big data in 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 online gasoline dry point quality prediction method based on big data described in Embodiment 1 are loaded.

[0175] 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.

[0176] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0177] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0178] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0179] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0180] 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 online quality prediction of gasoline dry point based on big data, characterized in that, Includes the following steps: By analyzing the process flow and mechanism, the key process reference numbers affecting the dry point quality index of gasoline were determined. After extracting and preprocessing the long-term historical data of the key process reference numbers, a pre-constructed BP neural network was trained. Based on the set prediction target, the key process reference data corresponding to the current prediction time is extracted and preprocessed, then input into a BP neural network to generate instantaneous prediction values, and then processed by sliding weighted average filtering to generate smooth online quality index prediction values.

2. The method for online quality prediction of gasoline dry point based on big data as described in claim 1, characterized in that, The process involves identifying key process parameters affecting gasoline dry point quality indicators through process flow and mechanism analysis, extracting and preprocessing long-term historical data of these key process parameters, and then training a pre-constructed BP neural network, including: The process flow and mechanism were analyzed to identify the key process variables affecting the dry point quality index of gasoline. Collect long-term historical time-series data of key process variables and extract typical process points through data mining. The extracted typical process points include process tag number data and the timestamps corresponding to the process tag number data. Within 24 hours after the typical process point timestamp, find the test timestamp that is closest to the typical process point timestamp and its corresponding gasoline dry point test value. The reverse timestamp is determined based on the test timestamp, and the reverse timestamp and the gasoline dry point test values ​​and process tag data within the preset time period before and after are fused according to the preset fusion rules. The data features are obtained by dimensionality reduction of the process tag data within the back-dated timestamp and the preset time period before and after by SVD, and then a prediction summary table is formed together with the median of the process tag data within the back-dated timestamp and the preset time period before and after. The BP neural network is trained based on the prediction summary table to obtain the BP neural network prediction model.

3. The method for online quality prediction of gasoline dry point based on big data as described in claim 2, characterized in that, The process features are obtained by dimensionality reduction of the back-derived timestamp and process tag number data within the preset time periods before and after the timestamp, and are then combined with the median of the back-derived timestamp and process tag number data within the preset time periods to form a prediction summary table, including: Standardize and transform the process tag data within the backward timestamp and the preset time periods before and after it; Singular value decomposition was performed on the standardized and square-transformed process reference data using SVD decomposition to extract data features. The data features, together with the median of the process tag data within the preset time periods before and after the timestamps, are used to form a prediction summary table.

4. The method for online quality prediction of gasoline dry point based on big data as described in claim 2, characterized in that, The step of training a pre-built BP neural network based on the prediction summary table to obtain a BP neural network prediction model includes: Construct a BP neural network; The BP neural network includes an input layer, a hidden layer, and an output layer. The input layer is used to read the data features and median corresponding to the process reference number and perform normalization processing. The hidden layer contains an activation function, which is used to perform nonlinear transformation on the input features and extract deep feature information. The output layer is used to output the predicted value. Determine the loss function and training parameters, and train the constructed BP neural network using the prediction summary table.

5. The method for online quality prediction of gasoline dry point based on big data as described in claim 2, characterized in that, The process involves extracting and preprocessing the key process reference data corresponding to the current prediction time based on the set prediction target, inputting it into a BP neural network to generate instantaneous prediction values, and then using a sliding weighted average filter to generate smooth online quality index prediction values, including: Set forecasting targets based on actual needs; Based on the preset target, determine the reverse timestamp; Extract and perform tag number operations on the reverse timestamp and the process tag number data within the preset time period before and after it to form a data matrix, and calculate the median of the process tag number data; The data features are obtained by extracting features from the data matrix using SVD, and then the data features are concatenated with the median of the process reference number to form a complete input vector. The input vector is fed into the BP neural network for prediction to obtain the instantaneous predicted value of the quality index; The instantaneous predicted value is filtered by a moving weighted average to obtain the smooth online quality index prediction value at the current prediction time. Prepare data and write it back to the real-time database.

6. The method for online quality prediction of gasoline dry point based on big data as described in claim 5, characterized in that, Before inputting the input vector into the BP neural network for prediction and obtaining the instantaneous predicted value of the quality indicator, the following steps are also included: Determine if all data features extracted by SVD are 0. If so, take the previous prediction value as the current instantaneous prediction value; otherwise, perform BP neural network prediction.

7. The method for online quality prediction of gasoline dry point based on big data as described in claim 5, characterized in that, The aforementioned sliding weighted average filtering of instantaneous predicted values ​​refers to performing sliding weighted average filtering on the instantaneous predicted values ​​corresponding to several prediction times prior to the current prediction time, and using the filtering result as the smoothed online quality index prediction value for the current prediction time.

8. A gasoline dry point online quality prediction system based on big data, characterized in that, include: The offline training module is used to determine the key process reference numbers that affect the dry point quality index of gasoline through process flow and process mechanism analysis, and to train a pre-constructed BP neural network after extracting and preprocessing the long-term historical data of the key process reference numbers. The online prediction module is used to extract and preprocess the key process reference data corresponding to the current prediction time based on the set prediction target, input it into the BP neural network to generate instantaneous prediction values, and then generate smooth online quality index prediction values ​​through sliding weighted average filtering.

9. 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 7.

10. A computing device, characterized in that, include: One or more processors and a memory, wherein the memory stores one or more programs and is 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 7.

Images