A petrochemical intelligent report generation method and system based on a professional model

Abstract

This invention discloses a method and system for generating intelligent petrochemical reports based on a professional model, belonging to the field of manufacturing business information management technology. The method includes: acquiring real-time monitoring data from a petrochemical enterprise's production equipment monitoring system; standardizing the format of the real-time monitoring data, detecting outliers, and completing missing values ​​to construct a petrochemical data set; identifying data categories, data correlations, and time trends based on a long short-term memory neural network; determining the report type and display format based on feature tags; generating personalized report templates by combining user role information and report requirement parameters; updating report content, reconstructing templates, and adjusting strategies based on data changes; and re-determining the report type and display format when data characteristics change significantly. Through the technical solution of this invention, intelligent generation and dynamic updating of petrochemical reports are achieved, improving the accuracy of report type matching, personalization adaptability, and report generation efficiency and real-time performance.

Description

Technical Field

[0001] This invention relates to the field of manufacturing business information management technology, and in particular to a method for generating intelligent petrochemical reports based on a professional model and a system for generating intelligent petrochemical reports based on a professional model. Background Technology

[0002] Petrochemical reports play a crucial role in the production management, operation monitoring, decision analysis, resource allocation, and compliance management of petrochemical enterprises. During production operations, petrochemical companies continuously generate real-time monitoring data, including temperature, pressure, flow rate, and liquid level. To facilitate production monitoring, operation analysis, and management decision-making, this data is typically compiled into corresponding petrochemical reports to reflect equipment operating status, process parameter changes, and anomalies. Currently, the generation of petrochemical reports usually relies on staff extracting relevant data from production equipment monitoring systems and then organizing and filling it out according to preset templates, or using fixed electronic templates to retrieve and display data from a database. In some scenarios, semi-automatic methods based on SQL databases and Excel macros are used for data import, statistics, and filling, but these still essentially rely on pre-set fixed templates and manual experience for processing.

[0003] However, existing methods for generating petrochemical reports mostly rely on fixed templates and manual configuration, resulting in low levels of automation. On one hand, petrochemical production data typically exhibits strong time-series characteristics and complex variable relationships. Data formats and quality vary across different equipment and monitoring points. Without a unified data preprocessing mechanism, outliers, missing values, or inconsistent formats can easily affect report generation results. On the other hand, existing solutions usually only directly display the collected data, lacking the ability to automatically identify data categories, data relationships, and time trends. This makes it difficult to automatically match suitable report types and display formats based on the characteristics of the real-time monitoring data itself. The selection of existing report types often still depends on the operator's experience, thus limiting the effectiveness and analytical relevance of the generated reports.

[0004] Furthermore, existing technologies typically use uniform report templates to output data to different users, failing to fully consider the differences in report content and presentation focus among different user roles. For example, operators are more concerned with real-time monitoring and alarm information, while technical analysts are more concerned with historical trends and analytical content. Without a personalized template generation mechanism based on role information and report requirement parameters, it is difficult to improve the matching degree between reports and actual use scenarios, and it is also difficult to meet the actual needs of different users in terms of data display depth, display style, and functional configuration.

[0005] Furthermore, existing reporting systems typically only support static generation or simple data replacement, lacking a dynamic update mechanism adapted to real-time monitoring data changes. When underlying data changes, existing solutions often fail to update report content promptly; when data characteristics change significantly, it is also difficult to further redefine report types, display formats, and template structures. In petrochemical production scenarios, real-time monitoring data updates frequently. If manual re-triggering of data collection, processing, and reporting processes is still required, it will be difficult to meet the timeliness requirements of reports. Simultaneously, existing technologies lack corresponding data processing strategies and report generation strategy adjustment mechanisms to adapt to changes in data scale or user needs, resulting in decreased processing efficiency and insufficient report generation efficiency and adaptability when dealing with large data volumes.

[0006] Therefore, existing technologies have at least the following problems: real-time monitoring data for petrochemical products lacks a unified and effective preprocessing mechanism; lacks the ability to automatically identify data categories, data correlations, and time trends, making it difficult to achieve intelligent recommendations for report types and display formats; lacks a personalized report template generation mechanism based on user roles and requirement parameters; and lacks the ability to dynamically update reports and adaptively adjust strategies in response to data changes and changes in data characteristics. Summary of the Invention

[0007] To address the aforementioned issues, this invention provides a method and system for generating intelligent petrochemical reports based on a professional model. By preprocessing real-time monitoring data from petrochemical enterprises and utilizing a long short-term memory neural network to automatically identify data categories, data correlations, and time trends, the system can intelligently determine the report type and display format based on the identification results. This reduces reliance on manual experience in report generation and solves the problem of inaccurate matching between reports and data features caused by existing methods that rely on manual experience to select report types. Furthermore, by combining the target user's role information and report requirement parameters to generate personalized report templates, and automatically updating reports when data changes and automatically reconstructing report types and template structures when data characteristics undergo significant changes, the system improves the automation, real-time performance, relevance, and adaptability of petrochemical report generation.

[0008] To achieve the above objectives, this invention provides a method for generating intelligent petrochemical reports based on a professional model, comprising: Acquire real-time monitoring data from the petrochemical enterprise's production equipment monitoring system and transmit the real-time monitoring data to a data storage system; The real-time monitoring data is preprocessed to unify the data format, detect outliers, fill in missing values, and construct a petrochemical data set; Using a long short-term memory neural network as a time-series feature recognition model, the time-series feature recognition model is pre-trained using the petrochemical data set; The real-time data from the petrochemical dataset is input into the pre-trained time-series feature recognition model, which outputs feature labels that characterize the data category, data correlation, and time trend. Based on the feature tags, the report type and display format corresponding to the feature tags are determined from the preset report type recommendation rules; Obtain the target user's role information and report requirement parameters, and generate a personalized report template based on the report type, the display format, the role information, and the report requirement parameters; The latest data corresponding to the personalized report template is obtained from the data source of the petrochemical enterprise, and the latest data is filled into the personalized report template to generate petrochemical reports; Monitor data changes in the data source, and when data changes are detected, obtain the latest data and update the petrochemical reports; After detecting data changes, feature labels are re-output based on the updated data. When a significant change in the feature labels is detected, the report type and display format determination, personalized report template generation, and report update are re-executed. Adaptively adjust data processing and report generation strategies based on changes in data volume and user needs.

[0009] In the above technical solution, preferably, the step of acquiring real-time monitoring data from the petrochemical enterprise production equipment monitoring system and transmitting the real-time monitoring data to the data storage system includes: Temperature, pressure, flow, and liquid level data of production equipment are collected through sensors and PLC industrial control systems. The temperature data, pressure data, flow rate data, and liquid level data are transmitted to the data storage system via the Modbus industrial network protocol. A distributed database system is used as the data storage system to store the real-time monitoring data.

[0010] In the above technical solution, preferably, the preprocessing includes: The real-time monitoring data is processed to unify the data format, and the various variables in the real-time monitoring data are standardized using Z-score. The DBSCAN algorithm is used to detect and remove outliers from the standardized data. For variables containing missing values, the correlation coefficient between the variables containing missing values ​​and other variables is calculated. The two variables with the largest correlation coefficients are selected to construct a multiple linear regression model. The regression coefficients of the multiple linear regression model are estimated using the least squares method. Based on the multiple linear regression model, the missing values ​​are filled in to obtain the petrochemical data set.

[0011] In the above technical solution, preferably, the step of using a long short-term memory neural network as a time-series feature recognition model and pre-training the time-series feature recognition model using the petrochemical dataset includes: The petrochemical dataset is divided into a training set, a validation set, and a test set in a ratio of 70%:15%:15%. Labels are set for the training set, the validation set, and the test set, and the labels include data category labels, data correlation labels, and time trend labels; The long short-term memory neural network is trained based on the training set, using the Adam optimizer, and an early stopping strategy is executed based on the validation set. The trained Long Short-Term Memory (LSTM) neural network is evaluated based on the test set, and the evaluated LSM neural network is used as a pre-trained temporal feature recognition model.

[0012] In the above technical solution, preferably, the preset report type recommendation rule includes a mapping relationship between feature tags and report types and display formats. The step of determining the report type and display format corresponding to the feature tags from the preset report type recommendation rule based on the feature tags includes: When the feature label represents a continuous trend and the time trend is a stable trend or a random trend, the report type is determined to be a scatter plot and the display format is a coordinate system. When the feature label represents a continuous trend and the time trend is an upward trend, a downward trend, or a periodic trend, the report type is determined to be a line chart, and the display format is a time axis. When the feature label represents a discrete type, the report type is determined to be a bar chart, and the display format is a classification axis.

[0013] In the above technical solution, preferably, the step of obtaining the target user's role information and report requirement parameters, and generating a personalized report template based on the report type, the display format, the role information, and the report requirement parameters, includes: Perform role classification on target users, wherein the role classification includes at least operator roles and engineer roles; Based on the role classification results, and in conjunction with the report type, the display format, and the report requirement parameters, determine the table header, column headings, data area layout, and chart positions in the report template; When the target user is an operator, the report template generates content that includes real-time data monitoring information, alarm information, and real-time data update binding relationships. When the target user is an engineer, the report template generates content that includes historical trend data, analysis tools, and data export and sharing functions.

[0014] In the above technical solution, preferably, the step of re-outputting feature labels based on the updated data after detecting data changes, and re-executing the report type and display format determination, personalized report template generation, and report update when a significant change in the feature labels is detected, includes: The feature labels are re-output based on the updated real-time monitoring data; When the change in the real-time monitoring data corresponding to the re-output feature label exceeds one-half of the current value of the corresponding real-time monitoring data before the update, it is determined that the feature label has changed significantly, and the report type and display format determination, personalized report template generation, and report update are re-executed. The adaptive adjustment of data processing and report generation strategies based on changes in data scale and user needs includes: When the data size is less than a preset threshold, a sequential processing method is used; when the data size is greater than the preset threshold, a parallel processing method is used; and the report generation process and report display focus are adjusted based on the user needs analysis results.

[0015] This invention also proposes a petrochemical intelligent report generation system based on a professional model, including a data acquisition module, a data storage module, a data preprocessing module, a model training module, a feature recognition module, a report recommendation module, a personalized report generation module, a report update module, and an adaptive optimization module; The data acquisition module is used to acquire real-time monitoring data from the petrochemical enterprise's production equipment monitoring system; The data storage module is connected to the data acquisition module and is used to receive and store the real-time monitoring data; The data preprocessing module is connected to the data storage module and is used to preprocess the real-time monitoring data to unify the data format, detect outliers, fill in missing values, and construct a petrochemical data set. The model training module is connected to the data preprocessing module and is used to use a long short-term memory neural network as a time-series feature recognition model and to pre-train the time-series feature recognition model using the petrochemical data set. The feature recognition module is connected to the data preprocessing module and the model training module respectively, and is used to input real-time data from the petrochemical dataset into the pre-trained time-series feature recognition model, and output feature labels that characterize the data category, data correlation and time trend. The report recommendation module is connected to the feature recognition module and is used to determine the report type and display format corresponding to the feature tag from the preset report type recommendation rules based on the feature tag; The personalized report generation module is connected to the report recommendation module and is used to obtain the target user's role information and report requirement parameters, and generate a personalized report template based on the report type, the display format, the role information and the report requirement parameters. The report update module is connected to the data storage module, the feature recognition module, the report recommendation module, and the personalized report generation module, respectively. It is used to obtain the latest data corresponding to the personalized report template from the data storage module and fill the personalized report template with the latest data to generate petrochemical reports. It is also used to monitor data changes in the data storage module and update the petrochemical reports when data changes are detected. Furthermore, it is used to re-execute the report type and display format determination, personalized report template generation, and report update when the feature tags change significantly. The adaptive optimization module is connected to the data preprocessing module and the report update module, and is used to adaptively adjust the data processing strategy and report generation strategy according to changes in data scale and user needs.

[0016] In the above technical solution, preferably, the data acquisition module includes a sensor interface and a PLC interface, and the data storage module includes a protocol transmission unit and a distributed database; The sensor interface and the PLC interface are used to collect temperature data, pressure data, flow data and liquid level data of the production equipment; The protocol transmission unit is used to transmit the real-time monitoring data to the distributed database via the Modbus industrial network protocol. The data preprocessing module includes a standardization unit, an outlier detection unit, and a missing value completion unit. The standardization unit is used to perform Z-score standardization on various variables in the real-time monitoring data. The outlier detection unit is used to detect outliers using the DBSCAN algorithm. The missing value completion unit is used to complete missing values ​​by using a multiple linear regression model and estimating regression coefficients using the least squares method.

[0017] In the above technical solution, preferably, the personalized report generation module includes a role recognition unit and a template generation unit, the report update module includes a report generation unit, a data monitoring unit and a reconstruction update unit, and the adaptive optimization module includes a data scale determination unit and a requirement analysis unit; The role recognition unit is used to perform operator role recognition or engineer role recognition for the target user; The template generation unit is used to generate real-time monitoring report templates or trend analysis report templates based on role recognition results, report types, display formats, and user requirement parameters. The report generation unit is used to obtain the latest data corresponding to the personalized report template from the data storage module, and fill the latest data into the personalized report template to generate a petrochemical report; The data monitoring unit is used to continuously track data changes in the data storage module and trigger report content updates when data changes are detected. The reconstruction and update unit is used to re-execute the determination of report type and display format, generation of personalized report templates, and report update when the feature labels change significantly. The data size determination unit is used to switch between sequential processing and parallel processing based on the data size. The requirement parsing unit is used to parse user requirement parameters and adjust the report generation process and report display focus based on the parsing results.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) By acquiring real-time monitoring data from the production equipment monitoring system of petrochemical enterprises, and by applying a unified data format to the real-time monitoring data, using the DBSCAN algorithm to detect outliers, and using regression interpolation to complete missing values, a petrochemical data set is constructed. This achieves unified organization and quality correction of multi-source heterogeneous petrochemical data, reduces the impact of inconsistent data formats, outlier interference, and missing values ​​on the report generation results, and provides a relatively stable data foundation for subsequent feature recognition and report generation.

[0019] (2) By using a long short-term memory neural network as a time-series feature recognition model, real-time data in the petrochemical dataset is identified, and feature labels representing data categories, data correlations and time trends are output. Based on the feature labels, the corresponding report type and display format are determined from the preset report type recommendation rules. This realizes the automatic adaptation of report type and display format to data features, reduces the dependence of report type selection on human experience judgment, and improves the matching degree between report display mode and petrochemical data features.

[0020] (3) By obtaining the target user's role information and report requirement parameters, and generating personalized report templates based on report type, display format, role information and report requirement parameters, the report content and display structure are adapted to the different user needs, which makes the generated petrochemical reports more in line with the focus of operators and technical analysts, and improves the relevance and practicality of the reports.

[0021] (4) Through a two-level linkage report update mechanism, petrochemical reports are updated when data changes are detected, and the report type and display format determination, personalized report template generation and report update are re-executed when the data characteristics change significantly. At the same time, the data processing strategy and report generation strategy are adaptively adjusted according to changes in data scale and user needs, realizing the linkage of report content update, report structure adjustment and processing strategy optimization, and improving the real-time performance, adaptability and processing efficiency of petrochemical report generation. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a method for generating intelligent petrochemical reports based on a professional model, as disclosed in one embodiment of the present invention. Figure 2 This is a schematic flowchart of a regression interpolation method disclosed in one embodiment of the present invention; Figure 3 This is a flowchart illustrating the DBSCAN algorithm disclosed in one embodiment of the present invention; Figure 4 This is a flowchart illustrating a long short-term memory neural network according to an embodiment of the present invention. Figure 5 This is a flowchart illustrating the optimization algorithm disclosed in one embodiment of the present invention. Detailed Implementation

[0023] 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 embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The present invention will now be described in further detail with reference to the accompanying drawings: like Figure 1 As shown, a petrochemical intelligent report generation method based on a professional model according to the present invention includes: Big data technology is used to continuously acquire real-time monitoring data from the production equipment monitoring systems of petrochemical enterprises and transmit the real-time monitoring data to the data storage system, providing a unified data source foundation for subsequent preprocessing and model training. Data generated during petrochemical production is typically generated at a high frequency and may exist in structured or unstructured forms, with sources covering various types of sensor equipment such as temperature, pressure, flow rate, and liquid level.

[0025] Real-time monitoring data undergoes preprocessing, which includes three aspects: First, standardizing the data format eliminates semantic ambiguity from different data sources, ensuring comparability across various data types. Second, outlier detection identifies and eliminates abnormal readings caused by equipment malfunctions, signal interference, etc., preventing them from contaminating subsequent analysis. Finally, missing values ​​are filled in to ensure the integrity of the dataset. This preprocessing standardizes the format of data from different sources, eliminates structural and representational differences between data sources, and constructs a complete petrochemical dataset, providing a high-quality data foundation for model pre-training and feature recognition.

[0026] Based on this petrochemical dataset, a Long Short-Term Memory (LSTM) neural network was used as the time-series feature recognition model. The model was pre-trained using the petrochemical dataset to enable it to recognize petrochemical time-series data. The core characteristic of petrochemical data lies in its temporal nature; indicators such as temperature, pressure, flow rate, and liquid level change continuously over time, and complex relationships exist between these variables. The LSTM, through its gating mechanism, can effectively capture long-term dependencies in time-series data, making it more suitable as a feature recognition model for petrochemical data compared to traditional statistical analysis methods or ordinary neural networks. Using the constructed petrochemical dataset, combined with prior knowledge and historical data from the petrochemical industry, the model was pre-trained to initially understand the basic patterns of petrochemical data in dimensions such as data type (continuous / discrete), data correlation (positive / negative correlation), and time trend (stable / increasing / decreasing / periodic / random).

[0027] Real-time data from the petrochemical dataset is input into a pre-trained time-series feature recognition model. The model automatically analyzes the input data and outputs feature labels representing data categories, data correlations, and time trends. For example, for a segment of continuous, stable real-time data with positive correlations between variables, the feature labels output by the model reflect the specific values ​​of the above three dimensions, providing objective data basis for subsequent report type recommendations.

[0028] Based on feature tags, the system determines the report type and display format corresponding to the feature tags from pre-built report type recommendation rules, thus eliminating reliance on manual chart selection for report generation. These recommendation rules pre-establish a mapping relationship between data features and report presentation formats, automatically matching the most suitable report type for intuitively presenting the data patterns based on the feature tags. This ensures that the report's display format adapts to the inherent characteristics of the data, avoiding subjective errors from manual selection.

[0029] After obtaining the report type and display format, the system acquires the target user's role information and report requirement parameters, and generates a personalized report template based on these parameters. Different user roles within petrochemical companies have fundamentally different needs for report content and style. By incorporating user role information into the report template generation process, targeted customization of report content and style is achieved, allowing the same set of data to be presented in a format best suited to a specific user role. This template not only defines the report's layout structure but also the subsequent data filling and display methods.

[0030] Using big data technology, the latest data corresponding to the personalized report template is obtained from the data sources of petrochemical enterprises. The latest data is then populated into the personalized report template according to the table header, column headings, data area layout and chart position defined in the report template to generate petrochemical reports.

[0031] During execution, the system continuously monitors data changes in the data source. Once a data update is detected in the data source, the latest data is immediately retrieved and populated into the existing report template to update petrochemical reports, ensuring that the reports always reflect the latest status of petrochemical production and meet the company's requirements for real-time reporting.

[0032] After detecting data changes, the time-series feature recognition model is called again to output feature labels based on the updated data. The re-output feature labels are then compared with the feature labels before the update. When a significant change in feature labels is detected, it indicates that the overall pattern of petrochemical data has undergone a fundamental change. At this point, relying solely on conventional data filling cannot guarantee the matching of report structure and data features. Therefore, the report type and display format determination, personalized report template generation, and report update are re-executed to achieve adaptive reconstruction of the report structure.

[0033] To address the dynamic changes in data volume in petrochemical enterprises, an adaptive optimization mechanism is designed. This mechanism automatically selects the optimal data processing strategy based on the current data volume and dynamically adjusts the report generation process and report display focus according to changes in user needs, ensuring that the system can maintain efficient report generation capabilities under different loads and demand scenarios.

[0034] In this implementation, by deeply integrating artificial intelligence technology with big data technology, and using the LSTM model to automatically identify petrochemical data features as the core driver, a complete intelligent method closed loop from data collection to dynamic report updates is constructed. This achieves automation and intelligence of the entire process of generating petrochemical reports, effectively overcoming the inherent defects of traditional manual and semi-automatic methods in terms of accuracy, real-time performance, and flexibility.

[0035] In the above embodiments, preferably, acquiring real-time monitoring data from the petrochemical enterprise's production equipment monitoring system and transmitting the real-time monitoring data to the data storage system includes: By deploying sensors at key nodes of the equipment and through a PLC industrial control system, temperature, pressure, flow, and liquid level data of the production equipment can be collected, enabling high-frequency continuous acquisition of these four types of real-time monitoring data. These data are all critical monitoring quantities in the petrochemical production process, directly reflecting equipment operating status and process changes.

[0036] Temperature, pressure, flow, and liquid level data are transmitted to the data storage system via the Modbus industrial network protocol. As a widely used industrial communication protocol in the petrochemical industry, Modbus offers excellent device compatibility and transmission reliability, ensuring a stable data transmission relationship between the production site and the data storage system. Data is transmitted in real-time without loss or distortion, meeting the real-time requirements of petrochemical production data transmission.

[0037] A distributed database system is used as the data storage system to store real-time monitoring data. The distributed database has the ability to efficiently store and quickly retrieve large-scale data, and can adapt to the characteristics of large data volume and high update frequency in petrochemical production scenarios. This allows the real-time monitoring data to be continuously written, quickly retrieved, and called, thus providing a unified data foundation for subsequent preprocessing, model training, feature recognition, and report generation.

[0038] In this implementation, a three-layer architecture consisting of a sensor and PLC acquisition layer, a Modbus protocol transmission layer, and a distributed database storage layer enables unified access and centralized storage of various real-time monitoring data from petrochemical sites. This provides complete hardware and protocol support for the real-time acquisition and reliable storage of petrochemical data, ensuring the quality of data sources for subsequent processing.

[0039] In the above embodiments, preferably, the preprocessing includes data format unification processing, standardization processing, outlier detection, and missing value completion.

[0040] The real-time monitoring data undergoes standardized format processing, unifying data from different sources and with varying field structures into a unified data organization. Z-score standardization is applied to each variable in the real-time monitoring data, converting the raw values ​​of each variable into standardized values ​​with a mean of 0 and a standard deviation of 1. This ensures comparability across different variables on a numerical scale, preventing significant differences in dimensions such as temperature, pressure, flow rate, and liquid level from affecting subsequent identification results. This standardization process ensures that different variables are given relatively consistent weights when entering subsequent analysis stages.

[0041] The formula for Z-score standardization is as follows: in, x This is the original data. μ The mean, σ The standard deviation is denoted as .

[0042] According to this formula, the specific process of Z-score standardization is as follows: subtract the mean of the variable from the original data and divide by its standard deviation, where the mean and standard deviation are calculated from the historical data of each variable. After standardization, variables with different dimensions are unified to the same scale, so that each variable has the same weight basis in the subsequent outlier detection and feature identification process, avoiding the influence of a certain variable on the accuracy of identification due to differences in dimensions. For example, after performing Z-score standardization on flow data 85, 90, 70, 80, 85 and temperature data 38, 39, 37, 41, 39 respectively, the standardized results of flow are 0.12, 0.70, -1.63, -0.47, 1.28, and the standardized results of temperature are -0.60, 0.15, -1.35, 1.65, 0.15, both of which are unified to a directly comparable scale.

[0043] Because petrochemical production sites may experience instantaneous abnormal equipment readings, data acquisition noise, or communication disturbances, directly inputting data containing outliers into the model without identifying and removing them can easily lead to deviations in subsequent model training and report display from actual operating conditions. Therefore, the DBSCAN algorithm is used to detect outliers in the standardized data. This algorithm clusters data points based on their density distribution, eliminating the need to pre-specify the number of clusters and effectively identifying outliers in petrochemical time-series data. After detection, data points marked as noise are removed as outliers, ensuring that all data entering subsequent processes represents normal production status data.

[0044] like Figure 2As shown, during data acquisition or transmission, monitoring data at some time points may be missing due to sensor malfunctions or network interruptions. For variables containing missing values, the correlation coefficients between these variables and other variables are calculated. The two variables with the highest correlation coefficients with the variable containing missing values ​​are selected as independent variables. A regression model is constructed using multiple linear regression analysis. The model expression includes an intercept term, two regression coefficients, and a random error term. The least squares method is used to estimate the regression coefficients of the multiple linear regression model to minimize the sum of squared residuals. The constructed multiple linear regression model is used to predict the values ​​at the missing locations, and the predicted values ​​are filled into the corresponding positions in the dataset to obtain a complete petrochemical dataset. When multiple variables contain missing values, they are filled in sequentially along the time axis. When processing the current variable, the missing values ​​of subsequent variables are temporarily set to zero to ensure the orderly completion process. Through correlation screening and regression completion, missing data can be recovered without destroying the original data structure, making the resulting petrochemical dataset more complete.

[0045] For example, consider a set of data that has not yet been standardized by Z-score, where temperature is 35, 39, and 41 degrees Celsius; pressure is 85, missing, and 100; flow rate is 78, 90, and 85; and liquid level is 60, 96, and 77. In this case, pressure has missing values, so pressure should be defined as the dependent variable, and temperature, flow rate, and liquid level should be defined as independent variables. If two or more variables have missing values ​​(i.e., multiple dependent variables), they need to be filled sequentially along the time axis. During the filling process, missing values ​​after the time axis are temporarily set to 0. For example, consider a set of data that has not yet been standardized by Z-score, where temperature is 35, 39, and 41 degrees Celsius; pressure is 85, missing, 100, and 87; flow rate is 78, 90, 85, and 96; and liquid level is 60, 96, 77, and 86. This data has two missing values. The missing pressure value will be filled in first by regression interpolation in chronological order. When filling in the missing pressure value by regression interpolation, the missing temperature value will be temporarily set to 0.

[0046] The formula for calculating the correlation coefficient r between any two variables is as follows: in, n It represents the number of observations, where X is the dependent variable and Y is the independent variable. i and Y i It is the i-th observation of variables X and Y. and It is the mean of X and Y. For example, when calculating the correlation coefficient between pressure and temperature, pressure is X and temperature is Y.

[0047] Specifically, the two variables with the highest correlation coefficients among the independent variables are used as independent variables to construct the regression model. Multiple linear regression analysis is employed. The expression for the multiple regression model is: in, It is the dependent variable. It is the independent variable with the highest correlation coefficient. It is the second largest independent variable in terms of correlation coefficient. It is the intercept term. These are regression coefficients, representing the degree of influence of the independent variable on the dependent variable. It is the random error term, representing the influence of other factors not included in the model on the model's performance. The impact of this. The parameters are estimated using the least squares method, with the goal of finding a set of regression coefficients. This minimizes the sum of squared residuals (RSS). Defined as observation value Compared with the predicted value The difference between them, namely ,and The formula for calculating RSS is: in, It is the number of observations. For the i-th residual, For the observed values, These are predicted values.

[0048] Specifically, such as Figure 3 As shown, the specific steps of the DBSCAN algorithm implementation are as follows: (1) Dynamically determine the range of values ​​for the radius parameter ε by plotting a scatter plot of the data. Select a reasonable value for k (5 in this invention), and calculate the distance from each point to its k-th nearest neighbor, called the k-distance. Arrange the k-distances of all points in ascending order and plot the k-distance graph. In the k-distance graph, find a clear elbow point, that is, the place where the distance value begins to increase sharply. The k-distance value corresponding to this elbow point can be used as a candidate value for ε. For example, if the elbow point appears at a distance value of 0.5, then you can try setting ε to 0.5.

[0049] (2) Select the minimum number of points MinPts according to the dimension of the data. In this invention, the value is 5.

[0050] (3) Calculate the Euclidean distance from each point in the dataset to other points. For each point, determine whether the number of points in its neighborhood reaches MinPts. Classify the data points according to the definitions of core points, boundary points, and noise points. A core point is a point that contains at least MinPts data points within its specified radius ε. In other words, a core point is a point with a sufficiently high density, and its ε neighborhood contains at least MinPts points (including itself). A boundary point is a point that contains fewer than MinPts data points within its specified radius ε, but it belongs to the ε neighborhood of at least one core point. In other words, a boundary point is not itself a core point, but it is located in the neighborhood of some core point. A noise point is a point that is neither a core point nor a boundary point. In other words, a noise point contains fewer than MinPts data points within its specified radius ε, and it does not belong to the ε neighborhood of any core point. The Euclidean distance formula is as follows: in, It is the dimension of each point. for The first point Dimension value, for The first point The value of the dimension.

[0051] (4) The last point marked as a noise value is an outlier. Delete the outlier.

[0052] In this implementation, Z-score standardization eliminates the influence of dimensions, DBSCAN eliminates abnormal interference, and multiple linear regression completes missing data. These three steps work in sequence to systematically ensure the standardization, purity, and completeness of the petrochemical data set, improve the availability and stability of the original monitoring data, provide high-quality data input for the feature recognition of the subsequent LSTM model, reduce the impact of abnormal and missing data on the quality of subsequent model recognition and report generation, and improve the accuracy of reports from the data source.

[0053] like Figure 4 As shown, in the above embodiment, preferably, a long short-term memory neural network is used as the time-series feature recognition model, and the time-series feature recognition model is pre-trained using a petrochemical dataset, including: The petrochemical dataset is divided into training, validation, and test sets in a 70%:15%:15% ratio. Taking a dataset of 1000 samples as an example, 700 samples are used for training, 150 for validation, and 150 for testing. The training set is used for learning model parameters, the validation set is used for hyperparameter tuning and overfitting monitoring during training, and the test set is used for final model performance evaluation. These three sets are independent of each other, ensuring the objectivity of the evaluation results. This division provides a clear division of labor between the training, validation, and testing phases, which is beneficial for balancing model fitting ability, training process control, and final recognition performance evaluation.

[0054] Three-digit feature labels were assigned to samples in the training, validation, and test sets. The first digit represents the data category, with 0 indicating continuous data and 1 indicating discrete data. The second digit represents the data correlation, with 0 indicating positive correlation and 1 indicating negative correlation. The third digit represents the time trend, with 0 indicating a stable trend, 1 indicating an upward trend, 2 indicating a downward trend, 3 indicating a periodic trend, and 4 indicating a stochastic trend. The values ​​of these three digits collectively describe the complete characteristics of a petrochemical time-series data set, providing a structured training objective for supervised learning. By setting these labels on the training data, the model can learn the comprehensive characteristics of petrochemical data in terms of category, variable relationships, and time changes.

[0055] For complex tasks, a four-layer LSTM is used with 32, 64, 128, and 256 neurons respectively. The activation function inside the LSTM is typically the tanh function, which maps input values ​​to between -1 and 1, helping the model learn non-linear relationships in the data. Two fully connected layers can be added after the LSTM layers, with 150 and 2 neurons respectively. The activation function of the output layer is determined based on the task. For classification tasks, the softmax activation function is used to transform the output into a probability distribution.

[0056] During model training, the Long Short-Term Memory (LSTM) neural network is trained on the training set, with the Adam optimizer used for parameter updates and the Cross-Entropy loss function as the model's loss function. The Adam optimizer combines the advantages of Adagrad and RMSProp, automatically adjusting the learning rate of each parameter based on the first and second moment estimates of the gradient. The initial learning rate can be set to 0.001. The training batch size is set to 256, and the number of training epochs is set to 20. It exhibits strong convergence stability in complex nonlinear scenarios such as petrochemical time-series data. Since the LSM neural network is suitable for processing time-dependent sequence data, it can memorize and identify historical changes in petrochemical monitoring data. During training, an early stopping strategy is implemented based on the validation set. When the validation set loss no longer decreases within a certain number of consecutive training epochs, training is terminated early to prevent overfitting on the training set due to overtraining and to ensure the model's generalization ability to unknown real-time data.

[0057] After training, the trained Long Short-Term Memory (LSTM) neural network is evaluated based on a test set. Accuracy is used as the evaluation metric to calculate the model's correct output of three-dimensional feature labels. The evaluated LSM is then used as a pre-trained temporal feature recognition model for subsequent feature label output of real-time data. This training and evaluation process ensures that the model subsequently deployed has relatively stable temporal recognition capabilities.

[0058] In this implementation, by dividing the petrochemical dataset into training, validation, and test sets, and employing a long short-term memory neural network pre-training process with an Adam optimizer and early stopping strategy, the learning effect and generalization ability of the model on petrochemical time-series data features can be improved. This enables the model to accurately identify the features of petrochemical data in the three dimensions of category, correlation, and trend, providing a reliable feature basis for subsequent intelligent recommendation of report types.

[0059] In the above embodiments, preferably, the preset report type recommendation rules include the mapping relationship between feature tags and report types and display formats.

[0060] Based on feature tags, the report type and display format corresponding to the feature tags are determined from the preset report type recommendation rules. This rule table comprehensively considers the optimal visualization presentation method corresponding to different combinations of data categories and time trends, including: When the feature labels represent continuous data and the time trend is stable or random, the report type is determined to be a scatter plot, and the display format is a coordinate system. The variable values ​​of this type of data fluctuate little over time or are randomly distributed. This mapping method is suitable for observing the discrete distribution of continuous data at a certain time or within a certain statistical window. It can intuitively present the distribution of each variable in the coordinate system and the correlation between variables, which is helpful in reflecting the fluctuation distribution of continuous data.

[0061] When the feature labels represent continuous data and the time trend is upward, downward, or cyclical, the report type is determined to be a line chart, and the display format is a time axis. This type of data has obvious time direction or cyclical patterns. Since the trend changes of continuous data are more suitable to be displayed in the time dimension, using a line chart can more intuitively show the upward, downward, or cyclical fluctuations of data over time. It can clearly present the trend of data evolution over time, making it easier for engineers to identify the trend direction, predict inflection points, and analyze cyclical patterns.

[0062] When feature labels represent discrete data, the report type is determined to be a bar chart, and the display format is a category axis. This approach is more suitable for intuitive comparison of data from different categories. Discrete data uses categories or ranks as the basic unit. Through this mapping relationship, the report type and display format can be matched with the data characteristics. By comparing the height of each category on the category axis, the differences in quantity or frequency between different categories can be intuitively presented, avoiding reliance on manual experience for chart selection.

[0063] Specifically, the recommended rules for report types are shown in the table below.

[0064] In this implementation, by pre-setting the mapping relationship between feature labels and report types and display formats, the output results of the time-series feature recognition model are directly associated with the report visualization format. This enables the system to automatically recommend appropriate chart types based on the identified data categories and time trends, improving the adaptability between the report display format and data features, and ensuring the consistency between the report presentation format and the inherent laws of the data.

[0065] In the above implementation, preferably, the target user's role information and report requirement parameters are obtained, and a personalized report template is generated based on the report type, display format, role information, and report requirement parameters, including: The target users are classified into roles, including at least operator and engineer roles. In petrochemical enterprises, the purpose and focus of different positions in the use of reports are fundamentally different. Operators need to grasp the real-time operating status of equipment and abnormal alarm information on the production site as soon as possible, while engineers need to conduct in-depth analysis based on historical trend data to identify potential equipment deterioration patterns or production bottlenecks.

[0066] Based on the role classification results, and in conjunction with the report type, display format, and report requirement parameters, the headers, column headings, data area layout, and chart positions in the report template are determined. In other words, template generation is not simply adding different titles to the same report, but rather first determining the report's structural framework based on roles and requirements, and then filling in the corresponding content within that framework.

[0067] When the target user is an operator, the report template generates real-time data monitoring information including key equipment parameters such as temperature, pressure, flow rate, and liquid level; alarm information triggered when data exceeds preset safety thresholds; and a real-time data update binding relationship established using a publisher-subscriber pattern, ensuring that the data in the report automatically refreshes as the data source changes, allowing operators to see the latest equipment status without manual triggering. The report style uses a clear grid layout, with color coding to distinguish data status: green indicates normal status, and red indicates alarm status. Key indicators are highlighted with bold text and icons, emphasizing intuitiveness and readability, enabling operators to quickly respond to abnormal situations. This type of template can also use a simpler layout, highlighting key indicators and alarm information so that operators can quickly view the current operating status.

[0068] When the target user is an engineer, the report template generates trend data content containing historical data from multiple time periods. It integrates analysis tools such as regression analysis, time series analysis, and cluster analysis, allowing engineers to perform data analysis directly on the report interface without switching to other tools. It features advanced interactive charts that support zooming, panning, and data filtering, as well as custom view functionality that allows engineers to customize data columns, chart types, and filter conditions. It also provides data export functionality, supporting export of report data to Excel, CSV, and other file formats, and offers report sharing capabilities for easy collaboration between engineers and colleagues. This template focuses more on viewing historical trends, analysis, and subsequent sharing applications, making it suitable for engineers to conduct more in-depth processing of operational trends, variable relationships, and analysis results.

[0069] In this implementation, by first determining the template structure and then filling in differentiated content according to different roles, the report template is adapted to different user roles and different report needs. This makes the report content and display focus more in line with the actual use scenarios of operators and engineers, significantly improving the relevance and practical value of the reports.

[0070] In the above embodiments, preferably, after detecting data changes, feature labels are re-output based on the updated data, and when significant changes in feature labels are detected, the determination of report type and display format, generation of personalized report templates, and report updates are re-executed, including: When the system detects data changes, it first re-outputs feature labels based on the updated real-time monitoring data. In other words, data changes do not only lead to a simple replacement of report content, but also trigger the system to re-evaluate the current data's category, correlation, and time trend to determine whether the data characteristics have undergone substantial changes.

[0071] When the change in the real-time monitoring data corresponding to the re-output feature label exceeds half of the current value of the corresponding real-time monitoring data before the update, it is determined that the feature label has changed significantly. The report type and display format determination, personalized report template generation, and report update are then re-executed to achieve adaptive restructuring of the report structure. For example, in petrochemical production, if a parameter of a certain equipment suddenly changes from a stable state to a continuous upward trend, the originally recommended scatter plot will be automatically replaced with a line chart that better reflects the trend change, and the report template will be updated accordingly to ensure that the report structure always matches the data characteristics. This determination mechanism can distinguish between general data updates and significant change scenarios requiring report structure restructuring.

[0072] Adaptively adjust data processing and report generation strategies based on changes in data volume and user needs, including: To determine data size, a data size calculation function is defined within the system, and a preset threshold is set as the basis for switching between sequential and parallel processing. When the data size is less than the preset threshold, sequential processing is used to reduce the overhead of parallel scheduling; when the data size is greater than the preset threshold, parallel processing is used to fully utilize multi-core computing resources and improve data processing efficiency. To address changes in user needs, a user requirement analysis module analyzes user configuration requirements regarding report content focus and display format, and dynamically adjusts the report generation process and report display focus accordingly, enabling the system to quickly adapt to changes in business needs and report requirements.

[0073] Specifically, such as Figure 5 As shown, the optimization algorithm design scheme is as follows: (1) Define a function in the program to calculate the data size. If the data is stored in a file, set it to 10MB; if it is data in a database, set it to 10,000 records. Set a threshold for the data size. For an ordinary computer, when the number of data records is less than 1,000, it is considered small-scale data; when it is greater than 1,000, it is considered large-scale data.

[0074] (2) In the report generation system, establish a user requirement parsing module. If user requirements exist in the form of a configuration file or user input, a parsing function can be written to obtain key information to adapt to user requirements.

[0075] (3) In the data scale judgment module, the data scale threshold can be dynamically adjusted according to the real-time performance of the system. If the current CPU and memory utilization of the system is low, the threshold for large-scale data can be appropriately reduced, and more parallel processing mode can be adopted to improve overall efficiency. This method is implemented by periodically monitoring the system resource usage and adjusting the threshold according to the statistical data of resource usage.

[0076] In this implementation, a clear trigger boundary is established between regular data updates and report structure reconstruction by using a quantified threshold for significant changes. This avoids the waste of resources caused by frequent reconstruction and prevents the problem of failing to update the report structure in time when data characteristics have undergone fundamental changes. The processing strategy is dynamically adjusted according to changes in data scale and user needs, further ensuring the overall operating efficiency of the system under different data scales and demand change scenarios.

[0077] This invention also proposes a petrochemical intelligent report generation system based on a professional model, including a data acquisition module, a data storage module, a data preprocessing module, a model training module, a feature recognition module, a report recommendation module, a personalized report generation module, a report update module, and an adaptive optimization module.

[0078] The data acquisition module is used to acquire real-time monitoring data from the production equipment monitoring system of petrochemical enterprises. It is the starting point of the system data flow and is responsible for introducing real-time status information from the production site into the system. The data storage module is connected to the data acquisition module to receive and store real-time monitoring data, providing a unified data access basis for all other modules in the system that need to access raw or processed data. The data preprocessing module is connected to the data storage module and is used to preprocess real-time monitoring data to unify the data format, detect outliers, fill in missing values, and build a petrochemical data set, thereby eliminating the interference of noise and incompleteness in the original data on subsequent modules. The model training module is connected to the data preprocessing module. It is used to use a long short-term memory neural network as a time series feature recognition model, and to pre-train the time series feature recognition model using a petrochemical dataset, outputting a pre-trained model with petrochemical data feature recognition capabilities. The feature recognition module is connected to the data preprocessing module and the model training module respectively. It is used to input real-time data from the petrochemical dataset into the pre-trained time-series feature recognition model and output feature labels that represent data categories, data correlations and time trends. It is the core functional module of the system's intelligence.

[0079] After feature recognition is completed, the report recommendation module determines the report type and display format corresponding to the feature tag from the preset report type recommendation rules based on the feature tag, thus completing the intelligent mapping from data features to report presentation format. The personalized report generation module is connected to the report recommendation module to obtain the target user's role information and report requirement parameters, and generate personalized report templates based on report type, display format, role information and report requirement parameters; The report update module is connected to the data storage module, feature recognition module, report recommendation module, and personalized report generation module. It is used to obtain the latest data corresponding to the personalized report template from the data storage module and fill the personalized report template with the latest data to generate petrochemical reports. It is also used to monitor data changes in the data storage module and update petrochemical reports when data changes are detected. In addition, it is used to re-execute the report type and display format determination, personalized report template generation, and report update when feature labels change significantly. The adaptive optimization module is connected to the data preprocessing module and the report update module. It is used to adaptively adjust the data processing strategy and report generation strategy according to changes in data scale and user needs, so as to ensure the operating efficiency of the entire system under different load and demand scenarios.

[0080] In this implementation, a modular architecture is used to achieve full functional coverage of the intelligent report generation system. Each module has a clear responsibility and a clear connection relationship, which not only ensures the integrity of the system functions, but also provides a good architectural foundation for the independent upgrade and expansion of subsequent modules.

[0081] In the above embodiments, preferably, the data acquisition module includes a sensor interface and a PLC interface, and the data storage module includes a protocol transmission unit and a distributed database; The sensor interface and PLC interface are used to collect temperature, pressure, flow and level data from production equipment, thereby connecting key process monitoring data from the petrochemical production site to the system; the PLC interface is used to establish a connection with the programmable logic controller. After the PLC system aggregates and processes the signals from multiple sensors, it transmits the collected equipment operation data to the system through the PLC interface.

[0082] The protocol transmission unit transmits real-time monitoring data to the distributed database via the Modbus industrial network protocol, ensuring reliable transmission and unified storage of the collected data. The Modbus protocol guarantees the real-time performance and reliability of data transmission in industrial environments. The distributed database, as the core storage component of the data storage module, possesses efficient storage and rapid retrieval capabilities for handling large-scale real-time data from petrochemical enterprises. It can be implemented using distributed storage systems such as Hadoop HDFS, supporting concurrent writing of multiple data streams and high-frequency reading requirements from subsequent modules.

[0083] The data preprocessing module includes a standardization unit, an outlier detection unit, and a missing value completion unit. The standardization unit performs Z-score standardization on various variables in the real-time monitoring data to eliminate dimensional differences and provide a unified data foundation for subsequent detection and completion. The outlier detection unit uses the DBSCAN algorithm to detect outliers, performs density clustering analysis on the standardized data, and marks and deletes data points identified as noise. The missing value completion unit uses a multiple linear regression model and least squares method to estimate regression coefficients to predict and complete missing values ​​in the detected data, outputting a complete petrochemical dataset.

[0084] In this implementation, by further refining the internal units of the data acquisition module, data storage module, and data preprocessing module, the system can form a clear functional division of labor in hardware interface access, protocol transmission, distributed storage, standardization, outlier detection, and missing value completion, thereby improving the completeness and executability of the system implementation.

[0085] In the above embodiments, preferably, the personalized report generation module includes a role recognition unit and a template generation unit, the report update module includes a report generation unit, a data monitoring unit and a reconstruction update unit, and the adaptive optimization module includes a data scale determination unit and a requirement analysis unit. The role recognition unit is used to perform operator role recognition or engineer role recognition for target users, determine the role type of the user based on the user's identity information, and output the role classification result.

[0086] The template generation unit is used to generate differentiated report templates based on role identification results, report type, display format, and user requirement parameters: for operator roles, it generates real-time monitoring report templates, which include real-time data monitoring information, alarm information, and real-time update binding relationships; for engineer roles, it generates trend analysis report templates, which include historical trend data, analysis tools, and data export and sharing functions.

[0087] The report generation unit is used to retrieve the latest data corresponding to the personalized report template from the data storage module and populate the personalized report template with the latest data to generate a complete petrochemical report that can be viewed by users for the first time.

[0088] The data monitoring unit is used to continuously track data changes in the data storage module and uses a real-time data monitoring mechanism to detect updates to the data source. Once a data change is detected, the report content is updated, and the latest data is re-populated into the existing report template to ensure the real-time nature of the report content.

[0089] The Reconstruction and Update Unit is used to re-execute the determination of report type and display format, generation of personalized report templates, and report updates when feature labels change significantly, so that reports can automatically adapt and adjust their presentation when data characteristics change fundamentally.

[0090] The data size determination unit is used to dynamically evaluate the current data size to be processed. It uses a preset data size threshold as the judgment benchmark. When the data size is lower than the threshold, it issues a sequential processing instruction to the data preprocessing module. When the data size exceeds the threshold, it switches to a parallel processing instruction. At the same time, by periodically collecting resource indicators such as system CPU utilization and memory utilization, it dynamically adjusts the threshold according to the current system resource usage to achieve a dynamic balance between resource utilization and processing efficiency.

[0091] The requirement parsing unit is used to parse the report requirement parameters submitted by users in the form of configuration files or interactive input, extract the user's specific requirements for the report content focus, display format and data filtering conditions, and output the parsing results to the report update module to drive the dynamic adjustment of the report generation process and the report display focus, so that the system can quickly respond to the report requirement updates caused by business changes.

[0092] In this implementation, by refining the three modules at the unit level, the role recognition and template generation are decoupled, the three types of report operations—initial generation, regular updates, and structural reconstruction—are grouped into independent units, and the data scale determination and requirement analysis functions are separated. This makes the report customization, generation, updating, and optimization in the system backend a set of functional units with clear responsibilities, improving the maintainability of the system and laying a good foundation for the independent optimization and functional expansion of each unit.

[0093] According to the petrochemical intelligent report generation method and system based on a professional model disclosed in the above embodiments, taking the generation of equipment operation status monitoring reports from the operator's perspective as an example, the process includes: Data on equipment operation status was collected from the production equipment monitoring system of a petrochemical enterprise. In one experiment, the collected temperature data was 350℃, the pressure data was 15MPa, and the flow rate data was 200m³ / h. 3 / h, liquid level data is 1.5m.

[0094] Data preprocessing: The collected data is cleaned, standardized, and formatted. Temperature, pressure, flow rate, and liquid level data are standardized to values ​​between 0 and 1.

[0095] Model selection and pre-training: The LSTM model was selected as the artificial intelligence model for petrochemical data, and pre-training was performed using pre-processed data.

[0096] Feature Recognition and Report Recommendation: A pre-trained model is used to identify features in petrochemical data and recommend appropriate report types and display formats. The model identifies the data as continuous and stable, with positive correlations between variables, recommending scatter plots for visualization.

[0097] Personalized customization and report generation: The report content can be customized according to the needs of different users, and report templates can be automatically generated to display clear and concise real-time data.

[0098] Real-time data and report updates: Utilizing big data technology, the latest data is acquired in real time and populated into report templates to generate complete reports. Simultaneously, a real-time data monitoring mechanism is established to continuously track data changes and re-identify data features and update reports based on these changes.

[0099] Adaptive optimization: Automatically adjusts data processing strategies based on data size and configuration to ensure efficient report generation that meets user needs. When dealing with large datasets, parallel processing is employed to improve overall efficiency.

[0100] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating intelligent petrochemical reports based on a professional model, characterized in that, include: Acquire real-time monitoring data from the petrochemical enterprise's production equipment monitoring system and transmit the real-time monitoring data to a data storage system; The real-time monitoring data is preprocessed to unify the data format, detect outliers, fill in missing values, and construct a petrochemical data set; Using a long short-term memory neural network as a time-series feature recognition model, the time-series feature recognition model is pre-trained using the petrochemical data set; The real-time data from the petrochemical dataset is input into the pre-trained time-series feature recognition model, which outputs feature labels that characterize the data category, data correlation, and time trend. Based on the feature tags, the report type and display format corresponding to the feature tags are determined from the preset report type recommendation rules; Obtain the target user's role information and report requirement parameters, and generate a personalized report template based on the report type, the display format, the role information, and the report requirement parameters; The latest data corresponding to the personalized report template is obtained from the data source of the petrochemical enterprise, and the latest data is filled into the personalized report template to generate petrochemical reports; Monitor data changes in the data source, and when data changes are detected, obtain the latest data and update the petrochemical reports; After detecting data changes, feature labels are re-output based on the updated data. When a significant change in the feature labels is detected, the report type and display format determination, personalized report template generation, and report update are re-executed. Adaptively adjust data processing and report generation strategies based on changes in data volume and user needs.

2. The method for generating intelligent petrochemical reports based on a professional model according to claim 1, characterized in that, The process of acquiring real-time monitoring data from the petrochemical enterprise's production equipment monitoring system and transmitting the real-time monitoring data to the data storage system includes: Temperature, pressure, flow, and liquid level data of production equipment are collected through sensors and PLC industrial control systems. The temperature data, pressure data, flow rate data, and liquid level data are transmitted to the data storage system via the Modbus industrial network protocol. A distributed database system is used as the data storage system to store the real-time monitoring data.

3. The method for generating intelligent petrochemical reports based on a professional model according to claim 1, characterized in that, The preprocessing includes: The real-time monitoring data is processed to unify the data format, and the various variables in the real-time monitoring data are standardized using Z-score. The DBSCAN algorithm is used to detect and remove outliers from the standardized data. For variables containing missing values, the correlation coefficient between the variables containing missing values ​​and other variables is calculated. The two variables with the largest correlation coefficients are selected to construct a multiple linear regression model. The regression coefficients of the multiple linear regression model are estimated using the least squares method. Based on the multiple linear regression model, the missing values ​​are filled in to obtain the petrochemical data set.

4. The method for generating intelligent petrochemical reports based on a professional model according to claim 1, characterized in that, The method of using a long short-term memory neural network as a time-series feature recognition model and pre-training the time-series feature recognition model using the petrochemical dataset includes: The petrochemical dataset is divided into a training set, a validation set, and a test set in a ratio of 70%:15%:15%. Labels are set for the training set, the validation set, and the test set, and the labels include data category labels, data correlation labels, and time trend labels; The long short-term memory neural network is trained based on the training set, using the Adam optimizer, and an early stopping strategy is executed based on the validation set. The trained Long Short-Term Memory (LSTM) neural network is evaluated based on the test set, and the evaluated LSM neural network is used as a pre-trained temporal feature recognition model.

5. The method for generating intelligent petrochemical reports based on a professional model according to claim 1, characterized in that, The preset report type recommendation rules include a mapping relationship between feature tags and report types and display formats. The step of determining the report type and display format corresponding to the feature tags from the preset report type recommendation rules based on the feature tags includes: When the feature label represents a continuous trend and the time trend is a stable trend or a random trend, the report type is determined to be a scatter plot and the display format is a coordinate system. When the feature label represents a continuous trend and the time trend is an upward trend, a downward trend, or a periodic trend, the report type is determined to be a line chart, and the display format is a time axis. When the feature label represents a discrete type, the report type is determined to be a bar chart, and the display format is a classification axis.

6. The method for generating intelligent petrochemical reports based on a professional model according to claim 1, characterized in that, The step of obtaining the target user's role information and report requirement parameters, and generating a personalized report template based on the report type, the display format, the role information, and the report requirement parameters, includes: Perform role classification on target users, wherein the role classification includes at least operator roles and engineer roles; Based on the role classification results, and in conjunction with the report type, the display format, and the report requirement parameters, determine the table header, column headings, data area layout, and chart positions in the report template; When the target user is an operator, the report template generates content that includes real-time data monitoring information, alarm information, and real-time data update binding relationships. When the target user is an engineer, the report template generates content that includes historical trend data, analysis tools, and data export and sharing functions.

7. The method for generating intelligent petrochemical reports based on a professional model according to claim 1, characterized in that, After detecting data changes, the feature labels are re-output based on the updated data. Furthermore, when a significant change in the feature labels is detected, the report type and display format determination, personalized report template generation, and report update are re-executed, including: The feature labels are re-output based on the updated real-time monitoring data; When the change in the real-time monitoring data corresponding to the re-output feature label exceeds one-half of the current value of the corresponding real-time monitoring data before the update, it is determined that the feature label has changed significantly, and the report type and display format determination, personalized report template generation, and report update are re-executed. The adaptive adjustment of data processing and report generation strategies based on changes in data scale and user needs includes: When the data size is less than a preset threshold, a sequential processing method is used; when the data size is greater than the preset threshold, a parallel processing method is used; and the report generation process and report display focus are adjusted based on the user needs analysis results.

8. A petrochemical intelligent report generation system based on a professional model, characterized in that, It includes a data acquisition module, a data storage module, a data preprocessing module, a model training module, a feature recognition module, a report recommendation module, a personalized report generation module, a report update module, and an adaptive optimization module; The data acquisition module is used to acquire real-time monitoring data from the petrochemical enterprise's production equipment monitoring system; The data storage module is connected to the data acquisition module and is used to receive and store the real-time monitoring data; The data preprocessing module is connected to the data storage module and is used to preprocess the real-time monitoring data to unify the data format, detect outliers, fill in missing values, and construct a petrochemical data set. The model training module is connected to the data preprocessing module and is used to use a long short-term memory neural network as a time-series feature recognition model and to pre-train the time-series feature recognition model using the petrochemical data set. The feature recognition module is connected to the data preprocessing module and the model training module respectively, and is used to input real-time data from the petrochemical dataset into the pre-trained time-series feature recognition model, and output feature labels that characterize the data category, data correlation and time trend. The report recommendation module is connected to the feature recognition module and is used to determine the report type and display format corresponding to the feature tag from the preset report type recommendation rules based on the feature tag. The personalized report generation module is connected to the report recommendation module and is used to obtain the target user's role information and report requirement parameters, and generate a personalized report template based on the report type, the display format, the role information and the report requirement parameters. The report update module is connected to the data storage module, the feature recognition module, the report recommendation module, and the personalized report generation module, respectively. It is used to obtain the latest data corresponding to the personalized report template from the data storage module and fill the personalized report template with the latest data to generate petrochemical reports. It is also used to monitor data changes in the data storage module and update the petrochemical reports when data changes are detected. Furthermore, it is used to re-execute the report type and display format determination, personalized report template generation, and report update when the feature tags change significantly. The adaptive optimization module is connected to the data preprocessing module and the report update module, and is used to adaptively adjust the data processing strategy and report generation strategy according to changes in data scale and user needs.

9. The petrochemical intelligent report generation system based on a professional model according to claim 8, characterized in that, The data acquisition module includes a sensor interface and a PLC interface, and the data storage module includes a protocol transmission unit and a distributed database. The sensor interface and the PLC interface are used to collect temperature data, pressure data, flow data and liquid level data of the production equipment; The protocol transmission unit is used to transmit the real-time monitoring data to the distributed database via the Modbus industrial network protocol. The data preprocessing module includes a standardization unit, an outlier detection unit, and a missing value completion unit. The standardization unit is used to perform Z-score standardization on various variables in the real-time monitoring data. The outlier detection unit is used to detect outliers using the DBSCAN algorithm. The missing value completion unit is used to complete missing values ​​by using a multiple linear regression model and estimating regression coefficients using the least squares method.

10. The petrochemical intelligent report generation system based on a professional model according to claim 8, characterized in that, The personalized report generation module includes a role recognition unit and a template generation unit; the report update module includes a report generation unit, a data monitoring unit, and a reconstruction and update unit; and the adaptive optimization module includes a data scale determination unit and a requirement analysis unit. The role recognition unit is used to perform operator role recognition or engineer role recognition for the target user; The template generation unit is used to generate real-time monitoring report templates or trend analysis report templates based on role recognition results, report types, display formats, and user requirement parameters. The report generation unit is used to obtain the latest data corresponding to the personalized report template from the data storage module, and fill the latest data into the personalized report template to generate a petrochemical report; The data monitoring unit is used to continuously track data changes in the data storage module and trigger report content updates when data changes are detected. The reconstruction and update unit is used to re-execute the determination of report type and display format, generation of personalized report templates, and report update when the feature labels change significantly. The data size determination unit is used to switch between sequential processing and parallel processing based on the data size. The requirement parsing unit is used to parse user requirement parameters and adjust the report generation process and report display focus based on the parsing results.

Images