Multivariable intelligent anomaly detection and early warning control method and system for flow production process

Abstract

The invention is suitable for the technical field of anomaly detection, and particularly relates to a multivariable intelligent anomaly detection and early warning control method and system for a flow production process, and the method comprises the steps: collecting the sensing data of key equipment under a normal working condition through sensing equipment which is integrated in the production process in advance, wherein the sensing data at least comprise temperature, flow and pressure, preprocessing the sensing data, and constructing a baseline model; and acquiring real-time data of the sensing equipment, inputting the real-time data into the baseline model, and outputting to obtain a deviation degree which at least comprises a normal level, an attention level, a warning level and a danger level. According to the method, the abnormal events are generated, each abnormal event can be recorded, traced and analyzed in detail, data support is provided for training and improvement of a baseline model, and meanwhile, a reliable data basis is provided for abnormal detection and intelligent early warning in the production process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of anomaly detection, and in particular to a multivariate intelligent anomaly detection and early warning control method and system for a process production process. BACKGROUND

[0002] Tobacco production is composed of multiple process production processes, and there are numerous process variables and complex coupling relationships. For example, in tobacco production, loose rehydration, cut tobacco drying, and addition of flavoring materials and the like play a decisive role in the quality of the final product. Any slight deviation in process parameters, such as instantaneous fluctuations in liquid flow, instability in the speed of drying medium (such as hot air), and gradual decline in the efficiency of heat exchange equipment, can directly affect the physical properties of cut tobacco (such as moisture content and filling value), uniform absorption of chemical components, and other core quality indicators, and thus significantly affect the smoking stability, burning characteristics, and sensory experience of the final consumer of the finished cigarette product.

[0003] Traditional anomaly monitoring methods mainly rely on setting fixed upper and lower threshold values for individual process variables to trigger alarms, or on periodic manual inspection and experience-based judgment, which has significant limitations, including delayed response, high misjudgment rate, and single treatment strategy.

[0004] Therefore, the technical problem of how to perform anomaly detection in a tobacco production process needs to be addressed. SUMMARY

[0005] The present application aims to provide a multivariate intelligent anomaly detection and early warning control method and system for a process production process to address the problem of how to perform anomaly detection in a tobacco production process as described in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] A multivariate intelligent anomaly detection and early warning control method for a process production process, the method comprising:

[0008] Collecting sensor data of key equipment under normal operating conditions using sensor devices integrated in advance in the production process, wherein the sensor data at least includes temperature, flow rate, and pressure, preprocessing the sensor data, and constructing a baseline model;

[0009] Collecting real-time data of the sensor devices and inputting them into the baseline model to output a deviation degree, wherein the deviation degree at least includes normal, attention, warning, and danger levels, and creating a knowledge base composed of historical failure cases, standard operating procedures, and emergency response rules;

[0010] Collecting environmental parameters in the production process, when the deviation degree is a warning level and a dangerous level, writing the environmental parameters and real-time data into a preset template, generating an abnormal event, traversing a knowledge base, and obtaining an emergency treatment rule;

[0011] Configuring an initial signal and a correction suggestion corresponding to the abnormal event, extracting the correction suggestion when the initial signal is included in the real-time data and the environmental parameters, and issuing the emergency treatment rule and the correction suggestion to a preset terminal.

[0012] Further, the step of preprocessing the sensing data and constructing a baseline model comprises:

[0013] Preprocessing the sensing data, wherein the preprocessing at least includes data cleaning, alignment, and uniform format;

[0014] Integrating historical sensing data, generating a training data set, training the baseline model, identifying incremental data of the training data set, and setting a training period.

[0015] Further, the step of collecting real-time data of the sensing device, inputting into the baseline model, and outputting to obtain the deviation degree comprises:

[0016] Constructing a sliding time window, determining a sliding step, configuring standard parameters of the sensing device, and calculating reconstruction errors of real-time parameters and standard parameters;

[0017] Dividing the reconstruction errors into several deviation degrees.

[0018] Further, the step of writing the environmental parameters and real-time data into a preset template, generating an abnormal event, traversing a knowledge base, and obtaining an emergency treatment rule comprises:

[0019] Creating evaluation rules corresponding to the deviation degree one by one;

[0020] When the environmental parameters and real-time data meet the evaluation rules, the corresponding deviation degree is activated, an abnormal event is generated, and the emergency treatment rule is started.

[0021] Further, the step of configuring an initial signal and a correction suggestion corresponding to the abnormal event comprises:

[0022] Storing the abnormal event in a database constructed in advance, and extracting an event summary;

[0023] Integrating all event summaries, generating an indexing mechanism, and embedding it into the database.

[0024] Further, the system comprises:

[0025] The construction module is configured to collect sensing data of key equipment under normal working conditions by using sensing devices integrated in advance in a production process, wherein the sensing data at least includes temperature, flow and pressure, pre-process the sensing data, and construct a baseline model;

[0026] The creation module is configured to collect real-time data of the sensing devices, input the real-time data into the baseline model, and output a deviation degree, wherein the deviation degree at least includes a normal level, a warning level and a danger level, and create a knowledge base composed of historical fault cases, standard operation procedures and emergency treatment rules;

[0027] The obtaining module is configured to collect environmental parameters in the production process, write the environmental parameters and the real-time data into a preset template when the deviation degree is the warning level or the danger level, generate an abnormal event, and traverse the knowledge base to obtain the emergency treatment rule;

[0028] The issuing module is configured to configure an initial signal and a correction suggestion corresponding to the abnormal event, extract the correction suggestion when the real-time data and the environmental parameters include the initial signal, and issue the emergency treatment rule and the correction suggestion to a preset terminal.

[0029] Further, the construction module comprises:

[0030] The pre-processing unit is configured to pre-process the sensing data, wherein the pre-processing at least includes data cleaning, alignment and uniform format;

[0031] The configuration unit is configured to integrate historical sensing data, generate a training data set, train the baseline model, identify incremental data of the training data set, and set a training period.

[0032] Further, the creation module comprises:

[0033] The calculation unit is configured to construct a sliding time window, determine a sliding step, configure standard parameters of the sensing devices, and calculate reconstruction errors of real-time parameters and the standard parameters;

[0034] The division unit is configured to divide the reconstruction errors into a plurality of deviation degrees.

[0035] Further, the obtaining module comprises:

[0036] The creation unit is configured to create evaluation rules corresponding to the deviation degrees;

[0037] The starting unit is configured to activate a corresponding deviation degree, generate an abnormal event, and start the emergency treatment rule when the environmental parameters and the real-time data meet the evaluation rules.

[0038] Further, the issuing module comprises:

[0039] an extraction unit configured to store the abnormal event into a database that is pre-constructed, and extract an event summary;

[0040] an embedding unit configured to integrate all the event summaries, generate an indexing mechanism, and embed into the database.

[0041] Compared with the prior art, the present application has the following beneficial effects:

[0042] By constructing and utilizing the baseline model, the trend of the production process from the normal state to the abnormal state can be perceived earlier by monitoring the slight deviation between multiple changes before a single variable reaches the alarm limit, early warning of potential problems is realized, by determining the deviation degree, not only the early warning can be sent in time, but also the targeted emergency disposal rules can be provided for the operators according to the detected abnormal level and characteristics, combined with the knowledge base, by generating the correction suggestions, the early warning and timely disposal of the production abnormality are realized, the adverse effects of the production abnormality on the final product quality are maximally avoided or mitigated, the safety accidents are prevented, by generating the abnormal event, the detailed record, traceability and analysis of each abnormal event are realized, the data support for the training and improvement of the baseline model is provided, and at the same time, the reliable data basis for the abnormal detection and intelligent early warning in the production process is provided. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 a flowchart of the multivariate intelligent abnormality detection and early warning control method for the process production process provided by the embodiment of the present application;

[0044] Figure 2 a first sub-flowchart of the multivariate intelligent abnormality detection and early warning control method for the process production process provided by the embodiment of the present application;

[0045] Figure 3 a second sub-flowchart of the multivariate intelligent abnormality detection and early warning control method for the process production process provided by the embodiment of the present application;

[0046] Figure 4 a third sub-flowchart of the multivariate intelligent abnormality detection and early warning control method for the process production process provided by the embodiment of the present application;

[0047] Figure 5 a fourth sub-flowchart of the multivariate intelligent abnormality detection and early warning control method for the process production process provided by the embodiment of the present application;

[0048] Figure 6 a composition block diagram of the multivariate intelligent abnormality detection and early warning control system for the process production process provided by the embodiment of the present application;

[0049] Figure 7 The component block diagram of the construction module in the multivariate intelligent anomaly detection and early warning control system for process production process provided by the embodiment of the present application is shown in the figure.

[0050] Figure 8 The component block diagram of the creation module in the multivariate intelligent anomaly detection and early warning control system for process production process provided by the embodiment of the present application is shown in the figure.

[0051] Figure 9 The component block diagram of the obtaining module in the multivariate intelligent anomaly detection and early warning control system for process production process provided by the embodiment of the present application is shown in the figure.

[0052] Figure 10 The component block diagram of the issuing module in the multivariate intelligent anomaly detection and early warning control system for process production process provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0054] In embodiment 1, Figure 1 The method for realizing the multivariate intelligent anomaly detection and early warning control of process production process provided by the embodiment of the present application is shown in the figure, which is described in detail as follows:

[0055] S100: Collect the sensing data of the key equipment under normal working conditions by using the sensing device integrated in the production process in advance, wherein the sensing data at least includes temperature, flow and pressure, pre-process the sensing data, and construct a baseline model.

[0056] Real-time collection of sensing data of each key section and key equipment in the process production process is realized by using the bottom-layer automation system (such as programmable logic controller, distributed control system) deployed in the production site in advance or various sensors directly installed on the key equipment; for example, in tobacco primary processing, the key sections include loose rehydration section, leaf conditioning and feeding section, cutting section, drying section and flavoring section, etc., and the sensing data includes but is not limited to temperature (such as material temperature, medium temperature), humidity (such as environment humidity, material moisture), pressure (such as steam pressure, air pressure), flow (such as liquid flow, air volume), speed (such as conveyor belt speed, motor speed), valve opening, material composition or concentration, equipment running state indication (such as current, vibration) and the like.

[0057] The sensor device is connected to the programmable logic controller or the distributed control system by using one or more industrial standard communication protocols to realize efficient and reliable acquisition of sensor data, and to ensure the integrity and real-time performance of the sensor data.

[0058] The acquired sensor data is preprocessed, including data cleaning (such as identifying and processing outliers or outliers, filling missing values, and removing noise interference), data transformation (such as data smoothing, data standardization or normalization to eliminate the influence of different variable dimensions), data synchronization (to ensure that different variable data are aligned in time), and necessary feature engineering (such as feature extraction, feature selection, and feature construction to highlight key information related to process state); a baseline model is constructed by using a machine learning algorithm, which can accurately describe and represent the dynamic behavior and correlation between the sensor data in the normal operation state of the production process.

[0059] S200: Collect real-time data of the sensor device and input it into the baseline model to output the deviation degree, which includes at least normal level, attention level, warning level and danger level, and create a knowledge base composed of historical fault cases, standard operating procedures and emergency disposal rules.

[0060] The real-time collected and preprocessed sensor data is input into the baseline model to determine whether a statistically significant abnormality has occurred in the current production process by calculating the deviation of the real-time data from the sensor data under normal operating conditions; in the baseline model, the root mean square difference between the real-time value of the sensor data at the current time and the standard parameter is calculated, and the difference is divided into several intervals, each interval corresponding to a deviation degree, which includes at least normal level, attention level, warning level and danger level; if the deviation degree is normal level, the root mean square difference is within the normal interval, and the corresponding set values are set for attention level, warning level and danger level, and when the root mean square difference is greater than the corresponding set value, the corresponding early warning mode is started.

[0061] For example, when the root mean square difference is greater than the set value corresponding to the attention level, but less than the set value corresponding to the warning level, and the duration is less than 3 minutes, the yellow early warning is started, and the prompt information is: "The drying process fluctuates slightly, please pay attention to the outlet moisture content and related parameters."

[0062] When the root mean square difference is greater than the set value corresponding to the warning level, but less than the set value corresponding to the danger level, or the yellow early warning state lasts more than 5 minutes, the orange early warning is started, and the prompt information is: "The drying process parameter deviates from the warning! The outlet moisture content has a high or low trend, and the possible cause analysis is: poor moisture removal or insufficient hot air.

[0063] When the root mean square difference value is greater than the set value corresponding to the dangerous level, or the real-time measurement value of the outlet tobacco moisture content has exceeded its corresponding normal fluctuation range, or the sensor data presents a continuous rapid deterioration trend (for example, judged by a short-term trend prediction model), or the orange early warning state lasts for more than 10 minutes, a red early warning is started, and the prompt information is: "The drying process is seriously abnormal, the outlet moisture content has exceeded the limit, take immediate action!", the prompt information is displayed in different colors on the visual monitoring interface of the drying machine operation station, accompanied by different levels of sound and light alarm, and a notification is pushed to the mobile terminal of the team leader and process engineer.

[0064] Integrate all historical failure cases, standard operating procedures and emergency disposal rules, etc. to build a knowledge base. Through the construction of the knowledge base, the response strategy can be quickly determined to promote the accumulation and continuous improvement of production knowledge.

[0065] S300: Collecting environmental parameters in the production process, when the deviation degree is warning level and dangerous level, writing the environmental parameters and real-time data into a preset template to generate an abnormal event, traversing the knowledge base to obtain an emergency disposal rule.

[0066] Collecting environmental parameters in the production process, wherein the environmental parameters include: occurrence time, duration, root mean square difference value and actual measures taken by the operator of the warning level and dangerous level, writing the environmental parameters and real-time data into a preset template to generate an abnormal event, wherein the preset template is prepared by professionals, and using the abnormal event, traversing the emergency disposal rule from the knowledge base.

[0067] For example, when an orange early warning occurs, find out the item with the largest difference between real-time data and standard parameters of sensor data, and judge whether the moisture exhaust door opening degree abnormality and the outlet moisture content rising trend present a correlation, wherein the correlation includes: positive correlation and negative correlation, if so, adjust in time, write the corresponding environmental parameters and real-time data into the template to generate an abnormal event, and set the corresponding emergency disposal rule, wherein the emergency disposal rule is: "Suggest checking whether the moisture exhaust door actuator is stuck or the sensor feedback is accurate, try to manually adjust the moisture exhaust door opening degree by 5%, and closely observe the outlet moisture content change."

[0068] When a red early warning occurs, if the item with the largest difference between the standard parameters is the outlet moisture content, the emergency disposal rule is determined as: "1. Immediately reduce the net belt speed by 10%; 2. Check and ensure that the hot air inlet temperature is at the set value; 3. If there is no improvement within 1 minute, consider suspending feeding and conducting a comprehensive investigation."

[0069] S400: configuring initial signals and correction suggestions corresponding to abnormal events, extracting the correction suggestions when the initial signals are contained in the real-time data and environmental parameters, and issuing the emergency treatment rules and correction suggestions to preset terminals.

[0070] Using historical data, abnormal events are determined, and corresponding initial signals are determined. According to the current warning level, the identified abnormal characteristics (such as the variable combination that contributes most to the abnormality, the preliminary diagnosis of the possible cause of the abnormality, etc.), the emergency treatment rules, the preventive measures (such as suggesting to check the wear of a specific equipment component) or the correction suggestions (such as suggesting to adjust the set value of a certain process parameter or change the addition rate of a certain material) are retrieved from the knowledge base, and sent to the mobile terminal of the team leader.

[0071] In embodiment 2, Figure 2 The implementation process of the multivariate intelligent abnormality detection and early warning control method for the process production process provided by the embodiment of the application is shown. The steps of preprocessing the sensor data and constructing the baseline model are described in detail as follows:

[0072] S101: preprocessing the sensor data, wherein the preprocessing at least includes data cleaning, alignment and uniform format.

[0073] The collected sensor data is preprocessed, including but not limited to data cleaning (such as filtering obvious noise, handling missing values caused by communication interruption), data alignment, data normalization or uniform format, etc. to form time series data suitable for input to the baseline model.

[0074] S102: integrating historical sensor data, generating a training data set, training the baseline model, identifying the incremental data of the training data set, and setting a training period.

[0075] Using the accumulated historical sensor data, a training data set is formed, and the constructed baseline model is trained. According to the set training period (for example, every month or every quarter), the latest normal production data is used for retraining or incremental updating to adapt to the slow drift of process conditions or the natural aging of equipment performance.

[0076] In embodiment 3, Figure 3 The implementation process of the multivariate intelligent abnormality detection and early warning control method for the process production process provided by the embodiment of the application is shown. The steps of collecting real-time data of the sensor device, inputting the real-time data into the baseline model, and outputting the deviation degree are described in detail as follows:

[0077] S201: constructing a sliding time window, determining a sliding step, configuring standard parameters of the sensor device, and calculating reconstruction errors of real-time parameters and standard parameters.

[0078] Two time points are selected, and a sliding time window is constructed; for example, the sensor data in the last 60 seconds is analyzed every 5 seconds, the time window is 60 seconds, the sliding step is 5 seconds, the standard parameters of the sensor device are configured, the root mean square difference between the real-time parameters and the standard parameters is calculated, and the root mean square difference is defined as the reconstruction error.

[0079] S202: The reconstruction error is divided into several deviation degrees.

[0080] The root mean square difference is divided into several intervals, and each interval corresponds to a deviation degree, wherein the deviation degree can be divided into high, medium and low.

[0081] In embodiment 4, Figure 4 The implementation process of the multivariate intelligent anomaly detection and early warning control method for a process production process provided by the embodiment of the application is shown, and the steps of writing the environmental parameters and real-time data into a preset template, generating an abnormal event, traversing a knowledge base, and obtaining an emergency disposal rule are described in detail as follows:

[0082] S301: An evaluation rule corresponding to each deviation degree is created.

[0083] A corresponding set value is set for each deviation degree, wherein the set value is selected from the sensor data, and each deviation degree corresponds to an evaluation rule, wherein the evaluation rule can be: when the root mean square difference is greater than the set value corresponding to the attention level, but less than the set value corresponding to the warning level, and the duration is less than 3 minutes, a yellow early warning is started.

[0084] S302: When the environmental parameters and real-time data meet the evaluation rule, the corresponding deviation degree is activated, an abnormal event is generated, and an emergency disposal rule is started.

[0085] When the environmental parameters and real-time data meet the evaluation rule, the corresponding deviation degree is activated, the environmental parameters and real-time data are written into a preset template, an abnormal event is generated, and an emergency disposal rule is started, wherein the emergency disposal rule includes sending prompt information.

[0086] In embodiment 5, Figure 5 The implementation process of the multivariate intelligent anomaly detection and early warning control method for a process production process provided by the embodiment of the application is shown, and the steps of configuring an initial signal and a correction suggestion corresponding to each abnormal event are described in detail as follows:

[0087] S401: The abnormal event is stored in a pre-constructed database, and an event summary is extracted.

[0088] The abnormal event is stored into a database, and a keyword is extracted from each abnormal event to generate an event summary.

[0089] S402: all event summaries are integrated to generate an index mechanism and embed into the database.

[0090] The index mechanism is constructed by using a multi-dimensional keyword combination (such as "time + location + event type") and an inverted index mode, so that the abnormal event can be quickly retrieved on demand; the index mechanism is embedded into the database structure in a modular manner and is associated with the abnormal event as an independent index table or mapping table, to ensure that the corresponding abnormal event can be quickly located by a keyword or condition during database query.

[0091] Figure 6 A component structure block diagram of a multivariate intelligent abnormality detection and early warning control system for a process production process is shown, and the multivariate intelligent abnormality detection and early warning control system 1 for the process production process comprises:

[0092] A construction module 11 is configured to collect sensing data of a key device under a normal working condition by using a sensing device integrated in advance in a production process, wherein the sensing data at least comprises temperature, flow and pressure, the sensing data is preprocessed, and a baseline model is constructed;

[0093] A creation module 12 is configured to collect real-time data of the sensing device, input the real-time data into the baseline model, and output a deviation degree, wherein the deviation degree at least comprises a normal level, a warning level and a danger level, and a knowledge base composed of historical fault cases, standard operation procedures and emergency disposal rules is created;

[0094] A obtaining module 13 is configured to collect environmental parameters in the production process, and when the deviation degree is the warning level and the danger level, the environmental parameters and the real-time data are written into a preset template to generate an abnormal event, and the emergency disposal rules are obtained by traversing the knowledge base;

[0095] An issuing module 14 is configured to configure an initial signal and a correction suggestion corresponding to the abnormal event, and when the real-time data and the environmental parameters contain the initial signal, the correction suggestion is extracted, and the emergency disposal rules and the correction suggestion are issued to a preset terminal.

[0096] Figure 7 A component structure block diagram of a multivariate intelligent abnormality detection and early warning control system for a process production process is shown, and the construction module 11 comprises:

[0097] A preprocessing unit 111 is configured to preprocess the sensing data, wherein the preprocessing at least comprises data cleaning, alignment and uniform format;

[0098] The configuration unit 112 is configured to integrate historical sensing data, generate a training data set, train the baseline model, identify incremental data of the training data set, and set a training period.

[0099] Figure 8 The component structure block diagram of the multivariate intelligent anomaly detection and early warning control system for the process production process provided by the embodiment of the application is shown, and the creating module 12 comprises:

[0100] The calculation unit 121 is configured to construct a sliding time window, determine a sliding step, configure standard parameters of a sensing device, and calculate reconstruction errors of real-time parameters and the standard parameters.

[0101] The division unit 122 is configured to divide the reconstruction errors into a plurality of deviation degrees.

[0102] Figure 9 The component structure block diagram of the multivariate intelligent anomaly detection and early warning control system for the process production process provided by the embodiment of the application is shown, and the obtaining module 13 comprises:

[0103] The creating unit 131 is configured to create evaluation rules corresponding to the deviation degrees one by one.

[0104] The starting unit 132 is configured to activate the corresponding deviation degree, generate an abnormal event, and start an emergency disposal rule when the environmental parameters and the real-time data satisfy the evaluation rules.

[0105] Figure 10 The component structure block diagram of the multivariate intelligent anomaly detection and early warning control system for the process production process provided by the embodiment of the application is shown, and the issuing module 14 comprises:

[0106] The abstracting unit 141 is configured to store the abnormal event into a database that has been constructed in advance, and extract an event abstract.

[0107] The embedding unit 142 is configured to integrate all the event abstracts, generate an indexing mechanism, and embed the indexing mechanism into the database.

[0108] The constructing module 11 is mainly configured to complete the step S100, the creating module 12 is mainly configured to complete the step S200, the obtaining module 13 is mainly configured to complete the step S300, and the issuing module 14 is mainly configured to complete the step S400.

[0109] The preprocessing unit 111 is mainly configured to complete the step S101, and the configuration unit 112 is mainly configured to complete the step S102.

[0110] The computing unit 121 is mainly configured to complete the step S201, and the dividing unit 122 is mainly configured to complete the step S202.

[0111] The creating unit 131 is mainly configured to complete the step S301, and the starting unit 132 is mainly configured to complete the step S302.

[0112] The abstracting unit 141 is mainly configured to complete the step S401, and the embedding unit 142 is mainly configured to complete the step S402.

[0113] The technical features of the above-described embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above-described embodiments are not described, however, as long as the combinations of the technical features do not exist in contradiction, they should be considered as falling within the scope of the present disclosure.

[0114] The above-described embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, however, it should not be understood as a limitation on the patent scope of the present application. It should be pointed out that, for those skilled in the art, some modifications and improvements can be made without departing from the concept of the present application, and these should be considered as falling within the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

[0115] The above-described embodiments are only the preferred embodiments of the present application, and are not used to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be considered as falling within the protection scope of the present application.

Claims

1. A multivariate intelligent anomaly detection and early warning control method for a process production process, characterized in that: The method comprises: Using sensor equipment pre-integrated in the production process, collecting sensor data of key equipment under normal operating conditions, wherein the sensor data includes at least temperature, flow rate and pressure, pre-processing the sensor data and building a baseline model; Collect real-time data from sensor devices, input it into a baseline model, and output a degree of deviation, where the degree of deviation includes at least normal, caution, warning, and danger levels. This creates a knowledge base consisting of historical failure cases, standard operating procedures, and emergency response rules. Collect environmental parameters during the production process. When the deviation reaches the warning level or danger level, write the environmental parameters and real-time data into a preset template, generate an abnormal event, traverse the knowledge base, and obtain emergency response rules; An initial signal and correction suggestion corresponding to an abnormal event are configured. When the real-time data and environmental parameters contain an initial signal, the correction suggestion is extracted, and the emergency handling rules and correction suggestion are sent to a preset terminal.

2. The multivariate intelligent anomaly detection and early warning control method for process production according to claim 1 is characterized in that: The steps of preprocessing the sensor data and building a baseline model include: Preprocessing the sensor data, wherein the preprocessing at least includes: data cleaning, alignment and formatting; Integrate historical sensor data to generate a training data set, train the baseline model, identify incremental data of the training data set, and set a training cycle.

3. The multivariate intelligent anomaly detection and early warning control method for a process production process according to claim 1 is characterized in that: The steps of collecting real-time data from the sensing device, inputting it into the baseline model, and outputting the degree of deviation include: Construct a sliding time window, determine the sliding step size, configure the standard parameters of the sensor equipment, and calculate the reconstruction error between the real-time parameters and the standard parameters; The reconstruction error is divided into several deviation degrees.

4. The multivariate intelligent anomaly detection and early warning control method for a process production process according to claim 3 is characterized in that: The steps of writing the environmental parameters and real-time data into a preset template, generating an abnormal event, traversing the knowledge base, and obtaining emergency response rules include: Create evaluation rules that correspond to the degree of deviation; When the environmental parameters and real-time data meet the evaluation rules, the corresponding deviation degree is activated, an abnormal event is generated, and the emergency response rules are started.

5. The multivariate intelligent anomaly detection and early warning control method for a process production process according to claim 1, characterized in that: The step of configuring the initial signal and correction suggestion corresponding to the abnormal event includes: Storing the abnormal events in a pre-built database and extracting event summaries; All event summaries are integrated, an indexing mechanism is generated, and embedded into the database.

6. A multivariable intelligent anomaly detection and early warning control system for process production, characterized in that: The system comprises: A construction module is configured to utilize sensor equipment pre-integrated in the production process to collect sensor data of key equipment under normal operating conditions, wherein the sensor data includes at least temperature, flow rate, and pressure, pre-process the sensor data, and construct a baseline model; Create a module for collecting real-time data from sensor devices, inputting it into a baseline model, and outputting a degree of deviation, where the degree of deviation includes at least normal, caution, warning, and danger levels. Create a knowledge base consisting of historical failure cases, standard operating procedures, and emergency response rules. A module is used to collect environmental parameters during the production process. When the deviation level reaches the warning level or the danger level, the environmental parameters and real-time data are written into a preset template, an abnormal event is generated, and the knowledge base is traversed to obtain emergency response rules. The sending module is used to configure the initial signal and correction suggestion corresponding to the abnormal event one by one. When the initial signal is included in the real-time data and environmental parameters, the correction suggestion is extracted and the emergency response rules and correction suggestions are sent to the preset terminal.

7. The multivariable intelligent anomaly detection and early warning control system for a process production process according to claim 6, characterized in that: The building blocks include: A preprocessing unit, configured to preprocess the sensor data, wherein the preprocessing includes at least: data cleaning, alignment, and format unification; The configuration unit is used to integrate historical sensor data, generate a training data set, train the baseline model, identify incremental data of the training data set, and set a training cycle.

8. The multivariable intelligent anomaly detection and early warning control system for a process production process according to claim 6, characterized in that: The creation module includes: The calculation unit is used to construct a sliding time window, determine the sliding step size, configure the standard parameters of the sensor equipment, and calculate the reconstruction error between the real-time parameters and the standard parameters; The division unit is used to divide the reconstruction error into several deviation degrees.

9. The multivariable intelligent anomaly detection and early warning control system for a process production process according to claim 8, characterized in that: The obtaining module includes: A creation unit for creating evaluation rules corresponding to the degree of deviation; The starting unit is used to activate the corresponding deviation degree, generate an abnormal event, and start the emergency handling rules when the environmental parameters and real-time data meet the evaluation rules.

10. The multivariable intelligent anomaly detection and early warning control system for a process production process according to claim 6, characterized in that: The sending module includes: An extraction unit, configured to store the abnormal event in a pre-built database and extract an event summary; The embedding unit is used to integrate all event summaries, generate an index mechanism, and embed them into the database.

Images