Early warning information processing method and device applied to process production

Abstract

The application relates to the technical field of information processing, and discloses a pre-warning information processing method and device applied to process production, which comprises the following steps: collecting real-time data corresponding to process production and constructing a target data set; performing correlation analysis operation on all target data contained in the target data set through a target intelligent model to generate correlation analysis information; performing data classification operation on all target data contained in the target data set according to the correlation analysis information to obtain at least one data category; determining the pre-warning priority corresponding to each data category, and performing information processing operation on the information to be processed based on each pre-warning priority and the target intelligent model set in advance to obtain target pre-warning information. It can be seen that the application can intelligently process pre-warning information, and is favorable for improving the accuracy and reliability of pre-warning information processing.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a method and apparatus for processing early warning information applied to industrial production. Background Technology

[0002] In modern manufacturing processes, early warning systems are a crucial component for ensuring production safety and efficiency. Traditional early warning systems mostly trigger warnings based on preset thresholds; when monitored data exceeds or falls below these thresholds, the system generates an early warning. However, due to the complexity and variability of production processes, this simple threshold-based mechanism is prone to generating a large number of false alarms. Faced with a deluge of warning information, operators often struggle to quickly and accurately identify the truly critical warnings, resulting in key warning information being buried under a sea of ​​false alarms. This leads to technical problems of low accuracy and reliability in warning information processing.

[0003] Therefore, improving the accuracy and reliability of early warning information processing during the production process is of paramount importance. Summary of the Invention

[0004] This invention provides a method and apparatus for processing early warning information in process production, which can intelligently process early warning information and improve the accuracy and reliability of early warning information processing.

[0005] The first aspect of this invention discloses a method for processing early warning information applied to process production, the method comprising:

[0006] Collect real-time data corresponding to the production process, and construct a target dataset based on all the real-time data.

[0007] By using a pre-set target intelligent model, an association analysis operation is performed on all target data contained in the target dataset to generate association analysis information corresponding to the target dataset. The association analysis information includes association rule information and frequent itemset information for all target data contained in the target dataset.

[0008] Based on the association analysis information, a data classification operation is performed on all the target data contained in the target dataset to obtain at least one data category, wherein each data category includes at least one target data;

[0009] Determine the warning priority corresponding to each data category, and based on each warning priority and the pre-set target intelligent model, perform information processing operations on the information to be processed to obtain target warning information.

[0010] As an optional implementation, in the first aspect of the present invention, before performing correlation analysis on all target data contained in the target dataset using a pre-set target intelligent model to generate correlation analysis information corresponding to the target dataset, the method further includes:

[0011] Collect historical process data and equipment topology information, construct a multimodal pre-training dataset based on the historical process data and equipment topology information, and perform pre-training operations on the standby intelligent model based on the multimodal pre-training dataset to obtain the pre-trained intelligent model;

[0012] Based on a pre-designed bidirectional architecture, the temporal dependencies corresponding to the historical process data are determined, and the equipment topology association features corresponding to the equipment topology information are determined based on the temporal dependencies; wherein, the pre-designed bidirectional architecture includes a bidirectional Transformer architecture;

[0013] Based on the device topology association features and the pre-trained intelligent model, a target intelligent model is generated.

[0014] As an optional implementation, in the first aspect of the present invention, after the acquisition process generates corresponding real-time data, the method further includes:

[0015] Perform a data detection operation on the real-time data to obtain a data detection result, and determine whether the data detection result is used to indicate that there is abnormal data in all the real-time data;

[0016] When it is determined that the data detection result indicates the presence of abnormal data in all the real-time data, anomaly processing operation is performed on the abnormal data based on the pre-set forward propagation method to obtain the target processed data;

[0017] Perform batch processing on the target data to obtain batch processing results, and store the batch processing results in a predetermined target database to establish a data tracking relationship. Based on the data tracking relationship, update all the real-time data and trigger the operation of constructing a target dataset based on all the real-time data.

[0018] The data batch processing operation includes one or more of data alignment operations and data feature extraction operations; the predetermined target database includes one or more of time-series databases and relational databases.

[0019] As an optional implementation, in a first aspect of the invention, performing a data classification operation on all the target data contained in the target dataset to obtain at least one data category includes:

[0020] Based on pre-set association rules and all target data contained in the target dataset, generate data logical relationship information for all target data;

[0021] Extract data description information of all the target data, and based on the data description information and the data logical relationship information, determine the data mode corresponding to each target data and the mode weight corresponding to each data mode through a pre-determined fusion classification model;

[0022] Based on the data modality corresponding to each target data and the modality weight corresponding to each data modality, a data classification operation is performed on all the target data to obtain at least one data category.

[0023] As an optional implementation, in a first aspect of the present invention, the method further includes:

[0024] Obtain the warning feedback information corresponding to the target warning information, and construct a warning feedback state set based on the warning feedback information, wherein the warning feedback state set includes one or more of the following: equipment operating condition warning feedback state, process warning feedback state, and warning classification strategy feedback state.

[0025] The key alarm information in the target early warning information is identified, and based on the key alarm information and the early warning feedback state set, an experience replay pool corresponding to the target early warning information is constructed. Based on the experience replay pool, a model optimization operation is performed on the pre-set target intelligent model to obtain the optimized target intelligent model, and model optimization comparison information corresponding to the target intelligent model is generated.

[0026] As an optional implementation, in a first aspect of the present invention, the method further includes:

[0027] Based on the warning feedback information, determine the warning decision information, determine the warning confidence level corresponding to the warning decision information, and determine the warning priority labeling information based on the warning feedback information;

[0028] Obtain historical early warning information and the corresponding historical processing results, and determine historical feedback information based on the historical processing results;

[0029] Based on the warning confidence level, the warning priority calibration information, and the historical feedback information, a visual warning feedback information is generated.

[0030] The visual early warning feedback information includes one or more of the following: confidence heatmap feedback information, handling and source tracing view information, and dynamic priority calibration feedback information.

[0031] As an optional implementation, in a first aspect of the present invention, the step of performing a model optimization operation on the pre-set target intelligent model based on the experience replay pool to obtain the optimized target intelligent model includes:

[0032] The warning processing result is determined in the experience replay pool, and a reward signal is determined based on the warning processing result. Based on the predetermined reward function, the parameters to be adjusted in the predetermined target intelligent model are determined, and the parameter adjustment factor corresponding to each parameter to be adjusted is determined.

[0033] Based on each parameter to be adjusted and the parameter adjustment factor corresponding to each parameter to be adjusted, a model adjustment operation is performed on the pre-set target intelligent model to obtain the target adjusted model;

[0034] The target adjustment model and the target intelligent model are deployed to perform parallel processing early warning operations, and the first reward information corresponding to the target adjustment model and the second reward information corresponding to the target intelligent model are determined.

[0035] An information comparison operation is performed on the first reward information and the second reward information to obtain the information comparison result. Based on the information comparison result, the optimized target intelligent model is determined.

[0036] A second aspect of the present invention discloses an early warning information processing device for process production, the device comprising:

[0037] The data acquisition module is used to collect real-time data corresponding to the production process.

[0038] A building module is used to construct the target dataset based on all the aforementioned real-time data;

[0039] The analysis module is used to perform association analysis operations on all target data contained in the target dataset through a pre-set target intelligent model, and generate association analysis information corresponding to the target dataset. The association analysis information includes association rule information and frequent itemset information of all target data contained in the target dataset.

[0040] The classification module is used to perform a data classification operation on all the target data contained in the target dataset based on the association analysis information, to obtain at least one data category, wherein each data category includes at least one target data;

[0041] A determination module is used to determine the warning priority corresponding to each of the data categories;

[0042] The processing module is used to perform information processing operations on the information to be processed based on each of the warning priorities and the pre-set target intelligent model, so as to obtain the target warning information.

[0043] As an optional implementation, in a second aspect of the present invention, the acquisition module is further configured to acquire historical process data and equipment topology information before the analysis module performs correlation analysis on all target data contained in the target dataset through a pre-set target intelligent model to generate correlation analysis information corresponding to the target dataset.

[0044] The construction module is also used to construct a multimodal pre-training dataset based on the historical process data and the equipment topology information;

[0045] The device further includes:

[0046] The training module is used to perform pre-training operations on the standby intelligent model based on the multimodal pre-training dataset to obtain a pre-trained intelligent model;

[0047] The determining module is further configured to determine the temporal dependency relationship corresponding to the historical process data based on a pre-designed bidirectional architecture, and determine the equipment topology association features corresponding to the equipment topology information based on the temporal dependency relationship; wherein the pre-designed bidirectional architecture includes a bidirectional Transformer architecture;

[0048] The generation module is used to generate a target intelligent model based on the device topology association features and the pre-trained intelligent model.

[0049] As an optional implementation, in a second aspect of the invention, the apparatus further includes:

[0050] The detection module is used to perform data detection operations on the real-time data corresponding to the process production after the acquisition module acquires the real-time data, and obtain the data detection results.

[0051] The judgment module is used to determine whether the data detection result is used to indicate that there is abnormal data in all the real-time data;

[0052] The processing module is further configured to, when the judgment module determines that the data detection result indicates the presence of abnormal data in all the real-time data, perform anomaly processing operations on the abnormal data based on a pre-set forward propagation method to obtain target processing data; and perform data batch processing operations on the target processing data to obtain data batch processing results.

[0053] The update module is used to store the data batch processing results into a predetermined target database to establish a data tracking relationship, and update all the real-time data according to the data tracking relationship, and trigger the construction module to perform the operation of constructing a target dataset based on all the real-time data;

[0054] The data batch processing operation includes one or more of data alignment operations and data feature extraction operations; the predetermined target database includes one or more of time-series databases and relational databases.

[0055] As an optional implementation, in a second aspect of the present invention, the classification module performs a data classification operation on all the target data contained in the target dataset to obtain at least one data category in the following specific manner:

[0056] Based on pre-set association rules and all target data contained in the target dataset, generate data logical relationship information for all target data;

[0057] Extract data description information of all the target data, and based on the data description information and the data logical relationship information, determine the data mode corresponding to each target data and the mode weight corresponding to each data mode through a pre-determined fusion classification model;

[0058] Based on the data modality corresponding to each target data and the modality weight corresponding to each data modality, a data classification operation is performed on all the target data to obtain at least one data category.

[0059] As an optional implementation, in a second aspect of the invention, the apparatus further includes:

[0060] The acquisition module is used to acquire the warning feedback information corresponding to the target warning information;

[0061] The construction module is further configured to construct a set of early warning feedback states based on the early warning feedback information, wherein the set of early warning feedback states includes one or more of the following: equipment operating condition early warning feedback states, process early warning feedback states, and early warning classification strategy feedback states.

[0062] The determining module is also used to determine key alarm information in the target early warning information;

[0063] The construction module is also used to construct an experience replay pool corresponding to the target early warning information based on the key alarm information and the early warning feedback status set.

[0064] The generation module is further configured to perform model optimization operations on the pre-set target intelligent model based on the experience replay pool, to obtain the optimized target intelligent model, and to generate model optimization comparison information corresponding to the target intelligent model.

[0065] As an optional implementation, in a second aspect of the present invention, the determining module is further configured to determine early warning decision information based on the early warning feedback information, determine the early warning confidence level corresponding to the early warning decision information, and determine early warning priority calibration information based on the early warning feedback information;

[0066] The acquisition module is also used to acquire historical early warning information and the historical processing results corresponding to the historical early warning information;

[0067] The determining module is further configured to determine historical feedback information based on the historical processing results;

[0068] The generation module is also used to generate visualized early warning feedback information based on the early warning confidence level, the early warning priority calibration information, and the historical feedback information;

[0069] The visual early warning feedback information includes one or more of the following: confidence heatmap feedback information, handling and source tracing view information, and dynamic priority calibration feedback information.

[0070] As an optional implementation, in a second aspect of the present invention, the generation module performs a model optimization operation on the pre-set target intelligent model based on the experience replay pool to obtain the optimized target intelligent model in the following specific ways:

[0071] The warning processing result is determined in the experience replay pool, and a reward signal is determined based on the warning processing result. Based on the predetermined reward function, the parameters to be adjusted in the predetermined target intelligent model are determined, and the parameter adjustment factor corresponding to each parameter to be adjusted is determined.

[0072] Based on each parameter to be adjusted and the parameter adjustment factor corresponding to each parameter to be adjusted, a model adjustment operation is performed on the pre-set target intelligent model to obtain the target adjusted model;

[0073] The target adjustment model and the target intelligent model are deployed to perform parallel processing early warning operations, and the first reward information corresponding to the target adjustment model and the second reward information corresponding to the target intelligent model are determined.

[0074] An information comparison operation is performed on the first reward information and the second reward information to obtain the information comparison result. Based on the information comparison result, the optimized target intelligent model is determined.

[0075] A third aspect of the present invention discloses another early warning information processing device applied to process production, the device comprising:

[0076] Memory containing executable program code;

[0077] A processor coupled to the memory;

[0078] The processor calls the executable program code stored in the memory to execute some or all of the steps in the early warning information processing method for process production described in any of the first aspects of the present invention.

[0079] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the early warning information processing method for process production described in any of the first aspects of the present invention.

[0080] Compared with the prior art, the present invention has the following beneficial effects:

[0081] In this embodiment of the invention, real-time data corresponding to the production process is collected and a target dataset is constructed. A target intelligent model is used to perform correlation analysis on all target data contained in the target dataset to generate correlation analysis information. Based on the correlation analysis information, a data classification operation is performed on all target data contained in the target dataset to obtain at least one data category. The warning priority corresponding to each data category is determined, and based on each warning priority and the pre-set target intelligent model, information processing operations are performed on the information to be processed to obtain target warning information. Therefore, implementing this invention enables intelligent processing of warning information, which is beneficial to improving the accuracy and reliability of warning information processing. Attached Figure Description

[0082] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0083] Figure 1 This is a flowchart illustrating a method for processing early warning information in manufacturing processes, as disclosed in an embodiment of the present invention.

[0084] Figure 2 This is a flowchart illustrating another method for processing early warning information in manufacturing processes, as disclosed in an embodiment of the present invention.

[0085] Figure 3This is a schematic diagram of the structure of an early warning information processing device for process production disclosed in an embodiment of the present invention;

[0086] Figure 4 This is a schematic diagram of another early warning information processing device for process production disclosed in an embodiment of the present invention;

[0087] Figure 5 This is a schematic diagram of the structure of another early warning information processing device for process production disclosed in the embodiments of the present invention. Detailed Implementation

[0088] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0089] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0090] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0091] This invention discloses a method for processing early warning information applied to industrial processes. It enables intelligent processing of early warning information, thereby improving the accuracy and reliability of the processing. Detailed descriptions follow.

[0092] Example 1

[0093] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for processing early warning information in industrial processes, as disclosed in an embodiment of the present invention. Figure 1The described method for processing early warning information in production processes can be applied to an early warning information processing device for production processes. This device can be integrated into a cloud server or a local server; this embodiment of the invention does not impose any limitations. Figure 1 As shown, the early warning information processing method applied to process production can include the following operations.

[0094] 101. Collect real-time data corresponding to the production process, and construct the target dataset based on all real-time data.

[0095] In this embodiment of the invention, optionally, the real-time data collected corresponding to the production process may include real-time data of the target production line corresponding to the production process. The real-time data may include one or more of the following: DCS system data, sensor network data, and manually labeled early warning sample data. Further, the DCS system data may include the DCS system's control setpoints (SP) for the equipment, such as the set temperature of the temperature control loop and the set pressure of the pressure control loop; control output values ​​(OP), such as the opening degree of the regulating valve and the frequency output of the frequency converter; and measured values ​​(PV), i.e., the measured values ​​of the actual operating parameters of the equipment. Sensor network data may include real-time collection of physical quantity parameters of the equipment through various sensors, such as the equipment's vibration frequency, amplitude, displacement, motor torque, power factor, bearing temperature, and performance indicators such as the equipment's efficiency, power, and energy consumption. The manually labeled early warning sample data may include one or more of the following: early warning category data, early warning level data, and labeled timestamp data.

[0096] In this embodiment of the invention, optionally, the real-time data corresponding to the production process can be collected by deploying various sensors on the target production line, such as temperature sensors, pressure sensors, flow sensors, and other industrial IoT devices.

[0097] In this embodiment of the invention, optionally, the target dataset can be all real-time data.

[0098] 102. Using a pre-set target intelligent model, perform correlation analysis on all target data contained in the target dataset to generate correlation analysis information corresponding to the target dataset.

[0099] In this embodiment of the invention, the association analysis information includes association rule information and frequent itemset information for all target data contained in the target dataset.

[0100] In this embodiment of the invention, the target intelligent model may optionally include the DeepSeek large model.

[0101] In this embodiment of the invention, optionally, the association rule information can be a rule describing interesting associations or relationships between itemsets or events in a dataset, which is a technique for mining meaningful associations from a large amount of data; a frequent itemset refers to a set of items that frequently appear in a given dataset; where "frequent appearance" is measured by a certain frequency threshold (called minimum support), that is, when the number or proportion of times an itemset appears in the dataset reaches or exceeds the threshold, it is considered a frequent itemset.

[0102] In this embodiment of the invention, optionally, the association rule information and frequent itemset information can be generated by: 1) generating frequent itemsets based on the improved Apriori algorithm (with dynamically adjusted support threshold, initial value 0.3), calculating itemset similarity through pre-trained model embedding vectors, and merging semantically equivalent items (such as "abnormal pump vibration" and "equipment resonance alarm"). The Apriori algorithm is a classic association rule mining algorithm used to discover frequent itemsets and association rules in a dataset.

[0103] In this embodiment of the invention, optionally, the above-mentioned method of performing correlation analysis on all target data contained in the target dataset using a pre-set target intelligent model to generate correlation analysis information corresponding to the target dataset may include:

[0104] Based on the Apriori algorithm and a pre-set target intelligent model, the similarity of itemsets is calculated by embedding vectors to obtain frequent itemset information. The device topology is modeled based on graph attention network (GAT), and cross-sensor association rule information is generated through the Transformer decoder. Based on the frequent itemset information and association rule information, the association analysis information corresponding to the target dataset is generated.

[0105] 103. Based on the association analysis information, perform data classification operations on all target data contained in the target dataset to obtain at least one data category.

[0106] In this embodiment of the invention, each data category includes at least one target data.

[0107] In this embodiment of the invention, the number of data categories may be one or more, and this embodiment of the invention does not impose a specific limitation.

[0108] 104. Determine the warning priority corresponding to each data category, and based on each warning priority and the pre-set target intelligent model, perform information processing operations on the information to be processed to obtain target warning information.

[0109] In this embodiment of the invention, optionally, the warning priority corresponding to each data category can be obtained by comprehensively considering factors such as the urgency of the warning information, the degree of impact on production, the historical frequency of occurrence, and potential risks.

[0110] In this embodiment of the invention, optionally, determining the warning priority corresponding to each data category may include:

[0111] Determine the category parameters corresponding to each data category, and determine the weight coefficients corresponding to each category parameter. The category parameters include one or more of the following: warning urgency parameter, impact parameter, historical occurrence frequency parameter, potential risk parameter, current production line status parameter, and warning timeliness parameter.

[0112] Based on the weight coefficients corresponding to each category parameter, calculate the priority score for each category parameter, and determine the warning priority for each data category based on the priority score.

[0113] In this embodiment of the invention, the priority score can be calculated as follows: Priority score = Σ (factor weight × factor score).

[0114] In this embodiment of the invention, optionally, the above-mentioned information processing operation performed on the information to be processed based on each early warning priority and a pre-set target intelligent model to obtain target early warning information may include:

[0115] Based on each early warning priority and a pre-set target intelligent model, the information to be processed is comprehensively evaluated to obtain the information evaluation result.

[0116] Based on the information evaluation results, the weight of each piece of information to be processed is determined by combining the Analytic Hierarchy Process (AHP), and the information to be processed is classified and graded to obtain the target early warning information corresponding to the information to be processed.

[0117] In this embodiment of the invention, optionally, for example, factors such as the urgency of the warning information, its impact on production, historical frequency of occurrence, potential risks, the current status of the production line, the availability of operators, and the timeliness of the warning information are comprehensively considered, and a priority score for each warning information is calculated based on a weighting coefficient to determine the priority level (high, medium, low); furthermore, the warning priority can be dynamically adjusted according to the real-time status of the production line (such as raising the equipment overheat warning level when operating at full load).

[0118] In this embodiment of the invention, optionally, a two-stage fine-tuning strategy can be adopted for the target intelligent model: the first stage freezes the bottom feature extraction layer and only fine-tunes the top classifier to adapt to the specific early warning label system; the second stage opens all network parameters and uses a dynamic learning rate strategy to optimize the data augmentation module end-to-end, expanding the training samples through time series interpolation, noise injection, and working condition simulation to improve the model's robustness to data-sparse scenarios. Furthermore, the robustness of the model can be further improved by providing annotation tools for engineers to select key alarm events and by incremental training with small samples (using Adapter fine-tuning technology, freezing 95% of the original parameters).

[0119] It is evident that implementation Figure 1 The described early warning information processing method applied to process production can collect real-time data corresponding to process production and construct a target dataset. It then performs correlation analysis on the target data using a target intelligent model to obtain correlation analysis information. Based on the correlation analysis information, it performs data classification on the target data in the target dataset to obtain data categories and determines the early warning priority corresponding to each data category. Combining this with the target intelligent model, it performs information processing operations on the information to be processed to obtain target early warning information. This method can quickly identify association rules and frequent itemsets in the data, perform rapid and accurate correlation analysis, effectively identify and eliminate false alarms, significantly reduce the number of early warnings that operators need to handle, thereby improving early warning response speed and overall processing efficiency. Furthermore, the association rules and frequent itemsets mined by the target intelligent model can more accurately identify real faults or abnormal situations, effectively reducing the occurrence of false alarms. Meanwhile, by predicting potential fault trends, the reliability and efficiency of the early warning system are improved. Furthermore, based on the discovered association rules, early warning information can be automatically classified, graded, and prioritized, achieving intelligent early warning management and improving the accuracy and relevance of early warning processing. The system also allows for flexible responses to changes in different process production environments and early warning data characteristics. By adjusting support and confidence thresholds, it can adapt to different early warning data processing needs, demonstrating good scalability and adaptability. This not only significantly improves the efficiency and accuracy of early warning processing and reduces the false alarm rate, but also achieves intelligent early warning management, enhancing the system's scalability and adaptability, improving the overall efficiency and safety of process production, and further enabling intelligent processing of early warning information, which is beneficial for improving the accuracy and reliability of early warning information processing.

[0120] Example 2

[0121] Please see Figure 2 , Figure 2 This is a flowchart illustrating another early warning information processing method for process production disclosed in an embodiment of the present invention. Figure 2The described method for processing early warning information in production processes can be applied to an early warning information processing device for production processes. This device can be integrated into a cloud server or a local server; this embodiment of the invention does not impose any limitations. Figure 2 As shown, the early warning information processing method applied to process production may include the following operations:

[0122] 201. Collect real-time data corresponding to the production process, and construct the target dataset based on all real-time data.

[0123] 202. Collect historical process data and equipment topology information, construct a multimodal pre-training dataset based on the historical process data and equipment topology information, and perform pre-training operations on the standby intelligent model based on the multimodal pre-training dataset to obtain the pre-trained intelligent model.

[0124] In this embodiment of the invention, the historical process data may optionally include one or more of the following: equipment parameter data, environmental data, operation log data, and historical early warning record data.

[0125] In this embodiment of the invention, optionally, the equipment topology information may include one or more of the following: equipment connection relationship information, equipment layout information, equipment location information, equipment function information, and equipment dependency relationship information corresponding to the process production.

[0126] In this embodiment of the invention, optionally, the multimodal pre-training dataset includes at least all historical process data and equipment topology information.

[0127] In this embodiment of the invention, optionally, the above-mentioned pre-training operation on the standby intelligent model based on the multimodal pre-training dataset to obtain the pre-trained intelligent model may include:

[0128] Based on a multimodal pre-training dataset, self-supervised learning is used to pre-train the backup intelligent model to obtain a pre-trained intelligent model.

[0129] 203. Based on the pre-designed bidirectional architecture, determine the temporal dependencies corresponding to historical process data, and determine the equipment topology association features corresponding to equipment topology information based on the temporal dependencies.

[0130] In this embodiment of the invention, the pre-designed bidirectional architecture includes a bidirectional Transformer architecture.

[0131] In this embodiment of the invention, optionally, the above-mentioned determination of the temporal dependency relationship corresponding to historical process data based on a pre-designed bidirectional architecture, and the determination of the equipment topology association features corresponding to the equipment topology information based on the temporal dependency relationship, may include:

[0132] Based on the bidirectional Transformer architecture, the temporal dependencies between process parameters are learned through mask prediction tasks. Based on the temporal dependencies and multimodal pre-trained datasets, graph neural network branches are constructed. Based on the graph neural network branches, the device topology association features corresponding to the device topology information are captured.

[0133] 204. Generate the target intelligent model based on the equipment topology association features and the pre-trained intelligent model.

[0134] In this embodiment of the invention, optionally, the generation of the target intelligent model based on the device topology association features and the pre-trained intelligent model may include:

[0135] Based on the device topology association characteristics and temporal dependencies, training operations are performed on the pre-trained intelligent model to train it to convergence and obtain the target intelligent model.

[0136] In a further optional embodiment of the present invention, the above method may also include introducing a contrastive learning mechanism to perform adversarial representation learning on normal and abnormal operating condition data in the feature space, thereby enhancing the model's sensitivity to potential anomalies.

[0137] 205. Using a pre-set target intelligent model, perform correlation analysis on all target data contained in the target dataset to generate correlation analysis information corresponding to the target dataset.

[0138] 206. Based on the association analysis information, perform data classification operations on all target data contained in the target dataset to obtain at least one data category.

[0139] 207. Determine the warning priority corresponding to each data category, and based on each warning priority and the pre-set target intelligent model, perform information processing operations on the information to be processed to obtain target warning information.

[0140] In this embodiment of the invention, for detailed descriptions of steps 201 and 205-207, please refer to the other descriptions of steps 101-104 in Embodiment 1. These descriptions will not be repeated in this embodiment of the invention.

[0141] It is evident that implementation Figure 2The described early warning information processing method for process generation can collect historical process data and equipment topology information to construct a multimodal pre-training dataset, and then perform pre-training operations on a backup intelligent model to obtain a pre-trained intelligent model. Furthermore, based on a pre-designed bidirectional architecture, it determines temporal dependencies and thus identifies equipment topology association features corresponding to the equipment topology information. Based on these equipment topology association features and the pre-trained intelligent model, a target intelligent model is generated. By collecting historical process data and equipment topology information to construct a multimodal pre-training dataset containing rich information, and performing pre-training operations on the backup intelligent model, the model learns general feature representations on large-scale multimodal data, improving the model's adaptability and flexibility. Moreover, based on a bidirectional Transformer architecture, it can effectively capture temporal dependencies in historical process data and determine the associations between equipment by analyzing equipment topology information. This feature not only focuses on the status of individual devices but also understands the mutual influence and synergy between devices. Combining device topology association features with pre-trained intelligent models, the generated target intelligent model can more accurately perform association analysis operations, thereby generating more accurate and meaningful association analysis information. Through the target intelligent model, early warning data can be automatically classified, graded, and prioritized based on association analysis information, realizing intelligent early warning management, improving the efficiency and accuracy of early warning processing, enabling operators to respond to key early warning information more quickly, and improving the safety of equipment operation and process production. It not only enhances the model's generalization ability and adaptability but also strengthens the understanding and analysis capabilities of the process, improves the accuracy and efficiency of association analysis, optimizes the early warning information processing flow, and further enables intelligent processing of early warning information, which is conducive to improving the accuracy and reliability of early warning information processing.

[0142] In an optional embodiment, after acquiring real-time data corresponding to the production process, the method further includes:

[0143] Perform data detection operations on real-time data, obtain data detection results, and determine whether the data detection results can be used to indicate that there is abnormal data in all real-time data;

[0144] When it is determined that the data detection result indicates the presence of abnormal data in all real-time data, anomaly processing operations are performed on the abnormal data based on the pre-set forward propagation method to obtain the target processed data;

[0145] Perform batch processing on the target data to obtain batch processing results, and store the batch processing results in a pre-determined target database to establish data tracking relationships. Based on the data tracking relationships, update all real-time data and trigger the execution of the operation to build the target dataset based on all real-time data.

[0146] The data batch processing operations include one or more of data alignment operations and data feature extraction operations; the pre-determined target database includes one or more of time-series databases and relational databases.

[0147] In this embodiment of the invention, optionally, the above-mentioned data detection operation on real-time data to obtain data detection results may include: performing data detection operation on real-time data using the 3σ principle to obtain data detection results. The 3σ principle is a statistical empirical rule based on normal distribution, used to determine whether data is abnormal. In a normal distribution, most data is concentrated near the mean, and the probability of data appearing is lower the further away from the mean.

[0148] In this embodiment of the invention, the abnormal data may optionally include one or more of erroneous data and null data.

[0149] In this embodiment of the invention, optionally, the process can be terminated when it is determined that the data detection result indicates that there is no abnormal data among all real-time data.

[0150] In this embodiment of the invention, optionally, the above-mentioned anomaly processing operation on abnormal data based on a pre-set forward propagation method to obtain target processed data may include: determining the filling data corresponding to the abnormal data based on a pre-set forward propagation method, and performing anomaly processing operation on the abnormal data based on the filling data to obtain target processed data. For example, in time series data, if pressure data is missing at a certain moment, and the previous normal data is 0.8 MPa, then 0.8 MPa is used to fill the missing point to ensure data continuity.

[0151] In this embodiment of the invention, optionally, the above-mentioned data batch processing operation on the target processing data to obtain the data batch processing result may include: periodically performing data alignment operation on the target processing data to unify the timestamp granularity of the target processing data to generate data alignment result; and performing operations to extract statistical features and extract frequency domain features on the target processing data to generate data feature result; and obtaining the data batch processing result based on the data alignment result and the data feature result. Further, the data batch processing operation on the target processing data may be performed at a batch processing layer.

[0152] In this embodiment of the invention, optionally, the above-mentioned method of storing the data batch processing results into a predetermined target database to establish a data tracking relationship, and updating all real-time data according to the data tracking relationship, may include:

[0153] Based on the data batch processing results, the data alignment results are stored in a time-series database within a pre-determined target database, and the data feature results are stored in a relational database within the pre-determined target database to establish data tracing relationships. All real-time data is then updated according to these data tracing relationships. The data tracing relationships include a data lineage tracing mechanism. The time-series database includes an InfluxDB database, and the relational database includes a MySQL database.

[0154] In this embodiment of the invention, optionally, the time-series database is mainly used to store real-time data with timestamps, such as time-series records of sensor data such as temperature, pressure, and flow, which facilitates subsequent querying and analysis of data change trends in chronological order; the relational database is used to store structured data related to process production, such as equipment information, operation logs, maintenance records, etc., as well as the relationships between different data types.

[0155] As can be seen, implementing this optional embodiment can perform data detection operations on real-time data to obtain data detection results, determine whether the data detection results are used to indicate the presence of abnormal data in all real-time data; if so, perform anomaly handling operations on the abnormal data based on the forward propagation method to obtain target processing data; perform data batch processing operations on the target processing data to obtain data batch processing results, and store the data batch processing results into a pre-determined target database to establish a data tracking relationship, and update all real-time data according to the data tracking relationship, enabling the detection of real-time data and utilizing 3σ The system identifies anomalous data based on principles and processes it using the forward propagation method, improving data accuracy and completeness. This contributes to the accuracy of subsequent early warning analysis and model training based on this data. Batch data processing operations, including data alignment and feature extraction, enable unified processing of real-time data of different types and timestamps, resulting in better consistency and comparability. This improves data storage efficiency and query performance, providing strong support for subsequent data analysis and mining. Timely updates of all real-time data based on data tracking relationships ensure data timeliness and accuracy, enhancing the timeliness and effectiveness of early warnings. Furthermore, it improves the system's ability to perform correlation analysis, intelligent classification, and hierarchical processing of early warning information, reducing false alarm rates and enhancing the overall performance and adaptability of the early warning system. Moreover, it enables intelligent processing of early warning information, further improving the accuracy and reliability of early warning information processing.

[0156] In another alternative embodiment, a data classification operation is performed on all target data contained in the target dataset to obtain at least one data category, including:

[0157] Based on pre-set association rules and all target data contained in the target dataset, generate data logical relationship information for all target data;

[0158] Extract the data description information of all target data. Based on the data description information and data logical relationship information, determine the data mode corresponding to each target data and the mode weight corresponding to each data mode through a pre-determined fusion classification model.

[0159] Based on the data modality corresponding to each target data and the modality weight corresponding to each data modality, a data classification operation is performed on all target data to obtain at least one data category.

[0160] In this optional embodiment, the process of generating data logical relationship information for all target data based on pre-set association rules and all target data included in the target dataset may include: analyzing all target data included in the target dataset based on pre-set association rules to determine the data association relationships and trend relationships between each target data, and generating data logical relationship information for all target data based on the data association relationships and trend relationships. For example, a positive correlation relationship may be used, and the specific manifestation of this logical relationship in the data may be recorded, such as an average increase in pressure of 0.5 MPa for every 10°C increase in temperature.

[0161] In this optional embodiment, optionally, the extraction of data description information from all target data, and the determination of the data modality corresponding to each target data and the modality weight corresponding to each data modality based on the data description information and data logical relationship information through a pre-determined fusion classification model, may include:

[0162] Extract the data description information for each target data, and input the data description information and data logical relationship information into a pre-determined fusion classification model to determine the data modality corresponding to each target data through the pre-determined fusion classification model; wherein, the fusion classification model may include a BERT model;

[0163] Based on the pre-determined gated fusion network and the data mode corresponding to each target data, determine the mode weight corresponding to each data mode.

[0164] In this optional embodiment, BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language representation model based on the Transformer architecture; Gated Fusion Network is a neural network architecture for multimodal data fusion, designed to effectively fuse information from different modalities (such as text, images, audio, etc.) to improve the model's ability to understand and process multimodal data.

[0165] In this optional embodiment, for example, for structured data features, an event logic graph can be generated based on association rules to generate data logic relationship information; for unstructured data features, fault descriptions can be extracted from maintenance record text (through the BERT model); and for the fusion classification module, a gated fusion network is used to dynamically allocate the modality weights of each modality to obtain the modality weights corresponding to each data modality.

[0166] In this optional embodiment, for example, the data such as temperature, pressure, and flow rate in the equipment operating parameter mode with higher weight can be further subdivided into different categories. For instance, based on the normal range and fluctuation of temperature data, it can be divided into normal operating temperature category, slightly abnormal temperature category, and severely abnormal temperature category. Similarly, for the maintenance record text in the equipment maintenance information mode, it can be divided into routine maintenance category, minor fault repair category, and major fault repair category based on the fault type and severity described in the text. The data in each category has similar data mode characteristics and similar importance in early warning processing. For example, under the equipment operating parameter mode, normal operating parameter category, slightly abnormal parameter category, and severely abnormal parameter category are obtained; under the equipment maintenance information mode, routine maintenance category, minor fault repair category, and major fault repair category are obtained.

[0167] As can be seen, implementing this optional embodiment can generate data logical relationship information for all target data based on pre-set association rules and all target data contained in the target dataset; extract data description information for all target data; and, based on the data description information and data logical relationship information, determine the data modality corresponding to each target data and the modality weight corresponding to each data modality through a fusion classification model; and perform data classification operations on all target data based on the data modality and modality weight corresponding to each data modality to obtain at least one data category. It can clarify the characteristics of each data point and its correlation with other data by performing logical relationship analysis and descriptive information extraction on the data in the target dataset, and generate data logical relationship information based on pre-set association rules. It can capture the inherent relationships between data, more accurately identify data anomalies, and combine data description information and logical relationship information to more comprehensively evaluate the characteristics of each data point, which is conducive to significantly improving the accuracy of classification. By assigning weights to each data modality, the model can dynamically adjust the classification strategy according to the importance of different modalities, which has strong adaptability and can effectively handle multimodal data. Furthermore, by assigning weights to each data modality, the system can dynamically adjust the weight allocation strategy according to the needs of different scenarios, thereby improving the system's flexibility and scalability. Through multi-dimensional analysis and dynamic weight adjustment, it can more accurately identify real abnormal data and intelligently process early warning information, which is conducive to improving the accuracy and reliability of early warning information processing.

[0168] In yet another optional embodiment, the method further includes:

[0169] Obtain the warning feedback information corresponding to the target warning information, and construct a warning feedback status set based on the warning feedback information. The warning feedback status set includes one or more of the following: equipment operating condition warning feedback status, process warning feedback status, and warning classification strategy feedback status.

[0170] The key alarm information in the target early warning information is identified, and based on the key alarm information and the set of early warning feedback states, an experience replay pool corresponding to the target early warning information is constructed. Based on the experience replay pool, a model optimization operation is performed on the pre-set target intelligent model to obtain the optimized target intelligent model, and model optimization comparison information corresponding to the target intelligent model is generated.

[0171] In this optional embodiment, the early warning feedback information may optionally include feedback information from users corresponding to the process production environment regarding the target early warning information. The early warning feedback information may include one or more of the following: verbal feedback, action feedback, and operational feedback. Equipment operating condition early warning feedback status may include operator feedback on the equipment's operating status; process early warning feedback status may include operator feedback on abnormal process parameters; and early warning grading strategy feedback status may include operator feedback on the priority of early warning information, such as confirming whether the early warning is a false alarm or whether the early warning level needs to be adjusted. For example, the operator might report, "This early warning is a false alarm; it is recommended to lower the early warning level."

[0172] In this optional embodiment, the key alarm information in the target warning information may include one or more of the following: alarm time information, alarm type information, alarm category information, alarm description information, alarm associated equipment information, and alarm associated process information.

[0173] In this optional embodiment, the above-mentioned construction of the experience replay pool corresponding to the target early warning information based on key alarm information and early warning feedback status set may include:

[0174] Based on key alarm information and early warning feedback status, an experience replay pool corresponding to the target early warning information is constructed using a TD-error priority sampling mechanism. The TD-error priority sampling mechanism includes temporal difference error, a key indicator for measuring sample importance, representing the difference between the predicted value of the current state and the target value. Its calculation formula is: TD_error = r + γ⋅V(s′)−V(s), where r is the reward value of the current sample; γ is the discount factor; V(s) is the value function of the current state s; and V(s′) is the value function of the next state s′. Furthermore, in reinforcement learning, a larger TD-error indicates that the current sample is more valuable for model updates and should therefore be prioritized for sampling.

[0175] In this optional embodiment, the above-mentioned model optimization operation on the pre-set target intelligent model to obtain the optimized target intelligent model and generate model optimization comparison information corresponding to the target intelligent model may include: performing a model comparison operation based on the optimized target intelligent model and the target intelligent model before optimization to generate model optimization comparison information; wherein, the model optimization comparison information includes one or more of the following: F1-score and false positive rate; wherein, F1-score is a comprehensive index used to measure the harmonic mean of the precision and recall of a classification model, and is an index that balances precision and recall, and is particularly suitable for imbalanced datasets; the false positive rate (FPR) refers to the proportion of negative samples that the model incorrectly predicts as positive samples, and it reflects the model's misclassification of negative samples.

[0176] As can be seen, implementing this optional embodiment can acquire the warning feedback information corresponding to the target warning information, construct a warning feedback state set based on the warning feedback information, determine the key alarm information in the target warning information, and construct an experience replay pool corresponding to the target warning information based on the key alarm information and the warning feedback state set. Based on the experience replay pool, a model optimization operation is performed on the pre-set target intelligent model to obtain an optimized target intelligent model, and model optimization comparison information corresponding to the target intelligent model is generated. This allows for dynamic adjustment of the warning strategy through the acquired warning feedback information, and the reinforcement learning mechanism based on the experience replay pool enables the model to learn better decision-making strategies from historical data, further improving the accuracy and timeliness of warnings. The system can dynamically adjust the warning strategy based on the information in the warning feedback state set, which is beneficial for improving the generation... The system's intelligent, accurate, and reliable early warning information is enhanced through a tiered early warning strategy that allows for dynamic adjustment of warning priorities based on operator feedback. Continuous optimization of the target intelligent model enables the system to better adapt to changes in the production process, maintaining high performance under various conditions. Based on the early warning feedback status set and key alarm information, the system can flexibly adjust early warning strategies to address different production scenarios and needs. By generating model optimization comparison information, the system can continuously evaluate and optimize its performance, ensuring high efficiency and stability during long-term operation, thereby improving system flexibility and scalability. Through multi-dimensional analysis and dynamic weight adjustment, it can more accurately identify truly abnormal data and intelligently process early warning information, improving the accuracy and reliability of early warning information processing.

[0177] In yet another optional embodiment, the method further includes:

[0178] Based on the early warning feedback information, determine the early warning decision information, determine the early warning confidence level corresponding to the early warning decision information, and determine the early warning priority labeling information based on the early warning feedback information;

[0179] Obtain historical early warning information and the corresponding historical processing results, and determine historical feedback information based on the historical processing results;

[0180] Based on the warning confidence level, warning priority labeling information, and historical feedback information, visualized warning feedback information is generated.

[0181] Among them, the visual early warning feedback information includes one or more of the following: confidence heatmap feedback information, handling and source tracing view information, and dynamic priority calibration feedback information.

[0182] In this optional embodiment, the early warning decision information may optionally include one or more of early warning type information and early warning description information. The early warning type information may include equipment fault information, process abnormality information, environmental alarm information, etc.; the early warning description information may include a specific description of the early warning, such as "the temperature of equipment A exceeds the threshold".

[0183] In this optional embodiment, the warning confidence level reflects the system's confidence in the accuracy of the warning information; the confidence level can be calculated based on the model's prediction probability, the accuracy of historical data, and operator feedback.

[0184] In this optional embodiment, the aforementioned determination of early warning priority labeling information based on early warning feedback information may include: analyzing early warning feedback information, determining the real-time load rate of the target production line corresponding to the process based on the early warning feedback information, and determining early warning priority labeling information based on the real-time load rate. For example, analyzing early warning feedback information may involve determining whether the actual operating status of the equipment is consistent with the early warning information based on the equipment operating condition early warning feedback status, evaluating whether the actual changes in process parameters are consistent with the early warning information based on the process early warning feedback status, and judging whether the initial classification of the early warning information is accurate based on the early warning classification strategy feedback status; if the equipment operating condition feedback shows that the equipment is in an abnormal state, and the process parameter feedback shows that there is a potential risk in the process operation, then the priority of the early warning information is labeled as "high"; the early warning priority is further adjusted based on the real-time load rate of the production line. If the production line is at full load, and the early warning information is an equipment overheating early warning, then the priority of the early warning information is raised to the highest level.

[0185] In this optional embodiment, the aforementioned historical warning information may include one or more of the following: historical warning type information, historical warning description information, historical warning equipment information, and historical warning process information; the historical processing results corresponding to the historical warning information may include one or more of the following: historical processing method information, historical processing time information, and historical processing result information.

[0186] In this optional embodiment, the above-mentioned determination of historical feedback information based on historical processing results may include: determining the processing records and final results of similar historical warnings based on historical processing results, and determining historical feedback information based on the processing records and final results.

[0187] In this optional embodiment, the generation of visualized early warning feedback information based on early warning confidence level, early warning priority calibration information, and historical feedback information may include: generating a confidence level heatmap based on the early warning confidence level, and generating confidence level heatmap feedback information based on the confidence level heatmap; generating a handling traceability view based on historical feedback information; and generating dynamic priority calibration feedback information based on early warning priority calibration information. For example, the confidence level heatmap displays the credibility of the current early warning decision (0-100%); the handling traceability view displays the processing records and final results of similar historical early warnings; and the dynamic priority calibration information displays the automatic adjustment of the early warning level based on the real-time load rate of the production line (e.g., increasing the equipment overheating early warning level when at full load).

[0188] As can be seen, implementing this optional embodiment can determine early warning decision information and the corresponding early warning confidence level based on early warning feedback information; determine early warning priority labeling information based on early warning feedback information; acquire historical early warning information and the corresponding historical processing results; determine historical feedback information based on historical processing results; and generate visualized early warning feedback information based on early warning confidence level, early warning priority labeling information, and historical feedback information. Through visualized early warning feedback information, operators can intuitively understand the confidence level, priority, and historical processing status of early warnings. Dynamic priority labeling and a handling traceability view help operators quickly identify key early warnings, optimize resource allocation, and improve decision-making efficiency. Dynamically adjusting early warning priority and confidence level based on historical feedback information allows the system to better adapt to changes in the production environment, improving the accuracy and reliability of the early warning system. Visualized information effectively reduces the cognitive burden on operators, improves system usability and user satisfaction, and enables intelligent processing of early warning information, which is beneficial for improving the accuracy and reliability of early warning information processing. It also improves the intelligence and ease of interaction between early warning information and operators, enhancing user convenience and comfort.

[0189] In another optional embodiment, based on an experience replay pool, a model optimization operation is performed on a pre-set target intelligent model to obtain an optimized target intelligent model, including:

[0190] The warning processing result is determined in the experience replay pool, and a reward signal is determined based on the warning processing result. Based on the pre-determined reward function, the parameters to be adjusted in the pre-set target intelligent model are determined, as well as the parameter adjustment factor corresponding to each parameter to be adjusted is determined.

[0191] Based on each parameter to be adjusted and the parameter adjustment factor corresponding to each parameter to be adjusted, a model adjustment operation is performed on the pre-set target intelligent model to obtain the target adjusted model;

[0192] Deploy the target adjustment model and the target intelligent model to perform parallel processing early warning operations, and determine the first reward information corresponding to the target adjustment model and the second reward information corresponding to the target intelligent model;

[0193] Perform an information comparison operation on the first reward information and the second reward information to obtain the information comparison results. Based on the information comparison results, determine the optimized target intelligent model.

[0194] In this optional embodiment, the warning processing result may optionally include one of the following: warning confirmation result, warning false alarm result, and warning missed alarm result.

[0195] In this optional embodiment, the reward function, optionally, may include:

[0196] R = α·TP - β·FP - γ·FN;

[0197] Where TP: correct alert, FP: false alarm, FN: missed alert, α, β, γ are weighting coefficients, and R is the reward value.

[0198] In this optional embodiment, the above-mentioned determination of the parameters to be adjusted in the pre-set target intelligent model based on the pre-determined reward function, and the determination of the parameter adjustment factor corresponding to each parameter to be adjusted, may include: determining the parameters to be adjusted in the pre-set target intelligent model based on the pre-determined reward function, wherein the parameters to be adjusted include one or more of weight parameters, bias parameters, and learning rate parameters; determining the reward value based on the pre-determined reward function, and determining the parameter adjustment factor corresponding to each parameter to be adjusted based on the reward value. For example, if the reward signal is positive (correct warning), the weight of the relevant parameter is increased; if the reward signal is negative (false alarm or missed alarm), the weight of the relevant parameter is decreased.

[0199] In this optional embodiment, optionally, a model adjustment operation is performed on a pre-set target intelligent model based on each parameter to be adjusted and the parameter adjustment factor corresponding to each parameter to obtain a target adjusted model. This can include: adjusting each parameter to be adjusted to the value corresponding to the parameter adjustment factor corresponding to each parameter to obtain a target adjusted model. Further, a model adjustment operation can be performed on the target intelligent model using the Proximal Policy Optimization (PPO) algorithm to obtain a target adjusted model. For example, the PPO algorithm is used to update model parameters. The PPO algorithm optimizes the model's decision policy by maximizing the cumulative reward. The specific steps are as follows: Calculate TD-error: Calculate the temporal difference error based on samples in the experience replay pool; Update the policy network: Update the model's policy network parameters based on the TD-error and reward signal; Update the value network: Simultaneously update the value network parameters to estimate the value function of the state. This allows the Proximal Policy Optimization (PPO) algorithm to continuously adjust the decision policy of a large model. Updating model parameters through a combination of offline experience replay pool and online interaction improves the intelligence and accuracy of updating model parameters.

[0200] In this optional embodiment, the process of deploying the target adjustment model and the target intelligent model to perform parallel processing early warning operations and determine the first reward information corresponding to the target adjustment model and the second reward information corresponding to the target intelligent model may include:

[0201] The target adjustment model and the target intelligence model are deployed to perform parallel processing of early warning operations. For each early warning data, the two models generate early warning decisions respectively, and calculate the first reward information corresponding to the target adjustment model and the second reward information corresponding to the target intelligence model according to the pre-determined reward function.

[0202] In this optional embodiment, optionally, the above-mentioned information comparison operation on the first reward information and the second reward information to obtain the information comparison result, and the determination of the optimized target intelligent model based on the information comparison result, may include:

[0203] An information comparison operation is performed on the first and second reward information to obtain the comparison results. Based on these results, the target reward information is determined, and the model corresponding to the target reward information is identified as the optimized target intelligent model. For example, the comprehensive rewards of the two models are compared, and the model with the higher comprehensive reward is selected as the new benchmark model. For instance, if the comprehensive reward of the current version model is higher than that of the historical version model, the current version model is replaced with the new benchmark model; if the comprehensive reward of the historical version model is higher, the historical version model is retained as the benchmark.

[0204] As can be seen, implementing this optional embodiment can determine the early warning processing result in the experience replay pool, and determine the reward signal based on the early warning processing result. Based on the reward function, it determines the parameters to be adjusted and the corresponding parameter adjustment factors in the pre-set target intelligent model. Based on each parameter to be adjusted and the corresponding parameter adjustment factor, it performs a model adjustment operation on the pre-set target intelligent model to obtain the target adjusted model. It deploys the target adjusted model and the target intelligent model to perform parallel processing early warning operation, and determines the first reward information corresponding to the target adjusted model and the second reward information corresponding to the target intelligent model and compares them to obtain the information comparison result. Based on the information comparison result, it determines the optimized target intelligent model, which can dynamically adjust its own parameters using past experience. This approach enables the model to continuously learn and adapt to new situations, improving its adaptability to complex environments. Based on a pre-determined reward function for rewards and signals, the model can dynamically adjust the parameters to be adjusted and their corresponding adjustment factors. Through dynamic adjustment mechanisms, the model can optimize its decision-making strategy in a timely manner based on actual feedback. Through parallel processing and information comparison, the system can select a model version with higher overall rewards as a new benchmark, ensuring the reliability and stability of the model in practical applications. It can also continuously evaluate and select better model versions. By combining offline experience replay pools and online interaction to update model parameters, it utilizes historical data experience and can reflect changes in the current environment in a timely manner, ensuring continuous optimization of the model. By optimizing the model's decision-making strategy through reinforcement learning algorithms (such as PPO), the model can automatically adjust its behavior based on reward signals, improving its intelligence level. By optimizing the model's decision-making strategy, the system can more accurately identify key early warning information and further intelligently process the early warning information, which is beneficial to improving the accuracy and reliability of early warning information processing.

[0205] Example 3

[0206] Please see Figure 3 , Figure 3 This is a schematic diagram of a color difference automatic calibration and control device for multiple lamps disclosed in an embodiment of the present invention. Figure 3 As shown, the early warning information processing device applied to process production may include:

[0207] The data acquisition module 301 is used to acquire real-time data corresponding to the production process.

[0208] Module 302 is used to build the target dataset based on all real-time data;

[0209] The analysis module 303 is used to perform association analysis on all target data contained in the target dataset through a pre-set target intelligent model, and generate association analysis information corresponding to the target dataset. The association analysis information includes association rule information and frequent itemset information of all target data contained in the target dataset.

[0210] The classification module 304 is used to perform data classification operations on all target data contained in the target dataset based on association analysis information to obtain at least one data category, wherein each data category includes at least one target data;

[0211] Module 305 is used to determine the warning priority for each data category;

[0212] The processing module 306 is used to perform information processing operations on the information to be processed based on each early warning priority and a pre-set target intelligent model, so as to obtain the target early warning information.

[0213] It is evident that implementation Figure 3 The described device can collect real-time data corresponding to the process production and construct a target dataset. It performs correlation analysis on the target data through a target intelligent model to obtain correlation analysis information. Based on the correlation analysis information, it performs data classification on the target data in the target dataset to obtain data categories and determines the warning priority corresponding to each data category. Combined with the target intelligent model, it performs information processing operations on the information to be processed to obtain target warning information. It can quickly identify the correlation rules and frequent itemsets in the data, perform fast and accurate correlation analysis, effectively identify and eliminate false alarms, significantly reduce the number of warnings that operators need to handle, thereby improving the warning response speed and overall processing efficiency. Furthermore, the correlation rules and frequent itemsets mined by the target intelligent model can more accurately identify real faults or abnormal situations, effectively reducing the occurrence of false alarms. Meanwhile, by predicting potential fault trends, the reliability and efficiency of the early warning system are improved. Furthermore, based on the discovered association rules, early warning information can be automatically classified, graded, and prioritized, achieving intelligent early warning management and improving the accuracy and relevance of early warning processing. The system also allows for flexible responses to changes in different process production environments and early warning data characteristics. By adjusting support and confidence thresholds, it can adapt to different early warning data processing needs, demonstrating good scalability and adaptability. This not only significantly improves the efficiency and accuracy of early warning processing and reduces the false alarm rate, but also achieves intelligent early warning management, enhancing the system's scalability and adaptability, improving the overall efficiency and safety of process production, and further enabling intelligent processing of early warning information, which is beneficial for improving the accuracy and reliability of early warning information processing.

[0214] In an optional embodiment, such as Figure 4 As shown, the acquisition module 301 is also used to acquire historical process data and equipment topology information before the analysis module 303 performs correlation analysis on all target data contained in the target dataset through a pre-set target intelligent model and generates correlation analysis information corresponding to the target dataset.

[0215] Module 302 is also used to construct a multimodal pre-training dataset based on historical process data and equipment topology information;

[0216] The device also includes:

[0217] Training module 307 is used to perform pre-training operations on the standby intelligent model based on the multimodal pre-training dataset to obtain the pre-trained intelligent model;

[0218] The determination module 305 is also used to determine the temporal dependencies corresponding to historical process data based on a pre-designed bidirectional architecture, and to determine the equipment topology association features corresponding to equipment topology information based on the temporal dependencies; wherein, the pre-designed bidirectional architecture includes a bidirectional Transformer architecture.

[0219] The generation module 308 is used to generate a target intelligent model based on the device topology association features and the pre-trained intelligent model.

[0220] It is evident that implementation Figure 4The described device can collect historical process data and equipment topology information to construct a multimodal pre-training dataset, then perform pre-training operations on a backup intelligent model to obtain a pre-trained intelligent model. Furthermore, based on a pre-designed bidirectional architecture, it determines temporal dependencies and thus identifies equipment topology association features corresponding to the equipment topology information. Based on these equipment topology association features and the pre-trained intelligent model, it generates a target intelligent model. By collecting historical process data and equipment topology information to construct a richly informative multimodal pre-training dataset and performing pre-training operations on the backup intelligent model, the model learns general feature representations on large-scale multimodal data, improving its adaptability and flexibility. Moreover, based on a bidirectional Transformer architecture, it can effectively capture temporal dependencies in historical process data and determine the association features between equipment by analyzing equipment topology information. This not only enables… By focusing on the status of individual devices and understanding the mutual influence and synergy between them, and combining device topology correlation features with pre-trained intelligent models, the generated target intelligent model can more accurately perform correlation analysis operations, thereby generating more accurate and meaningful correlation analysis information. Through the target intelligent model, early warning data can be automatically classified, graded, and prioritized based on correlation analysis information, achieving intelligent early warning management, improving the efficiency and accuracy of early warning processing, enabling operators to respond to critical early warning information more quickly, and improving the safety of equipment operation and process production. This not only enhances the model's generalization ability and adaptability but also strengthens the understanding and analysis of the process, improving the accuracy and efficiency of correlation analysis, optimizing the early warning information processing flow, and further enabling intelligent processing of early warning information, thus improving the accuracy and reliability of early warning information processing.

[0221] In another alternative embodiment, such as Figure 4 As shown, the device also includes:

[0222] The detection module 309 is used to perform data detection operations on the real-time data corresponding to the process production after the acquisition module 301 acquires the real-time data, and obtain the data detection results.

[0223] The judgment module 310 is used to determine whether the data detection result is used to indicate that there is abnormal data in all real-time data;

[0224] The processing module 306 is also used to perform anomaly processing operations on the abnormal data based on a pre-set forward propagation method when the judgment module 310 determines that the data detection result indicates that there is abnormal data in all real-time data, so as to obtain the target processing data; and to perform data batch processing operations on the target processing data to obtain the data batch processing result.

[0225] The update module 311 is used to store the data batch processing results into a predetermined target database to establish a data tracking relationship, and update all real-time data according to the data tracking relationship, and trigger the construction module 302 to perform the operation of building a target dataset based on all real-time data.

[0226] The data batch processing operations include one or more of data alignment operations and data feature extraction operations; the pre-determined target database includes one or more of time-series databases and relational databases.

[0227] It is evident that implementation Figure 4 The described apparatus can perform data detection operations on real-time data to obtain data detection results, determine whether the data detection results indicate the presence of abnormal data in all real-time data; if so, it performs anomaly handling operations on the abnormal data based on the forward propagation method to obtain target processing data; it performs data batch processing operations on the target processing data to obtain data batch processing results, and stores the data batch processing results into a pre-determined target database to establish a data tracking relationship, and updates all real-time data according to the data tracking relationship. It is capable of detecting real-time data and utilizing 3σ... The system identifies anomalous data based on principles and processes it using the forward propagation method, improving data accuracy and completeness. This contributes to the accuracy of subsequent early warning analysis and model training based on this data. Batch data processing operations, including data alignment and feature extraction, enable unified processing of real-time data of different types and timestamps, resulting in better consistency and comparability. This improves data storage efficiency and query performance, providing strong support for subsequent data analysis and mining. Timely updates of all real-time data based on data tracking relationships ensure data timeliness and accuracy, enhancing the timeliness and effectiveness of early warnings. Furthermore, it improves the system's ability to perform correlation analysis, intelligent classification, and hierarchical processing of early warning information, reducing false alarm rates and enhancing the overall performance and adaptability of the early warning system. Moreover, it enables intelligent processing of early warning information, further improving the accuracy and reliability of early warning information processing.

[0228] In yet another alternative embodiment, such as Figure 4 As shown, the classification module 304 performs data classification operations on all target data contained in the target dataset to obtain at least one data category in the following specific ways:

[0229] Based on pre-set association rules and all target data contained in the target dataset, generate data logical relationship information for all target data;

[0230] Extract the data description information of all target data. Based on the data description information and data logical relationship information, determine the data mode corresponding to each target data and the mode weight corresponding to each data mode through a pre-determined fusion classification model.

[0231] Based on the data modality corresponding to each target data and the modality weight corresponding to each data modality, a data classification operation is performed on all target data to obtain at least one data category.

[0232] It is evident that implementation Figure 4 The described apparatus is capable of generating data logical relationship information for all target data in a target dataset based on pre-set association rules; extracting data description information for all target data; determining the data modality and modality weight for each target data based on the data description information and data logical relationship information through a fusion classification model; and performing data classification operations on all target data based on the data modality and modality weight for each data modality to obtain at least one data category. It can clearly define the characteristics of each data point and its correlation with other data by performing logical relationship analysis and descriptive information extraction on the data in the target dataset, and can capture data logical relationship information based on pre-set association rules. The inherent connections between data enable more accurate identification of data anomalies. The fusion classification model combines data description information and logical relationship information, enabling a more comprehensive evaluation of the characteristics of each data point, which significantly improves classification accuracy. By assigning weights to each data modality, the model can dynamically adjust the classification strategy according to the importance of different modalities, exhibiting strong adaptability and effectively handling multimodal data. Furthermore, by assigning weights to each data modality, the system can dynamically adjust the weight allocation strategy according to the needs of different scenarios, thereby improving the system's flexibility and scalability. Through multi-dimensional analysis and dynamic weight adjustment, it can more accurately identify truly abnormal data and intelligently process early warning information, thus improving the accuracy and reliability of early warning information processing.

[0233] In yet another alternative embodiment, such as Figure 4 As shown, the device also includes:

[0234] The acquisition module 312 is used to acquire the warning feedback information corresponding to the target warning information;

[0235] The construction module 302 is also used to construct a set of early warning feedback states based on the early warning feedback information, wherein the set of early warning feedback states includes one or more of the following: equipment operating condition early warning feedback states, process early warning feedback states, and early warning classification strategy feedback states.

[0236] The determination module 305 is also used to determine key alarm information in the target early warning information;

[0237] Module 302 is also used to build an experience replay pool corresponding to the target early warning information based on key alarm information and early warning feedback status set;

[0238] The generation module 308 is also used to perform model optimization operations on a pre-set target intelligent model based on the experience replay pool, to obtain the optimized target intelligent model, and to generate model optimization comparison information corresponding to the target intelligent model.

[0239] It is evident that implementation Figure 4 The described device can acquire warning feedback information corresponding to target warning information, construct a warning feedback state set based on the warning feedback information, determine key alarm information in the target warning information, and construct an experience replay pool corresponding to the target warning information based on the key alarm information and the warning feedback state set. Based on the experience replay pool, it performs model optimization operations on a pre-set target intelligent model to obtain an optimized target intelligent model and generates model optimization comparison information corresponding to the target intelligent model. It can dynamically adjust the warning strategy through the acquired warning feedback information and, through a reinforcement learning mechanism based on the experience replay pool, enable the model to learn better decision-making strategies from historical data, further improving the accuracy and timeliness of warnings. The system can dynamically adjust the warning strategy based on the information in the warning feedback state set, which is beneficial for improving the generation of target warnings. The system enhances the intelligence, accuracy, and reliability of alarm information. Through a tiered alarm strategy and feedback mechanism, the system dynamically adjusts alarm priorities based on operator feedback. By continuously optimizing the target intelligent model, the system better adapts to changes in the production process, maintaining high performance under various conditions. Based on alarm feedback status sets and key alarm information, the system flexibly adjusts alarm strategies to address different production scenarios and needs. By generating model optimization comparison information, the system continuously evaluates and optimizes its performance, ensuring high efficiency and stability during long-term operation, thereby improving system flexibility and scalability. Through multi-dimensional analysis and dynamic weight adjustment, it can more accurately identify truly abnormal data and intelligently process alarm information, improving the accuracy and reliability of alarm information processing.

[0240] In yet another alternative embodiment, such as Figure 4 As shown, the determining module 305 is also used to determine the early warning decision information based on the early warning feedback information, determine the early warning confidence level corresponding to the early warning decision information, and determine the early warning priority calibration information based on the early warning feedback information;

[0241] The acquisition module 312 is also used to acquire historical early warning information and the historical processing results corresponding to the historical early warning information;

[0242] The determination module 305 is also used to determine historical feedback information based on historical processing results;

[0243] The generation module 308 is also used to generate visualized early warning feedback information based on early warning confidence, early warning priority calibration information and historical feedback information;

[0244] Among them, the visual early warning feedback information includes one or more of the following: confidence heatmap feedback information, handling and source tracing view information, and dynamic priority calibration feedback information.

[0245] It is evident that implementation Figure 4 The described device can determine early warning decision information and the corresponding early warning confidence level based on early warning feedback information; determine early warning priority calibration information based on early warning feedback information; acquire historical early warning information and corresponding historical processing results; determine historical feedback information based on historical processing results; and generate visualized early warning feedback information based on early warning confidence level, early warning priority calibration information, and historical feedback information. Through visualized early warning feedback information, operators can intuitively understand the confidence level, priority, and historical processing status of early warnings. Dynamic priority calibration and handling traceability views help operators quickly identify key early warnings, optimize resource allocation, and improve decision-making efficiency. Dynamically adjusting early warning priority and confidence level based on historical feedback information allows the system to better adapt to changes in the production environment, improving the accuracy and reliability of the early warning system. Visualized information effectively reduces the cognitive burden on operators, improves system usability and user satisfaction, and enables intelligent processing of early warning information, which is beneficial for improving the accuracy and reliability of early warning information processing. It also improves the intelligence and ease of interaction between early warning information and operators, enhancing user convenience and comfort.

[0246] In yet another alternative embodiment, such as Figure 4 As shown, the generation module 308 performs model optimization operations on a pre-set target intelligent model based on the experience replay pool, and the specific methods for obtaining the optimized target intelligent model include:

[0247] The warning processing result is determined in the experience replay pool, and a reward signal is determined based on the warning processing result. Based on the pre-determined reward function, the parameters to be adjusted in the pre-set target intelligent model are determined, as well as the parameter adjustment factor corresponding to each parameter to be adjusted is determined.

[0248] Based on each parameter to be adjusted and the parameter adjustment factor corresponding to each parameter to be adjusted, a model adjustment operation is performed on the pre-set target intelligent model to obtain the target adjusted model;

[0249] Deploy the target adjustment model and the target intelligent model to perform parallel processing early warning operations, and determine the first reward information corresponding to the target adjustment model and the second reward information corresponding to the target intelligent model;

[0250] Perform an information comparison operation on the first reward information and the second reward information to obtain the information comparison results. Based on the information comparison results, determine the optimized target intelligent model.

[0251] It is evident that implementation Figure 4 The described device can determine the early warning processing result in the experience replay pool and determine the early warning processing result as a reward signal. Based on the reward function, it determines the parameters to be adjusted and the corresponding parameter adjustment factors in the pre-set target intelligent model. Based on each parameter to be adjusted and the corresponding parameter adjustment factor, it performs a model adjustment operation on the pre-set target intelligent model to obtain a target adjusted model. It deploys the target adjusted model and the target intelligent model to perform parallel processing early warning operation, determines the first reward information corresponding to the target adjusted model and the second reward information corresponding to the target intelligent model, and compares them to obtain information comparison results. Based on the information comparison results, it determines the optimized target intelligent model, which can dynamically adjust its own parameters using past experience. This approach enables the model to continuously learn and adapt to new situations, improving its adaptability to complex environments. Based on a pre-determined reward function for rewards and signals, the model can dynamically adjust the parameters to be adjusted and their corresponding adjustment factors. Through dynamic adjustment mechanisms, the model can optimize its decision-making strategy in a timely manner based on actual feedback. Through parallel processing and information comparison, the system can select a model version with higher overall rewards as a new benchmark, ensuring the reliability and stability of the model in practical applications. It can also continuously evaluate and select better model versions. By combining offline experience replay pools and online interaction to update model parameters, it utilizes historical data experience and can reflect changes in the current environment in a timely manner, ensuring continuous optimization of the model. By optimizing the model's decision-making strategy through reinforcement learning algorithms (such as PPO), the model can automatically adjust its behavior based on reward signals, improving its intelligence level. By optimizing the model's decision-making strategy, the system can more accurately identify key early warning information and further intelligently process the early warning information, which is beneficial to improving the accuracy and reliability of early warning information processing.

[0252] Example 4

[0253] Please see Figure 5 , Figure 5 This is a schematic diagram of another early warning information processing device for process production disclosed in an embodiment of the present invention. Figure 5 As shown, the early warning information processing device applied to process production may include:

[0254] Memory 401 storing executable program code;

[0255] Processor 402 coupled to memory 401;

[0256] The processor 402 calls the executable program code stored in the memory 401 to execute some or all of the steps in the early warning information processing applied to the process production in any of the embodiments of the present invention.

[0257] Example 5

[0258] This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute some or all of the steps in any of the early warning information processing methods for process production disclosed in Embodiment 1 of this invention.

[0259] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0260] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0261] Finally, it should be noted that the above embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing early warning information applied to industrial processes, characterized in that, The method includes: Collect real-time data corresponding to the production process, and construct a target dataset based on all the real-time data. By using a pre-set target intelligent model, an association analysis operation is performed on all target data contained in the target dataset to generate association analysis information corresponding to the target dataset. The association analysis information includes association rule information and frequent itemset information for all target data contained in the target dataset. Based on the association analysis information, a data classification operation is performed on all the target data contained in the target dataset to obtain at least one data category, wherein each data category includes at least one target data; Determine the warning priority corresponding to each data category, and based on each warning priority and the pre-set target intelligent model, perform information processing operations on the information to be processed to obtain target warning information; The step of performing a data classification operation on all the target data contained in the target dataset to obtain at least one data category includes: Based on pre-set association rules and all target data contained in the target dataset, generate data logical relationship information for all target data; Extract data description information from all the target data. Based on the data description information and the data logical relationship information, determine the data modality corresponding to each target data and the modality weight corresponding to each data modality through a pre-determined fusion classification model. The data modality includes one or more of the following: equipment operating parameter modality and equipment maintenance information modality. Based on the data modality corresponding to each target data and the modality weight corresponding to each data modality, a data classification operation is performed on all the target data to obtain at least one data category.

2. The early warning information processing method applied to process production according to claim 1, characterized in that, Before performing correlation analysis on all target data contained in the target dataset using a pre-set target intelligent model to generate correlation analysis information corresponding to the target dataset, the method further includes: Collect historical process data and equipment topology information, construct a multimodal pre-training dataset based on the historical process data and equipment topology information, and perform pre-training operations on the standby intelligent model based on the multimodal pre-training dataset to obtain the pre-trained intelligent model; Based on a pre-designed bidirectional architecture, the temporal dependencies corresponding to the historical process data are determined, and the equipment topology association features corresponding to the equipment topology information are determined based on the temporal dependencies; wherein, the pre-designed bidirectional architecture includes a bidirectional Transformer architecture; Based on the device topology association features and the pre-trained intelligent model, a target intelligent model is generated.

3. The early warning information processing method applied to process production according to claim 1, characterized in that, After acquiring the real-time data corresponding to the process, the method further includes: Perform a data detection operation on the real-time data to obtain a data detection result, and determine whether the data detection result is used to indicate that there is abnormal data in all the real-time data; When it is determined that the data detection result indicates the presence of abnormal data in all the real-time data, anomaly processing operation is performed on the abnormal data based on the pre-set forward propagation method to obtain the target processed data; Perform batch processing on the target data to obtain batch processing results, and store the batch processing results in a predetermined target database to establish a data tracking relationship. Based on the data tracking relationship, update all the real-time data and trigger the operation of constructing a target dataset based on all the real-time data. The data batch processing operation includes one or more of data alignment operations and data feature extraction operations; the predetermined target database includes one or more of time-series databases and relational databases.

4. The early warning information processing method applied to process production according to claim 2, characterized in that, The method further includes: Obtain the warning feedback information corresponding to the target warning information, and construct a warning feedback state set based on the warning feedback information, wherein the warning feedback state set includes one or more of the following: equipment operating condition warning feedback state, process warning feedback state, and warning classification strategy feedback state. The key alarm information in the target early warning information is identified, and based on the key alarm information and the early warning feedback state set, an experience replay pool corresponding to the target early warning information is constructed. Based on the experience replay pool, a model optimization operation is performed on the pre-set target intelligent model to obtain the optimized target intelligent model, and model optimization comparison information corresponding to the target intelligent model is generated.

5. The early warning information processing method applied to process production according to claim 4, characterized in that, The method further includes: Based on the warning feedback information, determine the warning decision information, determine the warning confidence level corresponding to the warning decision information, and determine the warning priority labeling information based on the warning feedback information; Obtain historical early warning information and the corresponding historical processing results, and determine historical feedback information based on the historical processing results; Based on the warning confidence level, the warning priority calibration information, and the historical feedback information, a visual warning feedback information is generated. The visual early warning feedback information includes one or more of the following: confidence heatmap feedback information, handling and source tracing view information, and dynamic priority calibration feedback information.

6. The early warning information processing method applied to process production according to claim 5, characterized in that, The step of performing model optimization operations on the pre-set target intelligent model based on the experience replay pool to obtain the optimized target intelligent model includes: The warning processing result is determined in the experience replay pool, and a reward signal is determined based on the warning processing result. Based on the predetermined reward function, the parameters to be adjusted in the predetermined target intelligent model are determined, and the parameter adjustment factor corresponding to each parameter to be adjusted is determined. Based on each parameter to be adjusted and the parameter adjustment factor corresponding to each parameter to be adjusted, a model adjustment operation is performed on the pre-set target intelligent model to obtain the target adjusted model; The target adjustment model and the target intelligent model are deployed to perform parallel processing early warning operations, and the first reward information corresponding to the target adjustment model and the second reward information corresponding to the target intelligent model are determined. An information comparison operation is performed on the first reward information and the second reward information to obtain the information comparison result. Based on the information comparison result, the optimized target intelligent model is determined.

7. A warning information processing device applied to process production, characterized in that, The device includes: The data acquisition module is used to collect real-time data corresponding to the production process. A building module is used to construct the target dataset based on all the aforementioned real-time data; The analysis module is used to perform association analysis operations on all target data contained in the target dataset through a pre-set target intelligent model, and generate association analysis information corresponding to the target dataset. The association analysis information includes association rule information and frequent itemset information of all target data contained in the target dataset. The classification module is used to perform a data classification operation on all the target data contained in the target dataset based on the association analysis information, to obtain at least one data category, wherein each data category includes at least one target data; A determination module is used to determine the warning priority corresponding to each of the data categories; The processing module is used to perform information processing operations on the information to be processed based on each of the warning priorities and the pre-set target intelligent model, so as to obtain the target warning information; The specific methods by which the classification module performs data classification operations on all the target data contained in the target dataset to obtain at least one data category include: Based on pre-set association rules and all target data contained in the target dataset, generate data logical relationship information for all target data; Extract data description information from all the target data. Based on the data description information and the data logical relationship information, determine the data modality corresponding to each target data and the modality weight corresponding to each data modality through a pre-determined fusion classification model. The data modality includes one or more of the following: equipment operating parameter modality and equipment maintenance information modality. Based on the data modality corresponding to each target data and the modality weight corresponding to each data modality, a data classification operation is performed on all the target data to obtain at least one data category.

8. A warning information processing device applied to process production, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the early warning information processing method applied to process production as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the early warning information processing method for process production as described in any one of claims 1-6.

Images