Method, system, electronic device and storage medium for detecting production risk events

Abstract

The application relates to a production risk event detection method, system, electronic device and storage medium, wherein the production risk event detection method comprises the following steps: collecting production parameter samples and risk event samples of multiple monitoring points on a production line; adopting a correlation analysis algorithm to calculate a time lag correlation relationship between the production parameters and the risk events, and obtaining a correlation matrix representing the correlation relationship between the production parameters and the risk events; adopting a clustering analysis algorithm to classify the risk events, and obtaining the types of the risk events; according to the types of the risk events and the correlation matrix, a corresponding multi-model fusion prediction model is constructed; and according to the prediction model, the real-time collected production parameters are subjected to risk prediction, and a corresponding risk prediction result is obtained. The production parameter and the risk event can be subjected to time lag correlation analysis, the correlation matrix is obtained, and according to the correlation matrix, a corresponding multi-model fusion prediction model is constructed, so that the accuracy of production risk event detection is improved.

Description

Technical Field

[0001] This application relates to the field of industrial safety production and risk management technology, and in particular to detection methods, systems, electronic devices and storage media for production risk events. Background Technology

[0002] In modern industrial production environments, a large number of sensors are deployed at various stages of the production line, generating a large amount of real-time data. This data contains rich information on the production operation status, and by analyzing this data, production risk events can be predicted during the production process.

[0003] Existing technologies employ static analysis. This approach fails to consider the dynamic changes in production parameters, and there is typically a time lag between these changes and the occurrence of risk events, resulting in low accuracy in detecting production risk events. Therefore, related technologies suffer from low accuracy in detecting production risk events.

[0004] There is currently no effective solution to the problem of low accuracy in detecting production risk events in related technologies. Summary of the Invention

[0005] This embodiment provides a method, system, electronic device, and storage medium for detecting production risk events, in order to solve the problem of low accuracy in detecting production risk events in related technologies.

[0006] Firstly, this embodiment provides a method for detecting production risk events, including:

[0007] Collect production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time series dataset;

[0008] Based on the time series dataset, the time-lag correlation between the production parameters and the risk events is calculated using an association analysis algorithm, resulting in an association matrix characterizing the correlation between the production parameters and the risk events.

[0009] Based on the correlation matrix, a clustering analysis algorithm is used to classify the risk events to obtain the types of the risk events;

[0010] Based on the type of the risk event and the correlation matrix, a corresponding multi-model fusion prediction model is constructed;

[0011] Based on the prediction model, risk prediction is performed on the real-time collected production parameters to obtain the corresponding risk prediction results, and a risk warning is generated based on the risk prediction results.

[0012] In some embodiments, the process of collecting production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time-series dataset includes:

[0013] The production parameter sample and the risk event sample are subjected to outlier detection and processing respectively to obtain the first production parameter sample and the first risk event sample;

[0014] The first production parameter sample is normalized to obtain the second production parameter sample;

[0015] Based on the timestamps, the second production parameter sample and the first risk event sample are time-aligned to obtain the time series dataset; the time series dataset includes the second production parameter sample and the first risk event sample.

[0016] In some embodiments, the step of calculating the time-lag correlation between the production parameters and the risk events using an association analysis algorithm based on the time series dataset to obtain an association matrix characterizing the correlation between the production parameters and the risk events includes:

[0017] The time series dataset is segmented using overlapping sliding time windows to obtain the time series dataset within each time window;

[0018] Based on the time series datasets within each time window, the correlation analysis algorithm is used to calculate the time-lag correlation between the production parameters and the risk events, thereby obtaining the correlation matrix.

[0019] In some embodiments, classifying the risk events using a clustering analysis algorithm based on the correlation matrix to obtain the types of the risk events includes:

[0020] Based on the correlation matrix, the feature vector of the risk event is obtained;

[0021] Based on the feature vectors, a similarity algorithm is used to calculate the similarity between the feature vectors of each risk event;

[0022] Based on the similarity, the clustering analysis algorithm is used to classify the risk events to obtain the types of the risk events.

[0023] In some embodiments, constructing a corresponding multi-model fusion prediction model based on the type of the risk event and the correlation matrix includes:

[0024] Based on the type of the risk event and the correlation matrix, construct multiple corresponding candidate models;

[0025] Each candidate model is evaluated based on a multi-dimensional evaluation index to obtain an evaluation score for each candidate model.

[0026] Based on the evaluation scores, multiple sub-models are selected from the multiple candidate models, and the multiple sub-models are constructed into the corresponding prediction model.

[0027] In some embodiments, the step of performing risk prediction on real-time collected production parameters based on the prediction model to obtain corresponding risk prediction results includes:

[0028] An overlapping sliding time window is used to extract features from the real-time acquired production parameters to obtain the feature vector of the real-time acquired production parameters.

[0029] Based on the feature vector of the real-time collected production parameters and the prediction model, the corresponding risk prediction result is obtained.

[0030] In some embodiments, obtaining the corresponding risk prediction result based on the feature vector of the real-time collected production parameters and the prediction model includes:

[0031] Based on the feature vectors of the real-time collected production parameters and the prediction model, the prediction results of each sub-model in the prediction model are obtained;

[0032] Based on the feature vector of the real-time collected production parameters and the prediction model, the weights of each sub-model in the prediction model are updated to obtain the updated weights of each sub-model.

[0033] Based on the prediction results of each sub-model and the updated weights of each sub-model, the corresponding risk prediction results are obtained.

[0034] Secondly, this embodiment provides a detection system for production risk events, including a data acquisition module, a time-delay correlation analysis module, a risk classification module, a prediction model construction module, and a risk prediction module, wherein:

[0035] The data acquisition module is used to collect production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time series dataset;

[0036] The time-delay correlation analysis module is used to calculate the time-delay correlation between the production parameters and the risk events based on the time series dataset using a correlation analysis algorithm, and to obtain a correlation matrix characterizing the correlation between the production parameters and the risk events.

[0037] The risk classification module is used to classify the risk events according to the correlation matrix using a clustering analysis algorithm to obtain the type of the risk event;

[0038] The prediction model building module is used to build a corresponding multi-model fusion prediction model based on the type of the risk event and the correlation matrix.

[0039] The risk prediction module is used to perform risk prediction on the real-time collected production parameters according to the prediction model, obtain the corresponding risk prediction results, and generate risk warnings based on the risk prediction results.

[0040] Thirdly, this embodiment provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the production risk event detection method described in the first aspect above.

[0041] Fourthly, this embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the production risk event detection method described in the first aspect above.

[0042] Compared with related technologies, the production risk event detection method provided in this embodiment constructs a time-series dataset by collecting production parameter samples and risk event samples from multiple monitoring points on the production line. Based on the time-series dataset, an association analysis algorithm is used to calculate the time-lag correlation between production parameters and risk events, obtaining an association matrix characterizing the correlation between production parameters and risk events. Based on the association matrix, a clustering analysis algorithm is used to classify risk events, obtaining their types. Based on the risk event types and the association matrix, a corresponding multi-model fusion prediction model is constructed. Based on the prediction model, risk prediction is performed on the real-time collected production parameters, obtaining corresponding risk prediction results, and a risk warning is generated based on the risk prediction results. This method can perform time-lag correlation analysis on production parameters and risk events, obtain an association matrix characterizing the correlation between production parameters and risk events, and construct a corresponding multi-model fusion prediction model based on this association matrix, thereby improving the accuracy of production risk event detection.

[0043] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0045] Figure 1 This is a hardware structure block diagram of the terminal of the production risk event detection method in this embodiment;

[0046] Figure 2 This is a flowchart of the production risk event detection method in this embodiment;

[0047] Figure 3 This is a flowchart of a method for detecting production risk events in some embodiments;

[0048] Figure 4 This is a flowchart of a method for detecting production risk events in some embodiments;

[0049] Figure 5 This is a structural block diagram of the production risk event detection system in this embodiment. Detailed Implementation

[0050] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0051] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this application do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific order of objects.

[0052] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. For example, it can run on a terminal. Figure 1 This is a hardware structure block diagram of the terminal for the production risk event detection method of this embodiment. For example... Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 and a memory 104 for storing data are also included. The processor 102 may be, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA). The terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that… Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown are illustrated.

[0053] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the production risk event detection method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the aforementioned method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0054] The transmission device 106 is used to receive or send data via a network. This network includes a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 can be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0055] This embodiment provides a method for detecting production risk events. Figure 2 This is a flowchart of the production risk event detection method in this embodiment, as follows: Figure 2 As shown, the process includes the following steps:

[0056] Step S201: Collect production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time series dataset.

[0057] Specifically, the production parameter sample includes equipment operating parameters, process parameters, environmental parameters, and personnel operation records; the risk event sample includes event type, occurrence time, scope of impact, and severity. Production parameter samples and risk event samples from multiple monitoring points on the production line are collected, and these samples are then constructed into a structured time-series dataset of production parameters and risk events in chronological order.

[0058] Step S202: Based on the time series dataset, use the association analysis algorithm to calculate the time-lag correlation between production parameters and risk events, and obtain the association matrix representing the correlation between production parameters and risk events.

[0059] Specifically, the association analysis algorithm includes three complementary association analysis methods: Pearson correlation coefficient, Spearman rank correlation coefficient, and maximum information coefficient. Based on the time series dataset, this association analysis algorithm is used to calculate the time-lag correlation between production parameters and risk events, obtaining an association matrix characterizing the correlation between production parameters and risk events.

[0060] Step S203: Based on the correlation matrix, a clustering analysis algorithm is used to classify the risk events and obtain the types of risk events.

[0061] Specifically, the clustering analysis algorithm is an improved density peak clustering algorithm. Based on the association matrix, feature vectors of risk events are obtained. Based on these feature vectors, a similarity algorithm is used to calculate the similarity between the feature vectors of each risk event. Based on the similarity, a clustering analysis algorithm is used to classify the risk events, resulting in their types. These risk event types can include cigarette manufacturing quality risks (including cigarette tip breakage, poor filter tip assembly, etc.), tobacco processing quality risks (including abnormal moisture content, fluctuations in shred width, etc.), equipment failure risks, and packaging quality risks (including poor sealing, printing misalignment, etc.).

[0062] Step S204: Based on the type of risk event and the correlation matrix, construct the corresponding multi-model fusion prediction model.

[0063] Specifically, based on the type of risk event and the correlation matrix, multiple candidate models are constructed. Each candidate model is evaluated using multi-dimensional evaluation metrics to obtain an evaluation score. Based on the evaluation scores, multiple sub-models are selected from the candidate models, and these sub-models are used to construct the corresponding prediction model.

[0064] Step S205: Based on the prediction model, perform risk prediction on the real-time collected production parameters, obtain the corresponding risk prediction results, and generate a risk warning based on the risk prediction results.

[0065] Specifically, an overlapping sliding time window is used to extract features from the real-time collected production parameters, resulting in feature vectors for these parameters. Based on these feature vectors and the prediction model, corresponding risk prediction results are obtained. Risk warnings are then generated based on the scope and severity of the risk events identified in these predictions.

[0066] In this embodiment, a time-series dataset is constructed by collecting production parameter samples and risk event samples from multiple monitoring points on the production line. Based on the time-series dataset, an association analysis algorithm is used to calculate the time-lag correlation between production parameters and risk events, obtaining an association matrix characterizing the correlation between production parameters and risk events. Based on the association matrix, a clustering analysis algorithm is used to classify risk events, obtaining the types of risk events. Based on the types of risk events and the association matrix, a corresponding multi-model fusion prediction model is constructed. Based on the prediction model, risk prediction is performed on the real-time collected production parameters, obtaining the corresponding risk prediction results, and a risk warning is generated based on the risk prediction results. This method can perform time-lag association analysis on production parameters and risk events, obtain an association matrix characterizing the correlation between production parameters and risk events, and construct a corresponding multi-model fusion prediction model based on this association matrix, thereby improving the accuracy of production risk event detection.

[0067] In some embodiments, production parameter samples and risk event samples from multiple monitoring points on the production line are collected to construct a time series dataset, including: performing outlier detection and processing on the production parameter samples and risk event samples respectively to obtain a first production parameter sample and a first risk event sample; normalizing the first production parameter sample to obtain a second production parameter sample; and performing time alignment processing on the second production parameter sample and the first risk event sample according to the timestamp to obtain a time series dataset; the time series dataset includes the second production parameter sample and the first risk event sample.

[0068] Specifically, the production parameter sample includes equipment operating parameters, process parameters, environmental parameters, and personnel operation records; the risk event sample includes event type, occurrence time, scope of impact, and severity. Outliers in the production parameter and risk event samples are identified using the interquartile range method. Missing values ​​are filled using moving median or Lagrange interpolation to obtain the first production parameter and first risk event samples. The first production parameter sample is normalized using the minimum-maximum standardization method to obtain the second production parameter sample. Based on timestamps, the second production parameter and first risk event samples are time-aligned to obtain a time-series dataset. This time-series dataset includes the second production parameter and first risk event samples.

[0069] For example, production parameter samples and risk event samples were collected from a tobacco company over a two-year period. The production parameter samples included parameters for the tobacco processing stage (e.g., cutting width, feeding temperature, moisture content), parameters for the cigarette rolling and splicing stage (e.g., cigarette density, loading speed, filter pressure drop), and equipment operating parameters (e.g., current, vibration, temperature); risk event samples included parameters for quality risks, equipment failures, and production efficiency risks. The collection frequency varied according to the importance of each process: 10 seconds / sample for the tobacco processing stage, 1 second / sample for the rolling and splicing stage, and 0.5 seconds / sample for the packaging stage. Outliers in the production parameter and risk event samples were identified using the interquartile range method. Missing values ​​were filled using moving median or Lagrange interpolation to obtain the first production parameter sample and the first risk event sample. The first production parameter sample was normalized using the minimum-maximum standardization method to obtain the second production parameter sample. The second production parameter sample and the first risk event sample were then time-aligned based on timestamps to obtain a time-series dataset.

[0070] In some of these embodiments, a correlation analysis algorithm is used to calculate the time-lag correlation between production parameters and risk events based on a time series dataset, thereby obtaining a correlation matrix that characterizes the correlation between production parameters and risk events. Figure 3 This is a flowchart of a production risk event detection method in some embodiments, such as Figure 3 As shown, the process includes the following steps:

[0071] Step S301: Using overlapping sliding time windows, the time series dataset is segmented to obtain the time series dataset within each time window;

[0072] Step S302: Based on the time series datasets within each time window, use the correlation analysis algorithm to calculate the time lag correlation between production parameters and risk events, and obtain the correlation matrix.

[0073] Specifically, the size of the time window can be set according to production characteristics, and the sliding step size of the time window can be dynamically adjusted. This sliding time window is used to capture the dynamic changes of production parameters in the time dimension. Overlapping sliding time windows are used to segment the time series dataset, obtaining time series datasets within each time window. The association analysis algorithm includes three complementary association analysis methods: Pearson correlation coefficient, Spearman rank correlation coefficient, and maximum information coefficient. Within each sliding time window, the correlation between production parameters and risk events is calculated using the three methods respectively. By shifting the production parameter sequences forward and backward, production parameter sequences with different time delays are constructed. The correlation coefficient between production parameters and risk events under each delay is calculated, and the time lag value corresponding to the maximum correlation coefficient is identified. Determine the optimal time lag relationship between changes in production parameters and the occurrence of risk events, specifically including: setting the maximum time lag range. ,exist Within the interval, with a fixed step size Values; for each production parameter sequence X and risk event sequence Y, calculate X at each time delay. The three correlation coefficients between Y and Y; for each correlation coefficient, find the maximum value and the corresponding time lag. Based on the results of the three correlation methods, the final optimal time-delay relationship is determined. The correlation degrees between each production parameter and risk event under different time-delay conditions are integrated into a three-dimensional correlation matrix. Where i represents production parameters and j represents the type of risk event. The value represents the time delay; the unit value of this three-dimensional correlation matrix is ​​the correlation strength, and the range of values ​​for the Pearson correlation coefficient and the Spearman rank correlation coefficient is [-1,1], while the range of values ​​for the maximum information coefficient is [0,1].

[0074] For example, a sliding time window of 15 minutes with a step size of 1 minute is used to segment the time series dataset to capture the dynamic changes in parameters. Within each time window, the correlation between production parameters and risk events is calculated using the Pearson correlation coefficient, Spearman rank correlation coefficient, and maximum information coefficient. The correlations between production parameters and risk events under different time lags are then integrated into a three-dimensional correlation matrix. The system extracts a set of key parameters whose correlation exceeds a threshold. For example, for the risk of abnormal cigarette density, the key parameters identified include tobacco moisture content, cigarette machine pressure, cigarette paper tension, and ambient humidity.

[0075] In some embodiments, risk events are classified using a clustering analysis algorithm based on the association matrix to obtain the types of risk events, including: obtaining feature vectors of risk events based on the association matrix; calculating the similarity between feature vectors of each risk event using a similarity algorithm based on the feature vectors; and classifying risk events using a clustering analysis algorithm based on the similarity to obtain the types of risk events.

[0076] Specifically, extract the correlation matrix. For each type of risk event j, construct a feature vector for that type of risk event from the data associated with that risk event j. This feature vector includes key parameters, correlation strength, time lag information, and parameter importance ranking. Key parameters refer to correlations exceeding a certain threshold. The parameter index set is defined as follows: correlation strength refers to the numerical correlation between each key parameter and the risk event type; time lag information refers to the optimal time lag between each key parameter and the risk event type; and parameter importance ranking refers to the parameter priority arranged in descending order of correlation strength. A similarity algorithm is used to calculate the similarity between the feature vectors of each risk event. This similarity algorithm can be cosine similarity or Euclidean distance similarity. The formula for calculating cosine similarity is:

[0077] ;

[0078] The formula for calculating Euclidean distance similarity is:

[0079] ;

[0080] The formula for calculating the overall similarity is:

[0081] ;

[0082] in , where is the weighting coefficient, used to balance the importance of directional similarity and distance similarity. An improved density peak clustering algorithm is used to cluster risk events, and the risk event types are determined based on the clustering results to establish a risk event type library. Specifically, the improved density peak clustering algorithm includes: for each risk event sample i, calculating its local density, which is expressed as:

[0083] ;

[0084] in, This is an indicator function, which is 1 when x > 0, and 0 otherwise; To truncate the distance, we take the top k% of similarity values ​​from all sample pairs in descending order. For each risk event sample i, we calculate the minimum distance to any higher-density sample, which is expressed as:

[0085] ;

[0086] For the sample with the highest local density, take ; Calculate the cluster center index for each risk event sample, which is expressed as:

[0087] ;

[0088] risk event samples are categorized as follows: Sort values ​​in descending order and construct Value distribution curve, calculation First-order difference sequence of values Find The locations where the absolute value of a cluster exceeds a threshold T are designated as steep descent points, and the number of samples corresponding to these steep descent points is determined as the optimal number of cluster centers. This is achieved by introducing... Value distribution curves and steep drop point detection methods, using first-order difference sequences The analysis automatically determines the optimal number of cluster centers. Non-centrally located risk event samples are assigned to the clusters belonging to the nearest high-density samples, forming the clustering results and thus determining the type of risk event. This improved density peak clustering algorithm can automatically adjust the clustering results after new data is added, thereby identifying the types of newly emerging risk events.

[0089] For example, feature vectors of risk events are constructed based on the association matrix, and the similarity between the feature vectors of each risk event is calculated using a combination of cosine similarity and Euclidean distance. An improved density peak clustering algorithm is used to automatically classify the risk events, yielding clustering results. Based on the clustering results, various risk events are categorized into four main risk types: cigarette manufacturing quality risk (including cigarette breakage, poor filter splicing, etc.), tobacco processing quality risk (including abnormal moisture content, fluctuations in shred width, etc.), equipment failure risk, and packaging quality risk (including poor sealing, printing misalignment, etc.).

[0090] In some of these embodiments, a corresponding multi-model fusion prediction model is constructed based on the type of risk event and the correlation matrix. Figure 4 This is a flowchart of a production risk event detection method in some embodiments, such as Figure 4 As shown, the process includes the following steps:

[0091] Step S401: Based on the type of risk event and the correlation matrix, construct multiple corresponding candidate models;

[0092] Step S402: Evaluate each candidate model according to the multi-dimensional evaluation indicators to obtain the evaluation score of each candidate model;

[0093] Step S403: Based on the evaluation scores, select multiple sub-models from multiple candidate models and construct the multiple sub-models into corresponding prediction models.

[0094] Specifically, the candidate models include autoregressive integral moving average (ARMA) models, exponential smoothing models, random forest models, gradient boosting tree models, long short-term memory (LSTM) network models, gated recurrent unit (GRU) network models, bidirectional LSM network models, and temporal convolutional network models. Multiple candidate models are constructed based on the type of risk event and the correlation matrix. Based on the correlation matrix, for each type of risk event, the set of parameters with correlation exceeding a threshold is extracted and sorted in descending order of correlation strength. This identifies the set of key parameters that have a significant impact on each type of risk event, and their influence weights are calculated. For each type of risk event, based on the key parameters and their time-lag characteristics, a feature combination scheme is constructed, including: feature combinations sorted by correlation, selecting the top k parameters by correlation; and feature combinations filtered by a threshold, selecting parameters with correlation exceeding a threshold. All parameters; feature combinations based on time lag grouping, grouping parameters according to different time lag ranges, and constructing a feature set for each group separately. All three feature combinations construct time series features, which include raw values, moving statistics, and rate of change features, and selectively add specific advanced features according to the type of risk event.

[0095] For each feature combination and each type of risk event, train each candidate model. Specifically, this may include: dividing historical data into training and validation sets in a 7:3 ratio to ensure a balanced distribution of risk event types in the two sets; automatically searching for the optimal hyperparameter combination using Bayesian optimization; evaluating the performance of the candidate model using five-fold cross-validation; and training the candidate model using the optimized hyperparameters and the complete training set to obtain the candidate model corresponding to each type of risk event.

[0096] A multi-dimensional evaluation index was used to comprehensively evaluate each candidate model, and an evaluation score was obtained for each candidate model. Based on the evaluation scores, each candidate model for each type of risk event is ranked, and the best-performing candidate model is added to the final model set. This multi-dimensional evaluation includes prediction accuracy metrics such as F1 score, AUC, precision, and recall; time-related metrics such as precision under different prediction lead times and prediction time window precision; stability metrics such as variance of continuous prediction results and sensitivity to input disturbances; and computational efficiency metrics such as prediction time complexity and resource consumption. Based on the evaluation scores of the selected candidate models, an initial weight is assigned to each selected model, expressed as follows:

[0097] ;

[0098] Where j represents each type of risk event, n represents each candidate model, and f represents each feature combination. The sum of the weights of each selected model is 1, thus obtaining the multi-model fusion prediction model corresponding to each type of risk event.

[0099] For example, a model ensemble incorporating multiple algorithms is constructed for the four identified risk types. For instance, for convolution quality risk events, an autoregressive integral moving average model, a gradient boosting tree model, and a long short-term memory network model are used. Three feature combination schemes are configured for each risk type: selecting the top ten parameters based on correlation ranking, selecting parameters with a correlation greater than 0.7 based on a threshold, and grouping parameters into short, medium, and long-term groups based on time lag. For each feature combination, time-series features are constructed, including raw values, moving average statistics, and high-level features. Each model is trained using a 7:3 ratio between the training and validation sets, and Bayesian optimization is used to automatically search for optimal hyperparameters. Five-fold cross-validation is used to evaluate the model performance. Multi-dimensional evaluation metrics are used to comprehensively evaluate each candidate model, resulting in an evaluation score. For example, the gradient boosting tree model for convolution quality risk performs best in terms of F1 score, AUC, computational efficiency, and stability, while the long short-term memory network model performs better in terms of early warning time. Initial weights are assigned to each selected model based on the comprehensive evaluation score. For example, the weights of the gradient boosting tree model, long short-term memory network model, and autoregressive integral moving average model for splicing quality risk are 0.42, 0.38, and 0.20, respectively.

[0100] In some of these embodiments, risk prediction is performed on the real-time collected production parameters according to the prediction model to obtain the corresponding risk prediction results. This includes: using overlapping sliding time windows to extract features from the real-time collected production parameters to obtain feature vectors of the real-time collected production parameters; and obtaining the corresponding risk prediction results based on the feature vectors of the real-time collected production parameters and the prediction model.

[0101] Specifically, using the same sampling frequency as the training data, current production parameter data is collected in real time to form a real-time monitoring data stream. The same data preprocessing method as the training data is used to preprocess the current production parameter data. Using the same sliding time window size and step size as the training data, features are extracted from the real-time collected production parameters to obtain feature vectors. These feature vectors include basic statistical features such as mean, standard deviation, maximum, and minimum values; time-series fluctuation features such as moving average deviation, volatility, and trend strength; multi-parameter interaction features such as parameter ratios (calculated as new features for physically or technologically related parameter pairs), parameter differences (differences between key parameter pairs), and parameter products (calculated as interaction features for parameters with interactive effects); and time lag features, parameter change rates, and cumulative change features. Based on the feature vectors of the real-time collected production parameters and the prediction model, the corresponding risk prediction results are obtained.

[0102] In some embodiments, the corresponding risk prediction result is obtained based on the feature vector of the real-time collected production parameters and the prediction model, including: obtaining the prediction result of each sub-model in the prediction model based on the feature vector of the real-time collected production parameters and the prediction model; updating the weights of each sub-model in the prediction model based on the feature vector of the real-time collected production parameters and the prediction model to obtain the updated weights of each sub-model; and obtaining the corresponding risk prediction result based on the prediction result of each sub-model and the updated weights of each sub-model.

[0103] Specifically, based on the feature vectors of the production parameters collected in real time and the prediction model, the prediction results of each sub-model in the prediction model are obtained. Based on the similarity between the feature vectors of the real-time collected production parameters and the feature vectors of the training data of each sub-model, the weights of each sub-model in the prediction model are updated to obtain the updated weights of each sub-model. The updated weights of each sub-model are expressed as follows:

[0104] ;

[0105] in This represents the similarity between the feature vectors of the real-time collected production parameters and the feature vectors of the training data of each sub-model. The updated weights of each sub-model are then normalized to obtain the final weights of each sub-model. These final weights of each sub-model are expressed as follows:

[0106] ;

[0107] Based on the prediction results of each sub-model and the updated weights of each sub-model, the corresponding risk prediction result is obtained. This risk prediction result is expressed as:

[0108] ;

[0109] Based on the scope and severity of the risk events predicted, the risk prediction results are converted into risk levels. Then, based on these risk levels and the time urgency of the prediction, tiered risk warning information is generated.

[0110] For example, production parameter data is collected in real time, and features are extracted using the same 15-minute sliding time window size as the training data. These features are then input into the corresponding prediction models for prediction. The prediction process is executed once every time interval. For a seam quality risk event, the three models give the following probabilities of occurrence within the next 20 minutes: Gradient Boosting Tree model predicts a probability of 0.78, Long Short-Term Memory Network model predicts a probability of 0.82, and Autoregressive Integral Moving Average model predicts a probability of 0.65. The similarity between the current seam process feature vector and historical data is calculated: 0.86 with the Long Short-Term Memory Network model training data, 0.74 with the Gradient Boosting Tree model training data, and 0.58 with the Autoregressive Integral Moving Average model training data. The weights are dynamically adjusted based on the similarity: the new weight for the Gradient Boosting Tree model is 0.31, the new weight for the Long Short-Term Memory Network model is 0.33, and the new weight for the Autoregressive Integral Moving Average model is 0.12. After normalization adjustment, the final weights of the gradient boosting tree model are 0.41, the long short-term memory network model is 0.43, and the autoregressive integral moving average model is 0.16. The weighted fusion prediction results show a probability of 0.78 for convolution quality risk. Combining the impact range and severity of the risk event, the prediction results are converted into risk levels. For convolution quality risk, a low-risk threshold of 0.5, a medium-risk threshold of 0.7, and a high-risk threshold of 0.85 are pre-set. The current predicted probability of 0.78 falls within the medium-risk range, and a corresponding risk warning is issued.

[0111] This embodiment also provides a detection system for production risk events, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. The terms "module," "unit," "subunit," etc., used below refer to combinations of software and / or hardware that perform a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0112] Figure 5 This is a structural block diagram of the production risk event detection system in this embodiment, as shown below. Figure 5 As shown, the production risk event detection system 50 includes: a data acquisition module 501, a time-delay correlation analysis module 502, a risk classification module 503, a prediction model construction module 504, and a risk prediction module 505, wherein:

[0113] The data acquisition module 501 is used to collect production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time series dataset.

[0114] The time-delay correlation analysis module 502 is used to calculate the time-delay correlation between production parameters and risk events based on the time series dataset using a correlation analysis algorithm, and obtain a correlation matrix that characterizes the correlation between production parameters and risk events.

[0115] The risk classification module 503 is used to classify risk events based on the correlation matrix using a clustering analysis algorithm to obtain the type of risk event.

[0116] The prediction model building module 504 is used to build a corresponding multi-model fusion prediction model based on the type of risk event and the correlation matrix.

[0117] The risk prediction module 505 is used to predict the risks of real-time collected production parameters based on the prediction model, obtain the corresponding risk prediction results, and generate risk warnings based on the risk prediction results.

[0118] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0119] This embodiment also provides an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0120] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0121] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated in this embodiment.

[0122] Furthermore, in conjunction with the production risk event detection methods provided in the above embodiments, this embodiment can also provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the production risk event detection methods in the above embodiments.

[0123] It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. All other embodiments derived by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0124] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0125] Obviously, the accompanying drawings are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar situations based on these drawings without any creative effort. Furthermore, it is understood that although the work done in this development process may be complex and lengthy, for those skilled in the art, certain design, manufacturing, or production modifications made based on the technical content disclosed in this application are merely conventional technical means and should not be considered as insufficient disclosure of this application.

[0126] The term "embodiment" in this application refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply that it is mutually exclusive with or independent of other embodiments. It will be clearly or implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0127] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of patent protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for detecting production risk events, characterized in that, include: Collect production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time series dataset; Based on the time series dataset, the time-lag correlation between the production parameters and the risk events is calculated using an association analysis algorithm, resulting in an association matrix characterizing the correlation between the production parameters and the risk events. Based on the correlation matrix, a clustering analysis algorithm is used to classify the risk events to obtain the types of the risk events; Based on the type of the risk event and the correlation matrix, a corresponding multi-model fusion prediction model is constructed; Based on the prediction model, risk prediction is performed on the real-time collected production parameters to obtain the corresponding risk prediction results, and a risk warning is generated based on the risk prediction results.

2. The method for detecting production risk events according to claim 1, characterized in that, The process involves collecting production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time-series dataset, including: Outlier detection and processing are performed on the production parameter sample and the risk event sample respectively to obtain the first production parameter sample and the first risk event sample; The first production parameter sample is normalized to obtain the second production parameter sample; Based on the timestamps, the second production parameter sample and the first risk event sample are time-aligned to obtain the time series dataset; the time series dataset includes the second production parameter sample and the first risk event sample.

3. The method for detecting production risk events according to claim 1, characterized in that, The step of calculating the time-lag correlation between the production parameters and the risk events using a correlation analysis algorithm based on the time series dataset, and obtaining a correlation matrix characterizing the correlation between the production parameters and the risk events, includes: The time series dataset is segmented using overlapping sliding time windows to obtain the time series dataset within each time window; Based on the time series datasets within each time window, the correlation analysis algorithm is used to calculate the time-lag correlation between the production parameters and the risk events, thereby obtaining the correlation matrix.

4. The method for detecting production risk events according to claim 1, characterized in that, The risk events are classified using a clustering analysis algorithm based on the correlation matrix to obtain the types of the risk events, including: Based on the correlation matrix, the feature vector of the risk event is obtained; Based on the feature vectors, a similarity algorithm is used to calculate the similarity between the feature vectors of each risk event; Based on the similarity, the clustering analysis algorithm is used to classify the risk events to obtain the types of the risk events.

5. The method for detecting production risk events according to claim 1, characterized in that, The step of constructing a corresponding multi-model fusion prediction model based on the type of the risk event and the correlation matrix includes: Based on the type of the risk event and the correlation matrix, construct multiple corresponding candidate models; Each candidate model is evaluated based on a multi-dimensional evaluation index to obtain an evaluation score for each candidate model. Based on the evaluation scores, multiple sub-models are selected from the multiple candidate models, and the multiple sub-models are constructed into the corresponding prediction model.

6. The method for detecting production risk events according to claim 1, characterized in that, The step of performing risk prediction on the real-time collected production parameters based on the prediction model to obtain the corresponding risk prediction results includes: An overlapping sliding time window is used to extract features from the real-time acquired production parameters to obtain the feature vector of the real-time acquired production parameters. Based on the feature vectors of the real-time collected production parameters and the prediction model, the corresponding risk prediction results are obtained.

7. The method for detecting production risk events according to claim 6, characterized in that, The step of obtaining the corresponding risk prediction result based on the feature vector of the real-time collected production parameters and the prediction model includes: Based on the feature vectors of the real-time collected production parameters and the prediction model, the prediction results of each sub-model in the prediction model are obtained; Based on the feature vector of the real-time collected production parameters and the prediction model, the weights of each sub-model in the prediction model are updated to obtain the updated weights of each sub-model. Based on the prediction results of each sub-model and the updated weights of each sub-model, the corresponding risk prediction results are obtained.

8. A detection system for production risk events, characterized in that, It includes a data acquisition module, a time-delay correlation analysis module, a risk classification module, a prediction model construction module, and a risk prediction module, among which: The data acquisition module is used to collect production parameter samples and risk event samples from multiple monitoring points on the production line to construct a time series dataset; The time-delay correlation analysis module is used to calculate the time-delay correlation between the production parameters and the risk events based on the time series dataset using a correlation analysis algorithm, and to obtain a correlation matrix characterizing the correlation between the production parameters and the risk events. The risk classification module is used to classify the risk events according to the correlation matrix using a clustering analysis algorithm to obtain the type of the risk event; The prediction model building module is used to build a corresponding multi-model fusion prediction model based on the type of the risk event and the correlation matrix. The risk prediction module is used to perform risk prediction on the real-time collected production parameters according to the prediction model, obtain the corresponding risk prediction results, and generate risk warnings based on the risk prediction results.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method for detecting production risk events according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the production risk event detection method according to claims 1 to 7.

Images