Real-time visual monitoring method and system for metering production business circulation state

Abstract

The invention relates to the technical field of production process monitoring and data processing, and discloses a real-time visual monitoring method and system for metering a production business flow state, and the method comprises the steps: collecting multi-process node heterogeneous data, and obtaining a complete integrated data set through standardized protocol conversion, timestamp synchronous alignment and dynamic time warping; detecting circulation state change by adopting a hidden Markov model and other time sequence analysis algorithms to obtain a state conversion sequence record; extracting a real-time index from the sequence, and determining the position of an abnormal circulation point when the change of the index exceeds a preset threshold value; a visual chart is generated in combination with the state health degree model, and a whole-process monitoring view is obtained; and a closed-loop feedback mechanism is constructed. According to the method, the production state can be accurately identified, circulation abnormity can be accurately positioned, transparent and intelligent monitoring of the whole production process is realized through a visual visual view and self-adaptive closed-loop feedback, and the state updating accuracy and the system operation efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of production process monitoring and data processing, in particular to a metering production business flow state real-time visualization monitoring method and system. BACKGROUND

[0002] In the modern precision manufacturing field such as metering production, in order to ensure product quality and production efficiency, sensors and manufacturing execution systems are usually deployed at each process node of the production line to collect and monitor various data in real time. Through the analysis of these data, enterprises expect to understand the production state in time, find potential problems, and make scientific decisions.

[0003] However, the existing technology still faces challenges in practical application. The data from different process nodes are usually heterogeneous, with different formats, protocols and physical dimensions, which brings great difficulty to building a unified and coherent data view, resulting in fragmented data. More importantly, due to the clock difference and network delay of different collection devices, the timestamps of these heterogeneous data are often inconsistent. This time misalignment seriously interferes with the analysis of the causal relationship of cross-process events and may lead to misjudgment of the production flow state.

[0004] In addition, most current monitoring methods focus on monitoring the threshold value of a single physical quantity and triggering an alarm when a parameter exceeds the preset range. This method can detect obvious single-point failures, but it is difficult to capture more subtle process abnormalities under the coupling of multiple parameters. It lacks deep modeling and analysis of the "state" and "conversion process" of business flow, so it cannot effectively identify dynamic flow problems such as production rhythm disorder and process connection difficulty, and it is also difficult to provide managers with a global and intuitive whole-process monitoring view. SUMMARY

[0005] In view of the shortcomings of the prior art, the present application provides a metering production business flow state real-time visualization monitoring method and system, which solves the problem of inaccurate production flow state analysis and untimely abnormal positioning caused by the difficulty of multi-source heterogeneous data integration and time alignment in the prior art.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: the present application provides a metering production business flow state real-time visualization monitoring method, which comprises a series of steps such as data collection and standardization, data clustering and association grouping, timestamp synchronization and consistency judgment, time series analysis and state conversion detection, abnormality detection and notification triggering, visualization and dynamic tracking, and feedback optimization and parameter adjustment, to form a closed-loop monitoring and optimization process.

[0007] The method first collects heterogeneous data including processing state and quality detection results through a sensor network deployed at multiple process nodes, and converts the data into a uniform format dataset using a standardized protocol. To eliminate differences in dimensions of different data, a standard score standardization process can be used, which has a mathematical expression of:

[0008] ;

[0009] wherein, is the standardized data, is the original data point, is the mean of the data set, is the standard deviation.

[0010] Subsequently, based on the uniform format dataset, a clustering algorithm is applied to group similar data types to determine data correlation groups. For example, the clustering algorithm can divide the dataset into clusters by minimizing the sum of squared errors within the cluster :

[0011] ;

[0012] wherein, is a data point, is the centroid of the cluster .

[0013] After obtaining the data correlation groups, since there may be differences in sampling times of different data sources, time stamps need to be synchronized to align all data entries. For example, the linear interpolation method can be used to calculate the data value at a certain time point :

[0014] ;

[0015] wherein, and are the data values at time points and , respectively. After alignment, a complete integrated dataset is obtained through consistency checking.

[0016] For the complete integrated dataset, a time series analysis algorithm is used to detect the flow state change of the production business. Specifically, a hidden Markov model can be used for modeling. Given an observation sequence, the Viterbi algorithm is used to solve the optimal state sequence, and the recursive relationship of the algorithm is:

[0017] ;

[0018] wherein, denotes the state at time Arrival state and the maximum probability of observing a partial sequence, is a state transition probability, is an observation probability. The optimal state sequence decoded by the algorithm is the state transition sequence record.

[0019] Then, the real-time update indicators are extracted from the state transition sequence record, and the instant notification mechanism is triggered when the indicator value exceeds the preset threshold, so as to determine the abnormal flow point position.

[0020] Through the determined abnormal flow point position, a visual chart is generated to show the dynamic tracking path. A state health degree indicator :

[0021] ;

[0022] wherein, is the normalized real-time feature of the i th standard, is the corresponding weight. According to the value of , color coding is performed to intuitively represent the state health degree in the whole process monitoring view.

[0023] Finally, according to the analysis result of the whole process monitoring view, a cyclic feedback mechanism is established. The mechanism uses a model such as support vector regression to map the state update features extracted in the monitoring view to the optimized sensor parameters, thereby constructing a parameter adjustment model. The goal of the model is to find a function whose deviation between the predicted value and the target value is minimized within a certain range. Through the parameter instructions output by the model, the sensor parameters of the data acquisition module are adjusted, and the feedback effect is judged to continuously optimize the accuracy of state update.

[0024] The second aspect of the application provides a real-time visualization monitoring system for metering production business flow state, which comprises:

[0025] A data acquisition and standardization module is used to collect heterogeneous data from multiple process nodes through a sensor network, convert different format data using a standardization protocol, and obtain a unified format data set;

[0026] A data clustering and association grouping module is used to group similar data types by applying a clustering algorithm based on the unified format data set, and determine data association groups for subsequent integration;

[0027] A timestamp synchronization and consistency judgment module is used to synchronize timestamps to align all data entries after obtaining the data association groups, and judge the consistency of the synchronized data to obtain a complete integrated data set;

[0028] ​The time sequence analysis and state transition detection module is configured to detect state transition changes by using a time sequence analysis algorithm for the complete integrated data set, and obtain state transition sequence records;

[0029] The abnormality detection and notification triggering module is configured to extract real-time update indicators from the state transition sequence records, and trigger an instant notification mechanism if the real-time update indicators exceed a preset threshold, so as to determine an abnormal flow point position;

[0030] The visualization and dynamic tracking module is configured to generate a visualization chart to show a dynamic tracking path through the abnormal flow point position, and obtain a whole-process monitoring view.

[0031] The feedback optimization and parameter adjustment module is configured to adjust sensor parameters of the data acquisition module according to the whole-process monitoring view, judge feedback effects, and continuously optimize state update accuracy.

[0032] The present application provides a kind of metrological production business flow state real-time visualization monitoring method and system.There are the following beneficial effects:

[0033] 1、The present application forms a high-quality complete integrated data set by standardizing conversion, time stamp synchronization alignment and deep consistency judgment based on dynamic time warping of multi-source heterogeneous data, and decodes based on this using hidden Markov model, which effectively solves the analysis misalignment problem caused by inconsistent data sources and time misplacement, thereby significantly improving the accuracy of metrological production business flow state recognition.

[0034] 2、The present application can instantly and accurately locate the specific abnormal flow occurrence point by extracting state duration, transition frequency and other dynamic indicators from the decoded state transition sequence records, and combining sliding window technology to perform real-time threshold judgment on the indicator change rate.Compared with the traditional abnormality detection method based on fixed threshold, the present application realizes fast and accurate locking from macroscopic abnormality perception to microscopic fault point, and enhances the effectiveness and pertinence of early warning.

[0035] 3、The present application builds a closed-loop feedback mechanism from the whole-process monitoring view to the front-end sensor parameters, automatically learns and optimizes the data acquisition strategy by using parameter adjustment models such as support vector regression, so that the whole monitoring system has the ability of self-adaptation and self-optimization, and can dynamically adjust the input and distribution of monitoring resources according to the actual health status of business flow, thereby ensuring the monitoring precision while improving the overall operation efficiency of the system. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The method flowchart of the present application;

[0037] Figure 2A system architecture diagram of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0039] Embodiment:

[0040] Please refer to the accompanying Figure 1 The embodiment of the present application provides a metering production business flow state real-time visualization monitoring method and system, which comprises:

[0041] S1: collecting heterogeneous data from multiple process nodes through a sensor network, converting different format data by using a standardized protocol to obtain a uniform format data set;

[0042] In the embodiment, to realize accurate monitoring of the metering production business flow state, firstly, the original data from the production site need to be systematically collected and preprocessed to construct a high-quality uniform format data set. The process is the basis for all subsequent analysis, detection and optimization, and the specific implementation manner is described as follows:

[0043] Firstly, the method collects various types of heterogeneous data reflecting the production process in real time through the sensor network deployed at multiple key process nodes of the production line. These data are diverse in source and format, and may include physical quantities for representing processing states and detection results for representing product quality. To solve the problem that the data formats in the original data set are different and the dimensions are not unified, the embodiment converts the data by using a standardized protocol to form a uniform format data set with unified structure and easy processing. Preferably, a commonly used data exchange format in the industry such as JSON or XML can be used as the standardized protocol. In the conversion process, unified meta-information such as process node ID, sensor ID and accurate timestamp is added to each data point.

[0044] Further, to eliminate the interference of different physical dimensions on the subsequent algorithm model, the converted data is normalized or standardized in the embodiment. Preferably, the Z-score standardization method is used, which converts the original data into a distribution with a mean of 0 and a standard deviation of 1. The mathematical expression is as follows:

[0045] ;

[0046] wherein, represents the standardized data point, representing the original data points, representing the mean value of the feature data set, and representing the standard deviation thereof. Through this step, all feature data is mapped into a unified scale space, providing a prerequisite for subsequent application of machine learning algorithms such as clustering and classification.

[0047] Considering that in actual industrial environments, data may be lost during acquisition or transmission, resulting in an incomplete data set. After generating a data set in a unified format, the embodiment first determines whether the data integrity is lower than a preset integrity threshold. If it is lower than the threshold, a data completion algorithm is started to fill in the missing values. Depending on the data characteristics, different filling strategies can be selected, for example, for data that is smooth in variation, mean or median filling can be used; for data with time sequence correlation, linear interpolation based on previous and subsequent time points or model-based regression interpolation method is preferred to generate a more reliable and complete complete data set.

[0048] After obtaining the complete data set, to automatically identify the macro running state of the process node from the massive data, the embodiment further uses a clustering algorithm to analyze the data representing the processing state. The K-means algorithm is preferably used, which aims to divide the data set into a preset number of clusters, so that the similarity of data points within the cluster is the highest, and the similarity of data points between clusters is the lowest. The objective is to minimize the sum of squared errors within all clusters, and the objective function can be expressed as:

[0049] ;

[0050] wherein, is the number of clusters, represents the th cluster, is a data point in the cluster , and is the centroid of the cluster . Through the clustering process, complex processing state data can be automatically summarized into several meaningful state classifications.

[0051] To further explore the internal relationship between the processing state and the product quality, after obtaining the state classification of the process node, the embodiment further calculates the correlation strength between the quality detection result of each process node and the processing state. The Pearson correlation coefficient is preferably used for measurement, and the calculation formula is:

[0052] ;

[0053] wherein, is the correlation coefficient, and sample points representing processing states and quality inspection results respectively, and are the mean values of the two groups of samples respectively. The value of the coefficient ranges from -1 to 1, and the closer the absolute value is to 1, the stronger the linear correlation between the two.

[0054] When it is determined that the correlation between the processing state and the quality inspection result is higher than a certain preset threshold, it indicates that the specific processing state has a significant impact on the product quality. On this basis, the present embodiment uses a supervised learning model to predict potential abnormal process nodes. Preferably, a support vector machine algorithm can be used. By using historical data for training, the processing state as input features, and whether the corresponding product is qualified as labels, the SVM aims to find an optimal decision hyperplane. For a linearly separable binary classification problem, the optimization objective can be formalized as:

[0055] ;

[0056] where, is the normal vector of the hyperplane, is the bias term, is the input feature vector, is its corresponding class label. The trained model can be used to predict the possibility of abnormal process nodes according to real-time processing state data.

[0057] Through the above series of data collection, conversion, completion, clustering, correlation analysis and prediction steps, the present embodiment finally obtains a high-quality, standardized unified format data set. The data set not only maintains consistency in format and scale, but also contains preliminary classification of process state and correlation analysis results with product quality, providing a solid and information-rich data foundation for subsequent time series analysis, state transition detection and abnormal positioning steps.

[0058] S2: According to the unified format data set, apply clustering algorithm to group similar data types, determine data correlation groups for subsequent integration;

[0059] In the present embodiment, after obtaining the aforementioned unified format data set, in order to effectively integrate and analyze cross-process, multi-source data, it is necessary to first logically associate and group the data. The purpose of this step is to aggregate information describing the same production event or state, but physically dispersed in different data streams, to form cohesive data correlation groups, laying a foundation for subsequent timestamp synchronization and time series analysis.

[0060] Specifically, the method obtains data from the unified format data set generated from the foregoing steps. At this time, the data is in a unified format, but is still a collection of independent data streams from different process nodes and different sensors. To identify the internal relationship between these data streams, the embodiment first determines the priority of different data groups according to preset business logic and process knowledge by the analytic hierarchy process or a similar expert system method. For example, the priority of a sensor data group directly related to a product core quality indicator is higher than that of an auxiliary equipment monitoring data group. Through this process, a group priority sequence reflecting the importance and correlation of the data can be obtained.

[0061] According to the determined group priority sequence, the embodiment preferably uses a decision tree algorithm to automatically classify and group the data in the unified format data set. The decision tree takes the source, type, and priority of the data as the basis for classification, and allocates data entries to different data correlation groups through a series of judgment nodes. The advantage of this method is that the rules are clear, interpretable, and easy to adjust according to changes in the production process.

[0062] After completing the preliminary classification and grouping, to ensure the quality of the data correlation groups, the embodiment sets up a verification link. This link calculates the accuracy of the classification result and compares it with a preset classification accuracy threshold. If the classification accuracy is found to be lower than the threshold, it indicates that there may be noise or abnormal data points in the data set that interfere with the grouping process. At this time, the system will trigger data cleaning techniques, such as using statistical-based outlier detection methods or density-based clustering methods to identify and remove these abnormal data points, and then re-execute the classification and grouping process until the classification accuracy meets the requirements. In this way, a cleaned and accurately classified data correlation group, i.e., a cleaned data correlation group, can be obtained.

[0063] Further, considering that the data feature dimension in the cleaned data correlation group may be high, there is information redundancy, which is not conducive to the computational efficiency and accuracy of subsequent models. Therefore, the embodiment extracts key characteristics from the cleaned data correlation group and preferably uses principal component analysis to reduce the dimension of high-dimensional features. PCA finds a set of orthogonal bases (i.e., principal components) that can maximize the explanation of data variance by performing eigenvalue decomposition on the data covariance matrix.

[0064] The core is to solve the following characteristic equation:

[0065] ;

[0066] wherein, is the covariance matrix of the data characteristics, is the eigenvalue, is the corresponding eigenvector. By selecting the first The eigenvectors corresponding to the maximum eigenvalues form a projection matrix, which can project the original high-dimensional feature space into a low-dimensional subspace, thereby obtaining a reduced feature set that can represent the main information of the original data.

[0067] Through the above series of operations of prioritizing, classifying, grouping, cleaning and verifying, and dimensionality reduction on the data, the embodiment finally converts the scattered unified format data set into a series of structured and information-intensive data correlation groups. Each data correlation group contains key characteristics describing a specific production link or state, which provides high-quality input for precise timestamp alignment and deep data fusion integration in subsequent steps.

[0068] S3: After obtaining the data correlation group, synchronize the timestamps of all data entries, judge the consistency of the synchronized data to obtain a complete integrated data set;

[0069] In this embodiment, after obtaining the data correlation group containing logically related but possibly not time-aligned data through the foregoing steps, one of the core tasks of the method is to solve the timestamp inconsistency problem caused by different sensor sampling frequencies, network transmission delays, and system clock differences. This step aims to generate a time-synchronized and data-consistent complete integrated data set, providing high-quality data input for subsequent construction of accurate time series analysis models.

[0070] Specifically, the embodiment first performs timestamp synchronization and alignment operations on each data entry in the data correlation group. Since the collection time points of each data stream do not completely coincide, direct data integration will cause information misplacement. To solve this problem, the embodiment preferably uses linear interpolation to resample all data streams to a unified, high-resolution time grid. For any time point (the time point is between two actual collection points and ), its corresponding data value can be calculated as follows:

[0071] ;

[0072] where, and are the data values collected at time points and , respectively. Through this process, a preliminary time-synchronized data set that is aligned in time can be obtained.

[0073] However, merely achieving alignment of time points is not enough to guarantee consistency of data sequences in shape and trend. Events in certain production processes can exhibit slight stretching or offset in time in different data streams. To address this issue, the embodiment further performs deep consistency judgment and alignment on the time-synchronized data sets.

[0074] First, the sliding window technique is used to divide the continuous time series data into a series of time segment sets with or without overlap. This is intended to decompose the long sequence analysis problem into analysis of multiple short sequence segments, facilitating the capture of local features.

[0075] When processing the time segment set, if it is found that the data points within certain segments are unevenly distributed or missing, the data points can be first completed by means such as mean filling method to ensure the integrity of each time segment.

[0076] Subsequently, to effectively compare the internal patterns of different time segments, the embodiment extracts statistical features from the completed time segment set, and preferably uses principal component analysis to reduce the dimension of these statistical features to obtain a low-dimensional feature vector that can represent the core characteristics of each time segment.

[0077] Based on these low-dimensional feature vectors, the embodiment preferably uses the dynamic time warping algorithm to align the time series segments between different data groups. DTW is a method that can effectively measure the similarity between two time series, and is particularly suitable for processing sequences with time stretching. It finds an optimal warping path that minimizes the cumulative distance between corresponding points of two sequences, thereby achieving the best match of sequences in shape. Through the DTW algorithm, a more consistent aligned data set in sequence shape and trend can be obtained.

[0078] After completing the deep alignment, the embodiment extracts data consistency indicators from the aligned data set. For example, the minimum warping path distance calculated can be used as an indicator of consistency. If the consistency indicator meets the preset conditions, it indicates that the data streams within the current data correlation group have high consistency and correlation, and can be integrated.

[0079] Finally, for the data group that passes the consistency judgment, the embodiment uses data fusion technology to integrate it into a single multi-dimensional data stream, thereby obtaining a complete integrated data set. According to the requirements of the application scenario, the data fusion technology can choose weighted average method, Kalman filter or Bayesian inference, etc. This complete integrated data set has achieved high synchronization and consistency in time dimension and feature dimension, providing an ideal input for subsequent hidden Markov model analysis, ensuring the accuracy of state transition detection.

[0080] S4: For the complete integrated dataset, a time series analysis algorithm is used to detect changes in the flow state and obtain a state transition sequence record;

[0081] In this embodiment, after obtaining a time-synchronized and information-rich integrated dataset through the aforementioned steps, this method aims to perform in-depth analysis on the dataset to automatically and accurately identify the changes in the flow status of metering production operations across different process nodes. The purpose of this step is to transform the continuous, multi-dimensional sensor data stream into a discrete state sequence with clear business meaning, thereby providing structured input for subsequent anomaly detection and visual monitoring.

[0082] Specifically, this embodiment preferably employs a Hidden Markov Model (HMM) as the time series analysis algorithm. The reason for choosing an HMM is that the actual flow of production operations is a hidden state that cannot be directly observed by the monitoring system, while the complete integrated dataset generated in the aforementioned steps is the observation sequence produced by these hidden states. HMM provides a powerful probabilistic modeling framework for this type of problem.

[0083] To apply Hidden Markov Models (HMMs), the first step is to construct an HMM model that describes the characteristics of the current production workflow based on historical data or expert knowledge. A complete HMM model... Defined by three core parameter groups, namely .

[0084] First, the initial state probability distribution. It describes the beginning of the observation (i.e. (At any given moment), the production process exists in various possible hidden states. The probability of . Its mathematical expression is:

[0085] ;

[0086] in, This represents the hidden state of the system at the initial moment.

[0087] Secondly, it is the state transition probability matrix. This matrix describes the production process from a hidden state. Move to another hidden state The probability reflects the inherent logic and constraints of the production process. Its mathematical expression is:

[0088] ;

[0089] in, and These represent the system at time. and The hidden state.

[0090] Thirdly, the observation probability matrix. This matrix describes the state of the production process when it is in a specific hidden state. At that time, specific data vectors were observed. The probability of [something]. It establishes a bridge between invisible states and visible data. Its mathematical expression is:

[0091] ;

[0092] in, Represents time The observed values. These model parameters can be optimally obtained by training historical data using algorithms such as the Baum-Welch algorithm.

[0093] After the model is built, the complete integrated dataset (i.e., the observation sequence) acquired in real time is used. The goal of this embodiment is to decode the most likely corresponding hidden state sequence. To achieve this goal, this embodiment preferably employs the Viterbi algorithm. The Viterbi algorithm is a dynamic programming method that can efficiently find the globally optimal state path.

[0094] The algorithm defines a variable. To achieve this, the variable represents time. Arrival Status The probability value of the path with the highest probability among all paths. Its calculation process includes three stages: initialization, recursion, and termination backtracking.

[0095] First, perform initialization ( ):

[0096] ;

[0097] At the same time, define a backtracking path pointer. .

[0098] Then, perform the recursion (for ):

[0099] ;

[0100] in, Let be the total number of hidden states. While calculating the maximum probability, record the index of the previous state that maximizes that probability:

[0101] ;

[0102] Finally, at the end of the observation sequence ( The optimal path is found by backtracking. First, determine the last state:

[0103] ;

[0104] Then, proceed from back to front through the backtracking path pointers. Find the optimal preceding state:

[0105] ;

[0106] Through the decoding process of the Viterbi algorithm described above, the optimal state sequence is finally obtained. This is the state transition sequence record to be obtained in this embodiment. This record, in a structured and serialized manner, accurately describes each state experienced by the metering production operation within the observation period and the time point of its occurrence, providing a direct and reliable basis for subsequent real-time detection of abnormal flows.

[0107] S5: Extract real-time update indicators from the state transition sequence record. If the real-time update indicators exceed the preset threshold, trigger the instant notification mechanism to determine the location of the abnormal flow point.

[0108] In this embodiment, after obtaining the structured state transition sequence record through the aforementioned time-series analysis algorithm, this method needs to monitor and analyze the sequence in real time, aiming to identify potential abnormal flow behaviors from seemingly normal business flows in a timely and accurate manner. This step is the key link in transforming the decoded state sequence into an actionable early warning signal.

[0109] Specifically, the state transition sequence record itself is a discrete time series, and its dynamic characteristics are difficult to quantify through direct observation. Therefore, this embodiment first extracts one or more real-time updated indicators from the sequence that can reflect its dynamic behavior. Preferably, these indicators may include:

[0110] First, there is the state duration. This metric measures how long a specific production state is maintained continuously. An excessively long or short state duration usually indicates an abnormal production cycle, such as equipment jamming or process skipping.

[0111] Secondly, there is the state transition frequency. This metric measures how frequently the system switches between two or more states within a specific time window. An unusually high transition frequency may indicate that the system is oscillating unstablely between two states.

[0112] To achieve real-time monitoring of these indicators, this embodiment preferably employs a sliding window technique. This technique defines a window with a preset size. The time window is continuously scrolling forward along the state transition sequence record. At each window position, the system recalculates one or more of the aforementioned real-time update metrics. This generates a time series of these indicators themselves.

[0113] Furthermore, this embodiment focuses not on the absolute value of the indicator, but on its rate of change, because abnormal fluctuations typically manifest as sharp swings in the indicator within a short period. Therefore, this embodiment calculates the relative rate of change of the real-time updated indicator between two adjacent sliding windows. .

[0114] Its mathematical expression can be defined as:

[0115] ;

[0116] in, This represents the indicator value calculated within the current time window, while This represents the indicator value from the previous time window. This rate of change. It quantifies the drastic changes in the dynamic characteristics of the production state.

[0117] Subsequently, the system takes the absolute value of the rate of change. With a preset abnormal threshold Real-time comparisons are performed. This threshold can be derived based on historical normal production data; for example, it can be set as the 99th percentile of the historical rate of change distribution. If it is determined that... If so, it is considered that a potential abnormal flow event has occurred at the current point in time.

[0118] Once an anomaly is detected, this embodiment will immediately trigger an instant notification mechanism. This mechanism can be configured to send alarms to the monitoring center, record system logs, or link with other systems such as the Manufacturing Execution System via an application programming interface.

[0119] At the same time, and more importantly, the system precisely identifies and records the location of the anomaly. This includes not only the timestamp that triggered the alert, but also the complete contextual information that led to the anomaly, such as the specific metric name that changed drastically, the metric's value before and after the anomaly, and the state sequence segment currently covered by the sliding window. Through this step, an abstract alert is given specific time, state, and quantifiable metrics, thus providing precise location information for subsequent visualization and root cause analysis.

[0120] To further uncover the root causes of anomalous events, this embodiment can also perform in-depth analysis on multiple anomalous flow points accumulated over a period of time. Preferably, a clustering algorithm can be used to group these anomalous flow points to form an anomalous point group set. This aims to aggregate anomalous points with similar characteristics together.

[0121] For each set of anomalies, this embodiment can extract common features of the flow points within the group and use methods such as feature matching to identify recurring anomaly patterns, thereby obtaining an anomaly pattern sequence. In this way, this method can discover recurring systemic problems from isolated anomaly events, providing a deeper insight for process optimization or equipment maintenance, rather than just dealing with single alarms.

[0122] S6: Generate a visual chart to display the dynamic tracking path by identifying the location of abnormal flow points, and obtain a full-process monitoring view;

[0123] In this embodiment, after determining the location of abnormal flow points through the aforementioned steps, this method needs to transform these discrete, data-based alarm messages into a visual interface that provides management and operators with an intuitive and comprehensive situational awareness. This step aims to bridge the gap between raw data and human cognition, achieving real-time and transparent management of the production business flow status by generating a dynamic, full-process monitoring view.

[0124] Specifically, the method first generates a dynamic tracking path based on the state transition sequence records decoded in the aforementioned steps. This path characterizes the flow of a production object between various process nodes (i.e., states). Visually, this path can be represented as a series of nodes and directed edges. Each node represents a production state, and the directed edges between nodes represent a state transition. This path is updated in real-time based on the latest state transition sequence records.

[0125] To ensure that the dynamic tracking path not only displays the process flow but also reveals the process's health status, this embodiment introduces and constructs a health status model. This model aims to comprehensively evaluate multi-dimensional real-time features from a fully integrated dataset into a single, quantifiable health status indicator. This approach avoids the limitations of judging anomalies solely based on a single threshold, instead providing a more hierarchical evaluation system.

[0126] Preferably, the health status index This can be achieved by analyzing multiple standardized real-time features. We calculate it by weighted summation. Its mathematical expression is:

[0127] ;

[0128] in, Represents a point in time The The standardized value of a real-time feature, and The corresponding weight represents the feature, and the sum of all weights is 1 (i.e., These weights can be determined based on expert experience or through data-driven methods to reflect the contribution of different characteristics to the overall health of the process.

[0129] Calculate real-time health status indicators To convert these values ​​into intuitive visual signals, this embodiment employs a color encoding method. This method maps continuous health index values ​​to discrete color categories using a preset piecewise function. Preferably, the color encoding function Color(H_t) can be defined as follows:

[0130] ;

[0131] in, and These are two preset health thresholds that together define three health levels: normal (green), warning (yellow), and abnormal (red). These two thresholds can be scientifically set based on the statistical distribution of historical data (e.g., quantiles).

[0132] Finally, this embodiment integrates all the aforementioned visual elements using dynamic rendering technology to generate a full-process monitoring view. In this view, the color of nodes or edges along the dynamic tracking path changes dynamically based on their real-time calculated health status. When the health status of a certain segment deteriorates, its color changes from green to yellow, or even red, thus providing a clear visual warning to monitoring personnel.

[0133] Furthermore, the locations of abnormal flow points identified in the aforementioned steps will be explicitly marked on this view with special icons or highlights. This allows operators to not only see the overall health of the process but also to focus their attention immediately on the most critical abnormal events that have occurred.

[0134] Through the series of operations described above—constructing dynamic paths, modeling health, implementing color coding, and finally integrating rendering—this embodiment transforms the complex production data flow into an information-rich, hierarchical, and easy-to-understand end-to-end monitoring view. This view not only displays the real-time trajectory of the production flow but also reveals its potential risks and health trends, providing a basis for decision-making in subsequent closed-loop feedback and parameter adaptive optimization steps.

[0135] S7: Based on the full-process monitoring view, the data acquisition module is cyclically fed back to adjust the sensor parameters and judge the feedback effect to continuously optimize the accuracy of status updates.

[0136] In this embodiment, after generating a global and intuitive end-to-end monitoring view through the aforementioned steps, the purpose of this method is not merely to achieve passive monitoring and alarms, but to construct a self-optimizing and proactively adaptive closed-loop feedback control system. This step aims to utilize the macroscopic insights gained from the monitoring view to guide the front-end data acquisition process in reverse, and to continuously optimize the accuracy and sensitivity of the entire monitoring system's status updates by dynamically adjusting sensor parameters.

[0137] Specifically, this closed-loop feedback mechanism first extracts state update features that characterize the overall system performance in real time from the full-process monitoring view. The full-process monitoring view is essentially a high-level abstraction of high-dimensional production data after multi-layered analysis and visualization, making it difficult to directly use as input for the feedback model. Therefore, this embodiment employs streaming processing technology to parse the underlying data streams driving the view updates (e.g., state health sequences, abnormal transition point sequences), thereby obtaining a structured state update feature sequence. This feature sequence may preferably include: the duration distribution of specific states, the frequency of key state transitions, the volatility of state health indicators, or entropy values, etc.

[0138] After obtaining the state update feature sequence, this embodiment aims to establish a mapping model from system performance to optimal acquisition parameters. Therefore, this embodiment preferably uses the Support Vector Regression (SVR) algorithm to construct a parameter tuning model. SVR is particularly suitable for handling high-dimensional, nonlinear regression problems. This model uses the state update feature sequence as the input vector. The sensor parameters that need to be optimized are used as the output target values. .

[0139] The goal of the SVR model is to find a function This ensures that the function's predicted values ​​for all training samples are close to the true target values. The deviation is within the preset range. Within a given range, while minimizing model complexity, the optimization problem can be formally expressed as:

[0140] ;

[0141] And satisfy the following constraints:

[0142] ;

[0143] in, It is a regularization constant used to balance model complexity and training error. A non-sensitive region was defined, and and These are slack variables, allowing some sample points to fall outside the insensitive region. By training on historical data, this SVR model can learn the complex mapping relationship between the two.

[0144] After the model training is complete, when the system is running, it continuously inputs the real-time state update feature sequences extracted from the full-process monitoring view into the trained SVR model. The model then predicts and outputs a set of optimized sensor parameter sequences for the current state.

[0145] Furthermore, the system will automatically extract and generate executable configuration update instructions from the optimized sensor parameter sequence. These instructions will be synchronized to the corresponding data acquisition modules or specific sensor hardware on the production site in a secure and reliable manner through an instruction distribution mechanism, thereby completing the dynamic adjustment of the front-end data acquisition behavior.

[0146] Finally, to ensure the effectiveness of this closed-loop feedback and achieve continuous optimization, this embodiment also includes a feedback effect assessment step. After the sensor parameters are adjusted, the system continuously monitors the newly acquired data stream. Preferably, time series analysis methods can be used to quantitatively evaluate the feedback effect brought about by this parameter adjustment by calculating changes in indicators such as accuracy, stability, or confidence of the state update before and after the adjustment. This evaluation result is not only used to verify the correctness of the feedback, but it can also serve as new training data for further iteration and optimization of the SVR parameter adjustment model, thereby enabling the entire monitoring system to have adaptive and self-learning capabilities.

[0147] Please see the appendix Figure 2 A real-time visual monitoring system for the status of metering production operations includes the following steps:

[0148] The data acquisition and standardization module is used to collect heterogeneous data from multiple process nodes through a sensor network, and to convert data of different formats using a standardized protocol to obtain a unified format dataset.

[0149] The data clustering and association grouping module is used to group similar data types based on a unified format dataset using clustering algorithms, and to determine data association groups for subsequent integration.

[0150] The timestamp synchronization and consistency judgment module is used to synchronize timestamps to align all data entries after obtaining the data association group, and judge the consistency of the synchronized data to obtain a complete integrated dataset.

[0151] The time series analysis and state transition detection module is used to detect changes in the flow state of a complete integrated dataset using time series analysis algorithms, and obtain state transition sequence records.

[0152] The anomaly detection and notification triggering module is used to extract real-time updated indicators from the state transition sequence record. If the real-time updated indicators exceed the preset threshold, an instant notification mechanism is triggered to determine the location of the abnormal flow point.

[0153] The visualization and dynamic tracking module is used to generate visual charts to display the dynamic tracking path based on the location of abnormal flow points, thereby obtaining a full-process monitoring view;

[0154] The feedback optimization and parameter adjustment module is used to continuously optimize the accuracy of status updates by cyclically feeding back to the data acquisition module based on the full-process monitoring view and judging the feedback effect.

[0155] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time visual monitoring of the state of a metering production service flow, characterized in that, The method comprises the following steps: S1: Collecting heterogeneous data, converting different format data into a unified format data set by using a standardization protocol; S2: According to the unified format data set, applying a clustering algorithm to group similar data types, determining data correlation groups for subsequent integration; S3: After obtaining the data correlation groups, synchronizing time stamps to align all data entries, judging the consistency of the synchronized data to obtain a complete integrated data set; S4: For the complete integrated data set, using a time series analysis algorithm to detect state change, obtaining a state transition sequence record; S5: Extracting real-time update indicators from the state transition sequence record; S6: Generating a visual chart to display the dynamic tracking path according to the abnormal flow point position, obtaining a whole process monitoring view; S7: According to the whole process monitoring view, feedback to the data acquisition module to adjust the sensor parameters, and judge the feedback effect.

2. The method of claim 1, wherein, The step of obtaining a unified format data set specifically comprises: Obtaining processing state and quality detection results to generate an original data set; Converting the format of the original data set by using the standardization protocol; If the integrity of the unified format data set is lower than a preset threshold, filling in the missing values by using a data completion algorithm; According to the filled data set, using a K-means algorithm to perform clustering analysis on the processing state to determine the state classification of the process node; For the state classification, calculate the Pearson correlation coefficient of the quality detection result and the processing state to judge the correlation strength; If the correlation strength is higher than a preset threshold, predict the potential abnormal process node by using a support vector machine algorithm.

3. The method of claim 1, wherein the method further comprises: The step of determining data correlation groups for subsequent integration specifically comprises: Obtaining data correlation groups from the unified format data set, determining the priority of each group by using an analytic hierarchy process to obtain a group priority sequence; According to the group priority sequence, using a decision tree algorithm to classify the data correlation groups and judging the classification accuracy; If the classification accuracy is lower than a preset threshold, removing abnormal data points by using data cleaning technology to obtain cleaned data correlation groups; Extracting key characteristics from the cleaned data correlation groups, performing dimensionality reduction processing by using principal component analysis to obtain a reduced feature set; According to the reduced feature set, using data fusion technology to integrate multiple characteristics to generate a final integrated data set.

4. The method of claim 1, wherein, The step of judging the consistency of the synchronized data to obtain a complete integrated data set specifically comprises: Extracting time sequence features from the time-synchronized data set, dividing the time sequence by using a sliding window technology to obtain a time segment set; If the data points in the time segment set are unevenly distributed, fill in the missing data points by using the mean filling method; Extracting statistical features from the completed time segment set, using principal component analysis to perform dimensionality reduction processing on the statistical features; Aligning the time sequences between different groups by using a dynamic time warping algorithm to obtain an aligned data set; Obtaining a data consistency index from the aligned data set, if the consistency index meets a preset condition, integrating multiple groups of data by using data fusion technology to obtain the complete integrated data set.

5. The method of claim 1, wherein the method further comprises: The step of obtaining the state transition sequence record specifically includes: extracting a transition time point from the state transition sequence record, determining a switching time interval between states by using a time interval calculation method, and obtaining a time interval sequence; extracting a state switching frequency from the time interval sequence, analyzing a frequency distribution by using a Fourier transform, and obtaining a frequency distribution feature; obtaining a state change trend from the frequency distribution feature, smoothing a trend curve by using a sliding window technique if the trend fluctuation exceeds a preset threshold, and obtaining a smoothed trend sequence; detecting content consistency of the smoothed trend sequence by using a sequence alignment method, and obtaining an abnormal state transition point according to a consistency detection result.

6. The method of claim 1, wherein the method further comprises: The step of determining the abnormal flow transition point position specifically includes: obtaining the real-time update index from the state transition sequence record, calculating an index change rate by using a sliding window method, and obtaining an index change sequence; determining an abnormal index point by threshold comparison logic if any change rate in the index change sequence exceeds a preset threshold; determining the abnormal flow transition point position by using a timestamp alignment method according to the abnormal index point; grouping the abnormal flow transition points by using a clustering algorithm, and obtaining an abnormal point grouping set; extracting an in-group flow transition point feature from the abnormal point grouping set, identifying a repeated abnormal pattern by using a feature matching method, and obtaining an abnormal pattern sequence.

7. The method of claim 1, wherein the method further comprises: The step of obtaining the whole-process monitoring view specifically includes: obtaining a time sequence point from the abnormal flow transition point position, and arranging the dynamic tracking path by using a timestamp alignment method; extracting a state change rate from the dynamic tracking path, and calculating a state health degree of each time point by using a sliding window method; grading and labeling the state health degree by using a color coding method, and obtaining a color coding sequence; generating the whole-process monitoring view by using a dynamic rendering technique according to the color coding sequence.

8. The method of claim 1, wherein the method further comprises: The step of judging the feedback effect to continuously optimize the state update accuracy specifically includes: obtaining a state update data stream from the whole-process monitoring view, analyzing data stream characteristics by using a stream processing technique, and obtaining a state update feature sequence; constructing a parameter adjustment model by using a support vector machine algorithm for the state update feature sequence, and obtaining an optimized sensor parameter sequence; extracting a configuration update instruction from the optimized sensor parameter sequence, and synchronizing to a sensor module by an instruction distribution mechanism; calculating response consistency by using a time series analysis method according to the state response data after the update configuration, and taking the response consistency as a judgment basis of the feedback effect.

9. The method of claim 1, wherein the method further comprises: In the step of detecting the flow state change by using the time series analysis algorithm: the time series analysis algorithm is a hidden Markov model, and an optimal state sequence is obtained by decoding an observation sequence in the complete integrated data set by using a Viterbi algorithm as the state transition sequence record.

10. A real-time visual monitoring system for metering production business flow state according to the real-time visual monitoring method for metering production business flow state of any one of claims 1-9, characterized in that, The method includes the following steps: a data acquisition and standardization module, configured to acquire heterogeneous data from a plurality of process nodes by a sensor network, convert different format data by using a standardization protocol, and obtain a unified format data set; a data clustering and correlation grouping module configured to group similar data types by applying a clustering algorithm to the uniform format data set and determine data correlation groups for subsequent integration; a timestamp synchronization and consistency judgment module configured to synchronize timestamps to align all data entries and determine data consistency after synchronization to obtain a complete integrated data set; a time series analysis and state transition detection module configured to detect state changes by using a time series analysis algorithm on the complete integrated data set and obtain a state transition sequence record; an abnormality detection and notification triggering module configured to extract real-time update indicators from the state transition sequence record, trigger an instant notification mechanism if the real-time update indicators exceed a preset threshold, and determine an abnormal flow point position; a visualization and dynamic tracking module configured to generate a visualization chart to display a dynamic tracking path and obtain a whole-process monitoring view through the abnormal flow point position; a feedback optimization and parameter adjustment module configured to adjust sensor parameters by circular feedback to the data acquisition module according to the whole-process monitoring view, determine feedback effects, and continuously optimize state update accuracy.

Images