A steel production monitoring method and monitoring system based on the Internet of Things

Abstract

This invention provides a method and system for monitoring steel production based on the Internet of Things (IoT). It collects multi-physical field collaborative sensing data sets by deploying a multi-type sensor collaborative sensing network at target process nodes in the steel production process. Based on this multi-physical field collaborative sensing data set, a physical field coupling mapping model is constructed. By tracking the evolutionary correlation of cross-physical field characteristics, a nonlinear mapping relationship is established between the evolution trajectory of micro-organism structure and the trend of macro-production state changes. Through the physical field coupling mapping model, continuous feature correlation analysis is performed on the collected multi-physical field collaborative sensing data set based on the nonlinear mapping relationship. Production state assessment results are generated, and production parameter adjustment instructions with time-series control logic are generated. These instructions are then pushed to the steel production control system for execution of control operations. Using this invention can improve the overall accuracy and effectiveness of steel production monitoring.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing, and more specifically, to a method and system for monitoring steel production based on the Internet of Things. Background Technology

[0002] With the continuous development of the steel industry, steel production monitoring, through the collection, analysis, and evaluation of various physical parameters and status information in the production process, can achieve effective supervision and control of the production process. The steel production process involves complex physicochemical reactions, and existing detection methods are insufficient to fully reflect the true state of the production process. For example, comparing preset thresholds or using simple model analysis makes it difficult to establish a correlation between the evolution of microstructure and macroscopic production status. This results in a rather crude judgment of abnormal situations, failing to accurately determine the specific starting point, development trend, and potential scope of impact of the abnormality. Consequently, the generated production parameter adjustment instructions lack specificity and timeliness, making it difficult to accurately match the development process of the abnormality, thus affecting the stability of steel production and the consistency of product quality. Summary of the Invention

[0003] This invention provides a method and system for monitoring steel production based on the Internet of Things (IoT).

[0004] In a first aspect, embodiments of the present invention provide an Internet of Things (IoT)-based steel production monitoring method. The method includes: collecting a multi-physical field collaborative sensing data set containing temperature field distribution data, stress field distribution data, magnetic field intensity change data, and acoustic emission signal data through a multi-type sensor collaborative sensing network deployed at target process nodes in the steel production process; constructing a physical field coupling mapping model based on the multi-physical field collaborative sensing data set, and establishing a nonlinear mapping relationship between the microstructure evolution trajectory and the macroscopic production state change trend by tracking the evolution correlation of cross-physical field features; performing continuous feature correlation analysis on the collected multi-physical field collaborative sensing data set based on the nonlinear mapping relationship using the physical field coupling mapping model, and generating a production state assessment result; generating a production parameter adjustment instruction with time-series control logic based on the production state assessment result, and pushing the production parameter adjustment instruction to the steel production control system to execute the control operation.

[0005] Secondly, embodiments of the present invention provide a monitoring system, comprising: a memory storing a computer program; and a processor for loading the computer program to implement the IoT-based steel production monitoring method described above.

[0006] This invention provides an IoT-based steel production monitoring method that collects multi-physics field data (temperature, stress, magnetic fields, and acoustic emission signals) by deploying a multi-type sensor network at target process nodes. This enables the synchronous acquisition of multi-dimensional physical information during steel production. This multi-physics field collaborative acquisition method fully reflects the inherent coupling relationship between different physical fields, providing a more comprehensive data foundation for subsequent production status assessment. A physical field coupling mapping model is constructed based on the multi-physics field data. By tracking the evolutionary correlation of cross-physical field characteristics, a nonlinear mapping relationship between microscopic and macroscopic states is established, allowing the model to adaptively adjust with the production process. This effectively improves adaptability to complex production environments and avoids the limitations of traditional static models in dealing with changes in the production process. Through continuous feature correlation analysis of the collected data using this model, a production status assessment result is generated, including the onset time of anomalies, development rate, and impact range. Compared to traditional monitoring methods that can only determine whether anomalies occur, this method can more accurately pinpoint the time of anomaly occurrence, quantify the development trend of anomalies, and predict the impact range, providing more specific decision-making basis for control operations. Based on the abnormal information in the assessment results, production parameter adjustment instructions with time-series control logic are generated. This allows parameter adjustments to not only include specific values, but also to determine the execution sequence based on the rate of abnormal development and the scope of impact. This achieves precise matching between parameter adjustments and the abnormal evolution process, avoiding over-adjustment or response lag problems that may be caused by traditional static control, thereby improving the overall accuracy and effectiveness of steel production monitoring. Attached Figure Description

[0007] Figure 1 This is a flowchart of a steel production monitoring method based on the Internet of Things provided in an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of the composition of a monitoring system provided in an embodiment of the present invention. Detailed Implementation

[0009] Please see Figure 1 This is a flowchart of a steel production monitoring method based on the Internet of Things provided in an embodiment of the present invention. It is executed by a monitoring system and may specifically include the following steps: Step S100: Collect a multi-physics field collaborative sensing data set, including temperature field distribution data, stress field distribution data, magnetic field intensity change data, and acoustic emission signal data, by deploying a multi-type sensor collaborative sensing network at the target process node of the steel production process.

[0010] A multi-sensor collaborative sensing network is a network system composed of various types of sensors. These sensors work together to monitor the steel production process from different physical dimensions. Temperature field distribution data describes the spatial distribution of temperature during steel production, reflecting temperature differences at different locations. Stress field distribution data provides information on the spatial distribution of internal stress in steel; the magnitude and distribution of stress affect the mechanical properties and structural integrity of the steel. Magnetic field strength variation data records the changes in the magnetic field strength around the steel over time. In steel production, changes in the magnetic field may be related to the steel's magnetic properties, electromagnetic induction, and other factors. Acoustic emission signal data consists of elastic wave signals generated when steel is subjected to stress, deforms, or undergoes internal structural changes. Analysis of acoustic emission signals can reveal the internal damage and deformation process of the steel.

[0011] When deploying multiple types of sensors at target process nodes in the steel production process, temperature sensors can be deployed to collect temperature field distribution data. For stress field distribution data, strain gauge sensors can be installed on key structural components to measure the strain of the steel structure under stress. Magnetic field strength change data can be collected by Hall sensors, and acoustic emission signal data can be collected by acoustic emission sensors.

[0012] Step S200: Construct a physical field coupling mapping model based on a multi-physics field collaborative sensing data set, and establish a nonlinear mapping relationship between the evolution trajectory of micro-organizational structure and the trend of macro-production state changes by tracking the evolution correlation of cross-physical field characteristics.

[0013] The physical field coupling mapping model is a model that comprehensively considers the interactions and coupling relationships between multiple physical fields. It can integrate and analyze data from different physical fields, thereby revealing the intrinsic connections between them. The evolutionary correlation of cross-physical field characteristics refers to the correlation and influence between the changes of different physical field characteristics in time and space. The microstructure evolution trajectory describes the path of change in the internal microstructure of steel (such as grain size and phase composition) during the production process. The macroscopic production state change trend reflects the direction and extent of change in some macroscopic indicators (such as rolling force and cooling rate) during steel production.

[0014] In one implementation, step S200 may specifically include the following steps S210 to S260: Step S210: Perform spatiotemporal registration processing on the multi-physics field collaborative sensing data set, adjust the sliding step size of the time window according to the sampling frequency fluctuation of different sensors, calibrate the temperature field, stress field, magnetic field and acoustic emission signal data to a unified spatiotemporal reference system, and generate a spatiotemporal registration data stream.

[0015] Spatiotemporal registration is the process of uniformly calibrating data from different physical fields in time and space, enabling comparison and analysis of data from different physical fields within the same spatiotemporal coordinate system. The time window sliding step size is the interval at which the time window moves when processing time series data. Since the sampling frequencies of different sensors may fluctuate, the time window sliding step size needs to be adjusted according to these fluctuations to ensure data accuracy and consistency. A unified spatiotemporal reference system is a common spatiotemporal coordinate system into which all physical field data needs to be calibrated for subsequent analysis and processing. The spatiotemporal registration data stream is the data stream generated after spatiotemporal registration processing, containing all physical field data within the unified spatiotemporal reference system.

[0016] In one implementation, step S210 may specifically include the following steps S211 to S215: Step S211: Extract the original acquisition timestamps and spatial coordinate information of each physical field data in the multi-physics collaborative sensing data set, construct a spatiotemporal coordinate matrix, and mark the process node number corresponding to each data point.

[0017] The original acquisition timestamp records the specific time of data acquisition for each physical field, while the spatial coordinate information indicates the location of the data acquisition point in space. The spatiotemporal coordinate matrix is ​​a matrix that integrates time and spatial information, clearly showing the spatiotemporal distribution of each physical field's data. The process node number is a unique identifier assigned to distinguish different production process nodes; marking the process node number corresponding to each data point helps in subsequent data classification and analysis.

[0018] When extracting the raw acquisition timestamps and spatial coordinates, data from different types of sensors may have different formats. For example, data from a temperature sensor may include a timestamp, a temperature value, and corresponding spatial coordinates. By parsing the data format output by the sensor, the timestamps and spatial coordinates can be extracted. For all physical field data, the extracted timestamps and spatial coordinates are organized into a matrix, which is the spatiotemporal coordinate matrix. Simultaneously, each data point is labeled with a corresponding process node number based on the process node where the data acquisition point is located. For example, in different stages of steelmaking, there are different process nodes, such as converter steelmaking nodes and refining nodes; the acquired data points are labeled with the corresponding process node numbers.

[0019] Step S212: Monitor the current sampling frequency of each sensor, calculate the sampling frequency fluctuation value, and when the fluctuation value exceeds the preset fluctuation threshold, adjust the sliding step size of the time window according to the fluctuation ratio so that the window size is positively correlated with the current sampling frequency.

[0020] The sampling frequency is the number of times a sensor collects data per unit of time. The current sampling frequency reflects the actual sampling situation of the sensor at the current moment. The sampling frequency fluctuation value is the difference between the current sampling frequency and the set standard sampling frequency, measuring the degree of instability of the sampling frequency. The preset fluctuation threshold is a pre-set standard value; when the sampling frequency fluctuation value exceeds this threshold, it indicates that the fluctuation of the sampling frequency has reached a level requiring adjustment. The time window sliding step size is the interval at which the time window moves when processing time series data. Adjusting the time window sliding step size can match the data processing with the current sampling frequency.

[0021] Monitoring the current sampling frequency of each sensor can be achieved through feedback information from the sensor itself or through statistical analysis of the collected data. For example, the actual sampling frequency can be calculated by recording the number of data points collected by the sensor within a certain time period. When calculating the sampling frequency fluctuation value, the current sampling frequency is subtracted from the standard sampling frequency to obtain the fluctuation value. When the fluctuation value exceeds a preset fluctuation threshold, the sliding step size of the time window is adjusted according to the fluctuation ratio.

[0022] Step S213: For the temperature field distribution data, perform spatiotemporal interpolation on the original data according to the adjusted time window. Use Kriging interpolation to supplement missing coordinate point data in the spatial dimension and use linear interpolation to supplement missing time point data in the time dimension to generate continuous spatiotemporal temperature field data.

[0023] The adjusted time window size and sliding step are adjusted based on sampling frequency fluctuations. Spatiotemporal interpolation is an operation that supplements and improves data in both time and space, making the data more continuous in time and space. Kriging interpolation uses the spatial correlation of known points to predict the values ​​of unknown points. Linear interpolation performs linear fitting based on the values ​​of adjacent data points to obtain the values ​​of missing data points. Spatiotemporally continuous temperature field data is temperature field data that is continuous in both time and space after spatiotemporal interpolation processing.

[0024] When performing spatiotemporal interpolation on temperature field distribution data, the interpolation range is first determined based on the adjusted time window. Spatially, when missing coordinate points exist, Kriging interpolation is used. Specifically, the spatial distribution and correlation of known temperature data points are analyzed first, and a covariance function is constructed to describe the spatial correlation. Then, based on the covariance function and the known data points, the predicted temperature value for the missing coordinate points is calculated. Temporally, if missing time points exist, linear interpolation is used. For example, given the temperature values ​​of two adjacent time points, the temperature value for the missing time point is calculated based on the linear relationship between the time interval and temperature change. Through this spatiotemporal interpolation operation, continuous spatiotemporal temperature field data is generated, making the temperature field data more complete and continuous in space and time.

[0025] Step S214: Using the same window processing logic as the temperature field, perform strain transfer path completion on the stress field distribution data, perform eddy current effect spatiotemporal compensation on the magnetic field intensity change data, and perform spectral energy spatiotemporal alignment on the acoustic emission signal data to generate spatiotemporal continuous data for each physical field.

[0026] Window processing logic refers to the methods and rules for operating according to an adjusted time window when processing data. Strain transfer path completion supplements and improves stress field distribution data to address missing or incomplete strain information, thus more accurately describing the stress transfer process. Eddy current effect spatiotemporal compensation corrects and compensates for data deviations caused by eddy current effects in magnetic field strength variation data, making the magnetic field strength data more accurate in time and space. Spectral energy spatiotemporal alignment calibrates and matches the spectral energy of acoustic emission signal data in time and space, ensuring the consistency and accuracy of the acoustic emission signal data. When performing strain transfer path completion on stress field distribution data, the continuity of strain information in the stress field data is analyzed according to the adjusted time window. If missing information is found in the strain transfer path, the missing strain information can be supplemented by analyzing and inferring the strain conditions of adjacent data points and using appropriate interpolation or fitting methods. For magnetic field strength variation data, eddy current effects may cause deviations in the measured values ​​of magnetic field strength, requiring spatiotemporal compensation. This can be achieved by establishing a mathematical model of the eddy current effect, calculating the deviation caused by the eddy current effect based on known physical parameters and measurement data, and then performing corresponding compensation. When performing spatiotemporal alignment of the spectral energy of acoustic emission signal data, a temporal and spatial reference standard is first determined. Then, the spectral energy of the acoustic emission signal is adjusted and matched according to this standard to make the acoustic emission signal data at different time and space points comparable, and finally, spatiotemporal continuous data of each physical field are generated.

[0027] Step S215: Align the spatiotemporal continuous data of each physical field with the coordinates and timestamps of a unified spatiotemporal reference system to obtain a spatiotemporal registration data stream containing temperature field, stress field, magnetic field and acoustic emission signals. Each spatiotemporal coordinate point in the spatiotemporal registration data stream contains the monitoring values ​​of the four types of physical fields.

[0028] A unified spacetime reference frame is a common spacetime coordinate system. Continuous spacetime data from various physical fields need to be calibrated to this reference frame to achieve data uniformity and comparability. Coordinate and timestamp alignment matches data from different physical fields in time and space, ensuring a correspondence between different physical field data at the same spacetime coordinate point. The spacetime registration data stream is a data stream generated after spacetime registration processing, containing all physical field data within the unified spacetime reference frame. Each spacetime coordinate point includes monitored values ​​for temperature, stress, magnetic fields, and acoustic emission signals.

[0029] When aligning the spatiotemporal continuous data of various physical fields to coordinates and timestamps using a unified spatiotemporal reference system, the specific coordinates and time standard of the unified spatiotemporal reference system must first be determined. For example, using a fixed time point and spatial origin as a reference, the spatiotemporal continuous data of each physical field are transformed. For each spatiotemporal coordinate point, the monitored values ​​of temperature field, stress field, magnetic field, and acoustic emission signal are integrated so that they correspond at the same spatiotemporal coordinate point. For example, at a certain time and spatial coordinate point, the temperature value, stress value, magnetic field strength value, and acoustic emission signal strength value of that point are combined to form a dataset containing the monitored values ​​of the four types of physical fields. By performing this processing on all spatiotemporal coordinate points, a spatiotemporal registration data stream containing temperature field, stress field, magnetic field, and acoustic emission signal is finally obtained.

[0030] Step S220: Perform cross-physics field feature evolution tracking on the spatiotemporal registration data stream. By using a rolling time window, obtain the diffusion trend of the temperature field gradient, the transmission and attenuation process of the stress field strain, the time-varying characteristics of the eddy current response of the magnetic field strength, and the spectral energy migration trajectory of the acoustic emission signal to generate a multi-physics field feature evolution sequence.

[0031] Cross-physics feature evolution tracking involves monitoring and analyzing the temporal and spatial changes of different physical field features to identify their correlations and evolutionary patterns. A rolling time window observes data changes over different time periods by continuously moving the time window. The diffusion trend of the temperature field gradient describes the spatial rate of temperature change over time, reflecting the heat transfer and distribution process. The stress-strain transmission attenuation process refers to the gradual decrease in strain generated by stress as it propagates within steel with increasing distance and time. The time-varying characteristics of the eddy current response of magnetic field strength record the changes in magnetic field strength over time under the influence of eddy current effects. The spectral energy migration trajectory of acoustic emission signals is the path of the spectral energy of acoustic emission signals moving and changing in time and space. A multi-physics feature evolution sequence is a sequence recording the evolution of different physical field features over time. A rolling time window is used when performing cross-physics feature evolution tracking on a spatiotemporally registered data stream. For example, a fixed-size time window is set, and this window is continuously moved across the spatiotemporally registered data stream. Within each time window, the diffusion trend of the temperature field gradient is calculated. The temperature field gradient can be obtained by performing spatial difference operations on temperature field data, and then its changes within a time window can be observed to analyze its diffusion trend. For the strain transmission and attenuation process in the stress field, the changes in strain with time and space in the stress field data can be analyzed to calculate the strain attenuation rate and transmission velocity. When monitoring the time-varying characteristics of the eddy current response of magnetic field strength, the changes in magnetic field strength at different time points can be observed based on the magnetic field strength variation data and the known eddy current effect model. For the spectral energy migration trajectory of acoustic emission signals, spectral analysis of the acoustic emission signal data can be performed to identify the movement patterns of spectral energy in time and space.

[0032] Step S230: Track the correlation strength based on the multi-physics feature evolution sequence, calculate the correlation degree of cross-physics features by the feature covariance change rate within continuous time segments, and generate a time-varying curve of correlation strength that is updated with the production process.

[0033] Correlation strength tracking aims to identify how the degree of correlation between different physical field features changes over time. A continuous time segment is a period of time into which the entire production process is divided according to certain time intervals. The rate of change of feature covariance reflects how quickly the covariance between different features changes within a continuous time segment. Covariance is an indicator that measures the linear relationship between two variables; the rate of change of covariance allows for a more sensitive capture of the dynamic changes in the correlation between features. The degree of correlation across physical field features is a quantitative indicator used to describe the tightness of the correlation between different physical field features. The time-varying correlation strength curve is a curve that records the change in correlation strength as the production process progresses, visually demonstrating the dynamic evolution of the correlation between different physical field features during the production process.

[0034] In one implementation, step S230 may specifically include the following steps S231 to S236: Step S231: Divide the multiphysics feature evolution sequence into continuous non-overlapping time segments according to the production process. The length of each time segment is determined by the adjusted time window sliding step size.

[0035] The production process encompasses the entire steel production process from start to finish. Dividing the multiphysics characteristic evolution sequence according to the production process allows for the analysis of data across different time segments arranged chronologically. Continuous, non-overlapping time segments refer to time intervals that are consecutive in time and do not overlap. The adjusted time window sliding step size, determined by the interval between each movement of the time window based on sensor sampling frequency fluctuations, determines the length of each time segment.

[0036] When segmenting the multiphysics feature evolution sequence, based on the adjusted time window sliding step size, continuous non-overlapping time segments are divided from the start time of the sequence at fixed time intervals. For example, if the adjusted time window sliding step size is 10 minutes, then a time segment is divided every 10 minutes. Based on this, the multiphysics feature evolution sequence is segmented into multiple continuous non-overlapping time segments, each containing multiphysics feature data within that time period.

[0037] Step S232: Within each time segment, calculate the covariance between the temperature field gradient diffusion trend and the stress field strain transmission attenuation process. The covariance value characterizes the degree of synchronous change of the two types of features within the segment.

[0038] Covariance is a statistic used to measure the linear relationship between two variables. By calculating the covariance between the temperature gradient diffusion trend and the stress strain transmission attenuation process, we can understand the synchronous changes of these two characteristics in each time segment. The higher the degree of synchronous change, the stronger the correlation between the temperature gradient diffusion trend and the stress strain transmission attenuation process.

[0039] Within each time segment, data on the temperature field gradient diffusion trend and the stress field strain transfer attenuation process are first extracted. For example, by processing the temperature field data, a data sequence of the temperature field gradient diffusion trend is obtained, and by analyzing the stress field data, a data sequence of the stress field strain transfer attenuation process is obtained. Then, the covariance value of the two data sequences is calculated using a general covariance calculation formula. The calculated covariance value can characterize the degree of synchronization between the temperature field gradient diffusion trend and the stress field strain transfer attenuation process within that time segment.

[0040] Step S233: Calculate the covariance values ​​of temperature field and magnetic field, temperature field and acoustic emission signal, stress field and magnetic field, stress field and acoustic emission signal, and magnetic field and acoustic emission signal using the same method to obtain six sets of cross-physical field covariance sequences.

[0041] When calculating the covariance values ​​of temperature field and magnetic field, temperature field and acoustic emission signal, stress field and magnetic field, stress field and acoustic emission signal, and magnetic field and acoustic emission signal, the same method is used as when calculating the covariance values ​​of temperature field gradient diffusion trend and stress field strain transmission attenuation process. That is, the corresponding physical field feature data are first extracted from the multi-physics feature evolution sequence, and then the covariance value is calculated according to the covariance calculation formula.

[0042] For the temperature and magnetic fields, temperature field data and magnetic field strength variation data are extracted, and the covariance value is calculated according to the covariance calculation formula. Similarly, for the temperature field and acoustic emission signal, stress field and magnetic field, stress field and acoustic emission signal, and magnetic field and acoustic emission signal, the corresponding physical field feature data are extracted, and the covariance value is calculated. By performing this calculation for each time segment, six sets of cross-physical field covariance sequences can be obtained. Each set of sequences records the changes in the covariance values ​​of the corresponding two physical field features in different time segments.

[0043] Step S234: Perform difference operations on the covariance values ​​corresponding to adjacent time segments, calculate the covariance change rate, which is the difference between the covariance value of the next segment and the covariance value of the previous segment divided by the length of the time segment.

[0044] By performing difference operations on the covariance values ​​corresponding to adjacent time segments, the changes in covariance can be obtained. The rate of change of covariance is an indicator that measures how quickly the covariance changes over time, reflecting the degree of change in the correlation between different physical field characteristics within adjacent time segments. When calculating the rate of change of covariance, for each set of cross-physical field covariance sequences, starting from the second time segment, the difference between the covariance value of the subsequent segment and the covariance value of the previous segment is calculated sequentially. Then, this difference is divided by the length of the time segment to obtain the rate of change of covariance. For example, if the covariance value of the first time segment is C1, the covariance value of the second time segment is C2, and the time segment length is T, then the rate of change of covariance is (C2-C1) / T. By performing the above calculation on all adjacent time segments, the sequence of the rate of change of covariance for each set of cross-physical field covariance sequences can be obtained.

[0045] Step S235: Construct a correlation calculation function based on the covariance value and the rate of change of covariance. The correlation is obtained by weighted summation of the covariance value and the rate of change. The weight coefficient is determined based on the contribution of the correlation to the production anomaly early warning in historical data.

[0046] The correlation calculation function comprehensively considers both the covariance value and the rate of change of covariance. The weighted summation involves multiplying the covariance value and the rate of change of covariance by their respective weighting coefficients, and then summing the results. The weighting coefficients are determined based on the contribution of correlation to production anomaly early warning in historical data; the higher the contribution, the larger the corresponding weighting coefficient.

[0047] When constructing the correlation calculation function, let the covariance value be C, the rate of change of covariance be R, and the weighting coefficients be w1 and w2, respectively. Then, the formula for calculating the correlation degree D is D = w1C + w2R. The weighting coefficients w1 and w2 can be determined based on historical data analysis. For example, by analyzing historical production data, the contribution of the covariance value and the rate of change of covariance to anomaly warnings can be statistically analyzed when production anomalies occur. If the covariance value plays a more important role in anomaly warnings, then the value of w1 can be set relatively large; if the rate of change of covariance is more critical to anomaly warnings, then the value of w2 can be appropriately increased. In this way, the weighting coefficients are determined based on the contribution of the correlation degree to production anomaly warnings in historical data, and then the correlation degree is obtained by weighted summation of the covariance value and the rate of change.

[0048] Step S236: Arrange the correlation degree of each time segment according to the production time sequence to generate a time-varying curve set of correlation strength. Each time-varying curve of correlation strength in the time-varying curve set corresponds to a set of correlation strength changes over time across physical field characteristics.

[0049] The time-varying correlation strength curve set is a collection of multiple time-varying correlation strength curves. Each curve records the change in the correlation strength across a set of physical field features as the production process progresses. By arranging the correlation degrees of each time segment according to the production time sequence, the dynamic evolution of the correlation strength between different physical field features during the production process can be visually displayed.

[0050] When generating the time-varying curve set of correlation strength, for each group of cross-physical field features, the calculated correlation degrees of each time segment are arranged according to the production time sequence. For example, the correlation degrees of the first time segment and the second time segment are arranged sequentially. Then, a time-varying curve of correlation strength is plotted with time as the horizontal axis and correlation degree as the vertical axis. For six groups of cross-physical field features, six time-varying curves of correlation strength can be obtained, which form the set of time-varying curves of correlation strength. By observing the set of time-varying curves of correlation strength, the changing trend of correlation strength between different physical field features with the production process can be clearly seen.

[0051] Step S240: Adjust the hidden layer connection topology of the neural network according to the time-varying curve of the correlation strength, so that the connection weights between network nodes increase or decrease with the fluctuation of the time-varying curve of the correlation strength, and construct a coupled mapping network framework with topological structure.

[0052] The time-varying correlation strength curve records how the correlation strength between different physical field features changes over time. By adjusting the hidden layer connection topology of a neural network based on the time-varying correlation strength curve, the neural network can better capture the correlation relationships between different physical field features. The coupling mapping network framework is a network structure that couples and maps different physical field features, which can more effectively handle the complex relationships between multi-physics data.

[0053] When adjusting the hidden layer connection topology of a neural network based on the time-varying correlation strength curve, the fluctuations of the correlation strength curve are first analyzed. When the correlation strength increases, the connection weights between network nodes are increased accordingly to strengthen information transmission and correlation between different physical field features. When the correlation strength decreases, the connection weights are decreased to reduce the correlation between features. For example, if the correlation strength between temperature and stress fields increases within a certain time period, the connection weights between hidden layer nodes related to temperature and stress fields are increased. By continuously adjusting the connection weights according to the fluctuations of the time-varying correlation strength curve, a coupled mapping network framework with a topological structure is constructed. This network framework can better adapt to the dynamic correlations between different physical field features, improving the processing capability of multi-physics data.

[0054] Step S250: Construct a training sample sequence by synchronously collecting video streams of micro-organizational structure evolution and macro-production status recording curves during historical production processes. Input the training sample sequence into the coupling mapping network framework segment by segment according to the production time sequence. Correct the network connection weights through the backpropagation algorithm to generate a preliminary physical field coupling mapping model.

[0055] The microstructure evolution video stream records the changes in the microstructure of steel over time during the production process, providing a visual representation of the evolution of microstructures such as grain size and phase composition within the steel. The macroscopic production state recording curve reflects the changes in macroscopic indicators (such as rolling force and cooling rate) over time during steel production. The training sample sequence is a set of samples composed of microstructure evolution characteristics and macroscopic production state evolution characteristics, used to train the coupled mapping network framework. The preliminary physics field coupled mapping model is a model obtained after training, capable of initially reflecting the nonlinear mapping relationship between the microstructure evolution trajectory and the macroscopic production state change trend.

[0056] In one implementation, step S250 may specifically include the following steps S251 to S256: Step S251: Retrieve the microstructure evolution video stream and macro production status recording curve corresponding to the same production batch from the historical production database. The microstructure evolution video stream contains a sequence of metallographic microscope images collected at preset time intervals, and the macro production status recording curve contains data on the changes in rolling force, cooling rate and plate shape parameters at the same time interval.

[0057] The historical production database stores various data from the steel production process, recording detailed information for different production batches. The microstructure evolution video stream consists of a series of metallographic microscope images acquired at preset time intervals, allowing observation of changes in the steel's microstructure over time. The macroscopic production status recording curves record the changes in macroscopic indicators such as rolling force, cooling rate, and plate shape parameters over time, reflecting the macroscopic state of the steel production process.

[0058] When retrieving data from the historical production database, the video streams of the micro-organizational structure evolution and the macro-production status record curves corresponding to the same production batch are filtered out based on the identification information of the production batch.

[0059] Step S252: Extract image features from the video stream of microstructure evolution, track the movement trajectory of grain boundaries using an edge detection algorithm, calculate the phase composition ratio change rate using a grayscale analysis algorithm, and generate a microstructure evolution feature sequence.

[0060] Edge detection algorithms can track the movement of grain boundaries, thus understanding the growth and changes of grains during the production process. Gray-scale analysis algorithms obtain image information by analyzing the gray-scale values ​​of an image. These algorithms can calculate the rate of change in phase composition ratios, reflecting the changes in the content of different phases in the steel's microstructure. The microstructure evolution feature sequence is a sequence of microstructure features that records the evolution of the steel's microstructure during the production process.

[0061] When extracting image features from a video stream analyzing the evolution of microstructures, each metallurgical microscope image in the video stream is first processed. Edge detection algorithms, such as the Canny edge detection algorithm, are used to detect edges and identify grain boundaries. Then, by comparing images at different time points, the movement trajectory of the grain boundaries is tracked. For grayscale analysis algorithms, the proportion of different grayscale regions in the image is calculated, and the rate of change of phase composition proportion is calculated based on the correspondence between grayscale and phase composition. The extracted features, such as the grain boundary movement trajectory and the rate of change of phase composition proportion, are arranged in chronological order to generate a sequence of microstructure evolution features.

[0062] Step S253: Extract trends from the macroscopic production state recording curves, eliminate high-frequency noise by using sliding window filtering, and extract the trends of rolling force change, cooling rate change, and plate shape parameter change by using a trend line fitting algorithm to generate a macroscopic production state evolution feature sequence.

[0063] Trend extraction involves extracting long-term trends from time-series data. Sliding window filtering eliminates high-frequency noise by sliding a fixed-size window across the time-series data and performing averaging or other calculations on the data within the window. Trend line fitting algorithms can more clearly display data trends through trend line fitting. The macroscopic production state evolution feature sequence is a sequence composed of features such as the trends in rolling force, cooling rate, and plate shape parameters, reflecting the evolution of macroscopic states during steel production.

[0064] When extracting trends from macroscopic production state record curves, a sliding window filter is first used to process the curves. For example, a sliding window of size 5 is selected, and the rolling force data is averaged to eliminate high-frequency noise and smooth the curve. Then, a trend line fitting algorithm, such as least squares fitting, is used to fit the processed curve and extract the rolling force variation trend. The same method is used to process the cooling rate and shape parameter variation data to extract the cooling rate variation trend and the shape parameter variation trend. These extracted trends are arranged in chronological order to generate a macroscopic production state evolution feature sequence.

[0065] Step S254: Align the micro-organizational structure evolution feature sequence with the macro-production state evolution feature sequence by timestamp to obtain micro-macro feature pairs, and arrange them according to production time sequence to form a training sample sequence. Each sample in the training sample sequence contains multi-physics field feature data of the previous time segment and the micro-macro feature pair corresponding to the current time point.

[0066] Timestamps are identifiers that record the time of data collection. By aligning the sequences of microscopic organizational structure evolution features and macroscopic production state evolution features according to timestamps, the temporal correspondence between microscopic and macroscopic features can be ensured. A microscopic-macroscopic feature pair consists of a microscopic organizational structure evolution feature and a macroscopic production state evolution feature, reflecting the correspondence between the microscopic structure and macroscopic state of steel at the same point in time. The training sample sequence is a set of samples used to train the coupled mapping network framework. Each sample contains multiphysics feature data from the previous time segment and the corresponding microscopic-macroscopic feature pair at the current time point. This allows the network to learn the correlation between multiphysics data and microscopic-macroscopic features.

[0067] When aligning the micro-organizational structure evolution feature sequence with the macro-production state evolution feature sequence by timestamp, micro-features and macro-features at the same time point are combined into micro-macro-feature pairs based on the timestamp information. For example, at time point t, the corresponding feature in the micro-organizational structure evolution feature sequence is M. t The corresponding feature in the macroscopic production state evolution feature sequence is H. t They are then combined into micro-macro feature pairs (M t H t Then, these micro-macro feature pairs are arranged according to production time sequence, while multiphysics feature data from the previous time segment are added to each sample to generate a training sample sequence.

[0068] Step S255: Divide the training sample sequence into consecutive training segments according to the production time sequence. Each segment contains a preset number of samples. Input the training segments into the coupled mapping network framework segment by segment and calculate the micro-macro feature prediction values ​​output by the model through forward propagation.

[0069] A training segment is a continuous set of samples divided according to the production time sequence, with each segment containing a predetermined number of samples. Forward propagation is the process of transferring information from the input layer to the output layer in a neural network; the output of the model can be calculated through forward propagation. Micro-macro feature predictions are the values ​​of micro-organizational structure evolution features and macro-production state evolution features predicted by the coupled mapping network framework based on the input training segment data.

[0070] When dividing the training segments, based on a preset number of samples, consecutive training segments are sequentially divided starting from the beginning of the training sample sequence. Each training segment is then input into the coupled mapping network framework for forward propagation. The network calculates the predicted micro-macro feature values ​​based on the input multiphysics feature data and micro-macro feature pairs through the connection weights between neurons, ultimately outputting the predicted micro-macro feature values.

[0071] Step S256: Compare the predicted values ​​of micro-macro features with the actual micro-macro feature pairs, calculate the prediction error, and correct the network connection weights along the time axis using the backpropagation algorithm. Update the weights once for each input training segment until all training segments are processed, and generate a preliminary physical field coupling mapping model.

[0072] Prediction error is the difference between the predicted micro-macro feature values ​​output by the model and the actual micro-macro feature pairs. Calculating the prediction error allows for the evaluation of model performance. Backpropagation is an optimization algorithm used for neural network training. It calculates the error gradient based on the prediction error and propagates this gradient back into the network, adjusting the network's connection weights to make the model's output closer to the true values. In this step, the network connection weights are corrected along the time axis using the backpropagation algorithm, updating the weights after each training segment input, gradually optimizing the network's performance. The preliminary physics coupling mapping model is a model obtained after training that can initially reflect the nonlinear mapping relationship between the evolution trajectory of the micro-organizational structure and the changing trend of the macro-production state.

[0073] When comparing the predicted micro-macro feature pairs with the actual micro-macro feature pairs, the differences between them are calculated. For example, the mean squared error (MSE) is used to calculate the prediction error. Based on the calculated prediction error, the error gradient is calculated using the backpropagation algorithm. Then, the error gradient is backpropagated along the time axis into the network, adjusting the network's connection weights. The connection weights are updated once for each training segment input. This process is repeated until all training segments have been processed, at which point a preliminary physics coupling mapping model is generated.

[0074] Step S260: Optimize the preliminary physical field coupling mapping model. Adjust the network topology based on the latest collected multi-physics field collaborative sensing data and the prediction deviation of the model output until the fluctuation amplitude of the prediction deviation is lower than the convergence threshold within a continuous preset time period, and determine the final physical field coupling mapping model.

[0075] Prediction bias is the difference between the latest acquired multi-physics collaborative sensing data and the predicted value output by the model, reflecting the model's prediction accuracy. Network topology refers to the connection methods and structures between neurons in a neural network; adjusting the network topology can change the network's performance and function. The convergence threshold is a pre-set standard value; when the fluctuation range of the prediction bias is below this threshold, it indicates that the model has converged and achieved good performance.

[0076] When optimizing the preliminary physics coupling mapping model, the latest multi-physics collaborative sensing data is first collected and input into the preliminary physics coupling mapping model to obtain the model's output prediction value. Then, the prediction deviation between the latest collected data and the prediction value is calculated. Based on the prediction deviation, potential problems in the network topology are analyzed. If the prediction deviation is large, it may be necessary to adjust the network topology by adjusting the network connection weights, increasing or decreasing the number of neurons, etc. This process is repeated continuously, and the fluctuation range of the prediction deviation is continuously observed. When the fluctuation range of the prediction deviation is lower than the convergence threshold within a preset time period, it indicates that the model has converged, and the final physics coupling mapping model is determined at this point.

[0077] Step S300: Through the physical field coupling mapping model, based on the nonlinear mapping relationship, perform continuous feature correlation analysis on the collected multi-physical field collaborative sensing data set to generate production status assessment results.

[0078] The physical field coupling mapping model is a trained and optimized model that reflects the nonlinear mapping relationship between the evolution trajectory of micro-organism structure and the trend of macro-production state changes. Continuous feature correlation analysis involves the continuous monitoring and analysis of the correlation relationships between different physical field features in the collected multi-physics collaborative sensing data set. The production state assessment results provide comprehensive information on the steel production state, helping production managers understand whether there are any abnormalities in the production process, as well as the degree and scope of these abnormalities.

[0079] As one implementation method, the production status assessment results may include the production status assessment results of the anomaly start time, development rate, and scope of impact. Therefore, step S300 may specifically include the following steps S310~S380: Step S310: Input the collected multi-physics field collaborative sensing data set into the preprocessing layer of the physical field coupling mapping model, perform spatiotemporal registration processing consistent with the training phase, and generate a spatiotemporal registration data stream.

[0080] The preprocessing layer is a layer in the physics-field coupling mapping model used to preprocess the input data to improve the model's processing efficiency and accuracy. Spatiotemporal registration is the process of uniformly calibrating data from different physics fields in time and space. In this step, the same spatiotemporal registration process as in the training phase is performed to ensure that the input data and training data have the same spatiotemporal reference frame. The spatiotemporal registration data stream is the data stream generated after spatiotemporal registration processing, containing all physics field data under a unified spatiotemporal reference frame.

[0081] When inputting the collected multi-physics collaborative sensing data set into the preprocessing layer of the physics coupling mapping model, the spatiotemporal registration processing method used in the training phase is followed. For example, the original acquisition timestamps and spatial coordinate information of each physics field data are extracted, a spatiotemporal coordinate matrix is ​​constructed, and the process node number corresponding to each data point is marked. The current sampling frequency of each sensor is monitored, the sampling frequency fluctuation value is calculated, and when the fluctuation value exceeds a preset fluctuation threshold, the sliding step size of the time window is adjusted according to the fluctuation ratio. For temperature field distribution data, Kriging interpolation is used to supplement missing coordinate point data in the spatial dimension, and linear interpolation is used to supplement missing time point data in the temporal dimension. The stress field distribution data, magnetic field strength change data, and acoustic emission signal data are processed accordingly. Finally, the spatiotemporal continuous data of each physics field are aligned with the coordinates and timestamps of a unified spatiotemporal reference system to generate a spatiotemporal registration data stream.

[0082] Step S320: Call the physical field coupling mapping model to load the latest correlation strength time-varying curve, perform cross-physical field feature coupling on the spatiotemporal registration data stream, obtain the correlation relationship of different physical field features by adjusting the coupling weight, and generate the correlation feature stream.

[0083] The time-varying correlation strength curve records how the correlation strength between different physical field features changes over time. Loading the latest correlation strength curve allows the model to process data based on the latest correlation information. Cross-physical field feature coupling is the process of combining and associating features from different physical fields. By adjusting the coupling weights, the degree of correlation between different physical field features can be changed. The correlated feature stream is a data stream containing correlation information between different physical field features, generated after cross-physical field feature coupling.

[0084] In one implementation, step S320 may specifically include the following steps S321 to S325: Step S321: The physical field coupling mapping model receives the latest time-varying curve of correlation strength through the internal weight update module and analyzes the cross-physical field correlation strength value corresponding to each time point in the time-varying curve of correlation strength.

[0085] The internal weight update module is a component of the physics-field coupling mapping model. It receives and processes the latest time-varying correlation strength curve information and updates the network's coupling weights based on the correlation strength values ​​in the curves. The cross-physics-field correlation strength value is the correlation strength between different physics-field features recorded in the time-varying correlation strength curve at various time points. The internal weight update module of the physics-field coupling mapping model receives the latest time-varying correlation strength curve data. Then, it analyzes the curves to extract the cross-physics-field correlation strength values ​​corresponding to each time point. For example, by analyzing the data format of the curves, it finds the correlation strength values ​​between different physics-field features such as temperature field and stress field, and temperature field and magnetic field at each time point.

[0086] Step S322: Divide the spatiotemporal registration data stream into data segments that match the time granularity of the correlation intensity time-varying curve in chronological order. Each data segment contains multiphysics feature data within the current time window.

[0087] Temporal granularity refers to the resolution of data over time. Dividing the spatiotemporal registration data stream into segments that match the temporal granularity of the correlation strength time-varying curve ensures a temporal correspondence between the data segments and the correlation strength values. Each data segment contains multiphysics characteristic data within the current time window, facilitating subsequent data processing based on the correlation strength values. When segmenting the spatiotemporal registration data stream, the time length of each data segment is determined based on the temporal granularity of the correlation strength time-varying curve.

[0088] Step S323: For each data segment, extract the correlation strength value at the corresponding time point from the correlation strength time-varying curve to construct a coupling weight matrix. The element values ​​in the coupling weight matrix are updated with the correlation strength value.

[0089] The coupling weight matrix is ​​used to weight and combine different physical field features. The element values ​​in the matrix represent the coupling weights between different physical field features. By extracting the correlation strength values ​​at corresponding time points from the time-varying correlation strength curve to construct the coupling weight matrix, the coupling weights can be updated as the correlation strength changes, thus better reflecting the correlation relationships between different physical field features.

[0090] For each data segment, based on the time point of the data segment, the cross-physical field correlation strength value at the corresponding time point is extracted from the time-varying correlation strength curve. For example, for a data segment at time point t, the correlation strength values ​​between different physical field characteristics such as temperature field and stress field, and temperature field and magnetic field at time point t are extracted. These correlation strength values ​​are arranged into a matrix according to certain rules; this matrix is ​​the coupling weight matrix. When the correlation strength values ​​change, the element values ​​in the coupling weight matrix are also updated accordingly.

[0091] Step S324: Combine the temperature field features, stress field features, magnetic field features and acoustic emission signal feature vectors in the data segment with the coupling weight matrix to generate cross-physical field coupling feature vectors. During the coupling process, the rows and columns of the weight matrix correspond to different physical field feature dimensions.

[0092] Weighted combination is the process of multiplying different physical field feature vectors with a coupling weight matrix to obtain a cross-physical field coupled feature vector. Temperature field feature vectors, stress field feature vectors, magnetic field feature vectors, and acoustic emission signal feature vectors are vectors extracted from data segments representing different physical field features. The rows and columns of the coupling weight matrix correspond to different physical field feature dimensions. Through this correspondence, a weighted combination of different physical field features can be achieved. During weighted combination, the temperature field feature vector, stress field feature vector, magnetic field feature vector, and acoustic emission signal feature vector in the data segment are multiplied with the coupling weight matrix. For example, if the temperature field feature vector is T, the stress field feature vector is S, the magnetic field feature vector is M, the acoustic emission signal feature vector is A, and the coupling weight matrix is ​​W, then the formula for calculating the cross-physical field coupled feature vector C is C = W × [T; S; M; A]. T , where [T;S;M;A] T This means arranging the four eigenvectors into a matrix by columns, and generating cross-physical field coupled eigenvectors through such weighted combination.

[0093] Step S325: Concatenate the cross-physical field coupling feature vectors of continuous time segments in chronological order to obtain a temporally correlated feature stream. The temporal resolution of the correlated feature stream is consistent with that of the spatiotemporal registration data stream, and the dimension of the feature vector at each time point matches the dimension of the coupling weight matrix.

[0094] In this step, cross-physics coupling feature vectors from consecutive time segments are concatenated in chronological order to form a temporally correlated feature stream. Temporal resolution refers to the precision of the data over time; the temporal resolution of the correlated feature stream is consistent with the spatiotemporal registration data stream, ensuring the continuity and consistency of the data over time. The dimension of the feature vector at each time point matches the dimension of the coupling weight matrix, guaranteeing the effectiveness and accuracy of the correlated feature stream.

[0095] When concatenating cross-physics coupling feature vectors, the feature vectors of subsequent time segments are connected sequentially, starting with the cross-physics coupling feature vector of the first time segment. Since the cross-physics coupling feature vector of each time segment is generated based on the coupling weight matrix, the dimension of the feature vector at each time point matches the dimension of the coupling weight matrix. The temporal resolution of the correlated feature stream is the same as that of the spatiotemporal registration data stream, enabling the correlated feature stream to accurately reflect the temporal changes in the correlation between different physical field features.

[0096] Step S330: Input the associated feature flow into the nonlinear mapping layer of the physical field coupling mapping model, and calculate the micro-organization structure evolution prediction trajectory and macro-production state prediction trend line through the neural network of topological structure.

[0097] The nonlinear mapping layer is responsible for realizing the nonlinear mapping relationship between the evolution trajectory of the microstructure and the trend of changes in the macroscopic production state. The topologically structured neural network can better handle the complex relationships between multi-physics data. The predicted trajectory of the microstructure evolution is the future evolution path of the steel microstructure predicted based on the associated feature flow of the input. The predicted trend line of the macroscopic production state is the predicted trend of changes in macroscopic indicators (such as rolling force, cooling rate, etc.) over time during the steel production process.

[0098] After inputting the associated feature stream into the nonlinear mapping layer of the physical field coupling mapping model, the neural network with the topological structure performs calculations based on the input associated features. The neurons in the network perform weighted summation of the input features through connection weights, and after processing by an activation function, ultimately outputting a predicted trajectory for the evolution of the microstructure and a predicted trend line for the macroscopic production state. For example, based on the correlation between physical field features such as temperature and stress fields, the network predicts the grain growth in the microstructure of steel and the changing trend of rolling force in the macroscopic production state over a future period.

[0099] Step S340: Compare the predicted trajectory of microstructural evolution with the preset standard evolution trajectory in a rolling manner, and calculate the trajectory deviation through a sliding time window. The window size is adjusted with the prediction duration, and the deviation weight of the recent trajectory is higher than that of the long-term trajectory.

[0100] The preset standard evolution trajectory is the evolution trajectory of the steel microstructure during normal production, determined based on process standards and experience. It serves as a reference standard for comparison with the predicted microstructure evolution trajectory. Rolling comparison refers to continuously comparing the predicted trajectory and the standard trajectory by constantly moving the window along the time axis. A sliding time window is a window that slides along the time axis; by calculating the trajectory deviation within the window, the differences between the predicted trajectory and the standard trajectory at different time periods can be understood. Trajectory deviation is an indicator that measures the degree of difference between the predicted trajectory and the standard trajectory; the deviation weight of recent trajectories is higher than that of long-term trajectories, indicating a greater focus on recent trajectory differences.

[0101] During the rolling comparison, the initial size of the sliding time window is first determined. The window size is adjusted according to the prediction duration; as the prediction duration increases, the window size increases by a preset ratio, ensuring that the length of the predicted trajectory covered by the window is positively correlated with the prediction duration. For example, when the prediction duration is 1 hour, the window size is 10 minutes; when the prediction duration increases to 2 hours, the window size is adjusted to 20 minutes. Within each sliding time window, the trajectory deviation between the predicted trajectory of microstructural evolution and the preset standard evolution trajectory is calculated. To ensure that the deviation of recent trajectories has a higher weight than that of distant trajectories, the trajectory deviations are weighted and summed, with the weight coefficients increasing in reverse chronological order. For example, within a 10-minute window, the trajectory deviation weight for the last minute is 0.5, the weight for the second-to-last minute is 0.3, and so on. Through this rolling comparison and calculation, a trajectory deviation sequence is obtained.

[0102] In one implementation, step S340 may specifically include the following steps S341 to S345: Step S341: Retrieve the standard evolution trajectory of the microstructure corresponding to the current steel grade from the process standard database. The standard evolution trajectory includes grain growth trajectory, phase transformation trajectory and dislocation density change trajectory.

[0103] The process standard database stores data on steel production process standards, containing various process standard information for different steel grades during normal production. The microstructure standard evolution trajectory is the evolution path of the microstructure under normal conditions, determined for the currently produced steel grade. This includes information such as grain growth trajectory, phase transformation trajectory, and dislocation density change trajectory. The grain growth trajectory describes the changes in grain size and shape during production; the phase transformation trajectory records the transformation process between different phases in the steel microstructure; and the dislocation density change trajectory reflects the change in dislocation density within the steel over time.

[0104] When retrieving the standard evolution trajectory of microstructure from the process standard database, the corresponding standard evolution trajectory is searched in the database based on the identification information of the steel grade currently being produced. For example, if the steel grade currently being produced is Q235, the standard evolution trajectory of the microstructure of Q235 steel is retrieved from the database. This trajectory includes detailed information such as grain growth trajectory, phase transformation trajectory, and dislocation density change trajectory.

[0105] Step S342: Initialize the sliding time window size as the base window size. Increase the window size by a preset ratio as the prediction duration increases, so that the length of the predicted trajectory covered by the window is positively correlated with the prediction duration.

[0106] The base window size is the initial size of the pre-set sliding time window, providing a benchmark for window size adjustments. The prediction duration is the predicted length of a future period in the steel production process. As the prediction duration increases, the window size needs to be increased proportionally to more comprehensively compare the predicted trajectory of microstructure evolution with the standard evolution trajectory. This ensures that the length of the predicted trajectory covered by the window is positively correlated with the prediction duration, guaranteeing reasonable calculation of trajectory deviation under different prediction durations.

[0107] When initializing the sliding time window, its size is set to the base window size, for example, 5 minutes. As the prediction duration increases, assuming a preset ratio of 2 minutes for every additional hour of prediction duration, the window size increases from 5 minutes to 7 minutes. This ensures the window covers the predicted trajectory length matching the prediction duration, resulting in more accurate trajectory comparison.

[0108] Step S343: Input the predicted trajectory of microstructural evolution and the standard evolution trajectory into the trajectory comparison module, calculate the trajectory coordinate difference at the corresponding time point within the sliding time window, and generate a coordinate difference sequence.

[0109] The trajectory comparison module receives the predicted trajectory and the standard evolution trajectory of the microstructure as input. Within a sliding time window, the coordinates of corresponding time points of the two trajectories are compared, and the difference between them is calculated. The coordinate difference reflects the degree of deviation between the predicted trajectory and the standard trajectory at each time point. Arranging these differences in chronological order generates a coordinate difference sequence.

[0110] After the predicted trajectory of microstructural evolution is input into the trajectory comparison module, the module compares the coordinates of the two trajectories point by point within a sliding time window. For example, at a certain time t, the coordinates of the predicted trajectory are (x1, y1, z1), and the coordinates of the standard trajectory are (x2, y2, z2), then the coordinate difference is calculated as (x1-x2, y1-y2, z1-z2). This calculation is performed for each time point within the window, ultimately yielding a sequence of coordinate differences. This sequence visually demonstrates the deviation between the predicted trajectory and the standard trajectory within the sliding time window.

[0111] Step S344: Perform a weighted summation on the coordinate difference sequence, with the weight coefficients increasing in reverse chronological order to make the coordinate difference weight of recent trajectories higher than that of distant trajectories, thereby generating the trajectory deviation within the window.

[0112] In this step, in order to pay more attention to the deviation of recent trajectories, the weight coefficient is set to increase in reverse chronological order, that is, the weight of the coordinate difference of recent trajectories is higher. The trajectory deviation degree within the window is a comprehensive index that reflects the overall deviation degree between the predicted trajectory and the standard trajectory within the current sliding time window.

[0113] When performing a weighted sum on the coordinate difference sequence, assume that there are n time points within the sliding time window, the coordinate difference sequence is [d1, d2, …, dn], the weight coefficient sequence is [w1, w2, …, wn], and w1 < w2 < … < wn. Then the calculation formula for the trajectory deviation degree D within the window is . Through the weighted sum method, the coordinate differences of recent trajectories contribute more to the final trajectory deviation degree, and can better reflect the actual deviation between the predicted trajectory of the current microstructure evolution and the standard trajectory.

[0114] Step S345: The sliding time window moves in chronological order, and the trajectory deviation degree within each window is repeatedly calculated to generate a trajectory deviation degree sequence. Each element in the trajectory deviation degree sequence corresponds to the trajectory deviation degree value of a time window.

[0115] The sliding time window moving in chronological order means that the window slides backward sequentially on the time axis. The step size of each slide can be set according to the actual situation. Within each new sliding time window, the operations of steps S343 and S344 are repeated, that is, the trajectory coordinate differences at the corresponding time points are calculated, and a weighted sum is performed on the coordinate difference sequence to obtain the trajectory deviation degree within this window. Arranging the trajectory deviation degrees of each window in chronological order generates the trajectory deviation degree sequence.

[0116] When the sliding time window starts to move, starting from the initial position, the trajectory deviation degree within the window is calculated once every time it moves. For example, the trajectory deviation degree of the first window is D1, the trajectory deviation degree of the second window is D2, and so on. The finally obtained trajectory deviation degree sequence [D1, D2, …, Dm] can clearly show the change trend of the deviation between the predicted trajectory of the microstructure evolution and the standard trajectory over time in different time periods.

[0117] Step S350: The predicted trend line of the macro production state is compared with the preset standard trend line in a rolling manner, and the trend deviation degree is calculated using the same sliding time window to generate a macro trend deviation degree sequence.

[0118] The preset standard trend line is a trend line representing the change of the macroscopic production state under normal conditions during steel production, determined based on process standards and historical data. It serves as a reference standard for comparison with the predicted macroscopic production state trend line. The rolling comparison involves continuously moving the window along the time axis to compare the two trend lines. Using the same sliding time window as in step S340 ensures that microscopic and macroscopic data are analyzed on the same time scale. The trend deviation is an indicator that measures the degree of difference between the predicted macroscopic production state trend line and the standard trend line. Arranging the trend deviations calculated within each window in chronological order generates a macroscopic trend deviation sequence.

[0119] During the rolling comparison, the predicted trend line of the macroscopic production state and the preset standard trend line are input into the comparison module. Within the sliding time window, the numerical difference between the two trend lines at corresponding time points is calculated, for macroscopic indicators such as rolling force and cooling rate. Similar to calculating the deviation of the microscopic trajectory, these differences are weighted and summed, with the weighting coefficients increasing in reverse chronological order to highlight the deviation of the recent trend. As the sliding time window moves chronologically, the trend deviation within each window is repeatedly calculated, ultimately generating a macroscopic trend deviation sequence. This sequence reflects how the degree of deviation between the macroscopic production state and the standard state changes over time in different time periods.

[0120] Step S360: Generate a comprehensive deviation index by fusing the trajectory deviation and macro trend deviation sequences. When the comprehensive deviation index exceeds the preset warning threshold, determine the start time of the anomaly by tracing back the change process of the comprehensive deviation index.

[0121] The comprehensive deviation index is a composite indicator that integrates the deviation of the micro-organizational structure evolution trajectory and the trend deviation of the macro-production state, providing a more comprehensive reflection of the overall deviation in the steel production process. The preset warning threshold is a pre-defined standard value; when the comprehensive deviation index exceeds this threshold, it indicates a potential anomaly in the production process. Tracing the changes in the comprehensive deviation index involves looking back from the current moment to the point where the index began to show abnormal growth; this moment is the anomaly initiation point.

[0122] When integrating the trajectory deviation and macro-trend deviation sequences, a weighted average method can be used. Assuming the trajectory deviation sequence is [D1, D2, ..., Dm] and the macro-trend deviation sequence is [E1, E2, ..., Em], with weight coefficients a and b (a+b=1) respectively, the formula for calculating the comprehensive deviation index sequence [F1, F2, ..., Fm] is Fi=aDi+bEi. When the comprehensive deviation index Fi exceeds a preset warning threshold at a certain moment, the changes in the comprehensive deviation index are traced back from that moment. For example, by observing the growth trend and slope of the index, the moment when the comprehensive deviation index begins to rise significantly can be identified as the anomaly initiation moment.

[0123] Step S370: Calculate the anomaly development rate based on the rate of change of the comprehensive deviation index after the anomaly initiation time, predict the range of process nodes that may be affected within a preset time period by the spatial distribution relationship between the anomaly development rate and the current production process nodes, and determine the scope of the anomaly impact.

[0124] The anomaly development rate reflects the speed at which an anomaly develops during the production process and is calculated based on the rate of change of the comprehensive deviation index after the anomaly's inception. The spatial distribution of current production process nodes describes the location and interrelationships of each node within the production flow. By analyzing the anomaly development rate and the spatial distribution of process nodes, it is possible to predict which process nodes the anomaly might spread to within a predetermined timeframe, thus determining the scope of its impact. In calculating the anomaly development rate, data after the anomaly's inception time is extracted from the comprehensive deviation index sequence. The difference between adjacent time points is calculated to obtain the comprehensive deviation rate of change sequence. A moving average filter is applied to the comprehensive deviation rate of change sequence to eliminate high-frequency fluctuations, resulting in a smoothed rate of change sequence. The average value of this smoothed rate of change sequence is then determined as the anomaly development rate.

[0125] In one implementation, step S370 may specifically include the following steps S371 to S375: Step S371: Extract the index data after the abnormal start time from the comprehensive deviation index sequence, calculate the index difference between adjacent time points, and generate the comprehensive deviation change rate sequence.

[0126] After determining the initial time of the anomaly, data from that time onwards are selected from the comprehensive deviation index sequence. The difference between the index values ​​at adjacent time points reflects the change in comprehensive deviation over time. Arranging these differences in chronological order generates a comprehensive deviation change rate sequence. This sequence visually demonstrates the trend of comprehensive deviation changes after the anomaly occurs.

[0127] Step S372: Perform a moving average filter on the comprehensive deviation rate of change sequence to eliminate high-frequency fluctuations and generate a smooth rate of change sequence. The average value of the smooth rate of change sequence is determined as the abnormal development rate.

[0128] Moving average filtering eliminates high-frequency fluctuations in the data by averaging the data within a fixed-size window that slides across the sequence. A smoothed rate of change sequence, obtained after moving average filtering, better reflects the overall trend of the rate of change in overall deviation. Using the average of the smoothed rate of change sequence as the anomaly development rate allows for a more accurate measurement of the speed at which anomalies develop.

[0129] When applying a moving average filter to the comprehensive deviation rate of change sequence, a window size is set, for example, 3. For the sequence [ The first smoothing value is... The second smoothing value is This process is repeated to obtain the smoothed rate of change sequence. Then, the average value of the smoothed rate of change sequence is calculated. For example, if the sequence is [S1, S2, ..., Sn], the average value is... This average value is defined as the abnormal growth rate.

[0130] Step S373: Obtain the spatial distribution map of the current production process nodes, mark the coordinates of the initial process node corresponding to the start time of the anomaly, and calculate the spatial distance and material transfer time between the initial node and adjacent process nodes.

[0131] The spatial distribution map of the current production process nodes shows the spatial relationship between each process node in the production process. Marking the initial process node coordinates corresponding to the anomaly initiation time can pinpoint the specific location where the anomaly started. Calculating the spatial distance between the initial node and adjacent process nodes, as well as the material transfer time, helps analyze the possibility and speed of the anomaly's spread between process nodes.

[0132] After obtaining the spatial distribution map of the current production process nodes, the corresponding initial process node is determined based on the anomaly initiation time, and its coordinates are marked. For example, in the steelmaking production process, the anomaly initiation time corresponds to the converter process node, and the coordinates of this node are marked. Then, the spatial distance between this initial node and adjacent process nodes (such as the refining furnace node, continuous casting machine node, etc.) is calculated. This can be calculated by measuring the Euclidean distance between the spatial coordinates. Simultaneously, based on the material transport method and speed in the production process, the time required for material to be transported from the initial node to adjacent nodes is calculated. For example, by querying the operating parameters and process route of the material transport equipment, the time required for material to be transported from the converter to the refining furnace is determined.

[0133] Step S374: Calculate the anomaly diffusion coefficient based on the anomaly development rate and material transport time. The diffusion coefficient is the product of the anomaly development rate and the material transport time, representing the probability that the anomaly will spread from the initial node to adjacent nodes.

[0134] The anomaly diffusion coefficient reflects the likelihood of an anomaly spreading from the initial process node to adjacent process nodes. The faster the anomaly develops and the longer the material transport time, the larger the anomaly diffusion coefficient, indicating a higher probability of anomaly diffusion.

[0135] The anomaly diffusion coefficient is calculated based on the anomaly development rate obtained in step S372 and the material transport time obtained in step S373. Let the anomaly development rate be V and the material transport time be T, then the anomaly diffusion coefficient D = V × T.

[0136] Step S375: Compare the abnormal diffusion coefficient with the preset diffusion threshold. When the diffusion coefficient exceeds the threshold, mark the corresponding adjacent process node as a potentially affected node. Repeat this process until all potentially affected nodes are marked to obtain the abnormal influence range.

[0137] The preset diffusion threshold is a pre-defined standard value used to determine whether an anomaly will spread to adjacent process nodes. When the anomaly diffusion coefficient exceeds this threshold, it indicates that the anomaly is highly likely to spread to that adjacent process node, and it is marked as a potentially affected node. By continuously repeating this comparison and marking process, all adjacent process nodes are evaluated, and the scope of the anomaly's impact is ultimately determined.

[0138] When comparing the abnormal diffusion coefficient with a preset diffusion threshold, assuming the preset threshold is 0.8, if the abnormal diffusion coefficient of a neighboring process node is 1.2, exceeding the threshold, then that node is marked as a potentially affected node. Then, the same comparison and marking operation is performed on other neighboring process nodes. For nodes marked as potentially affected, the abnormal diffusion of their neighboring nodes is analyzed until all potentially affected nodes have been marked. Ultimately, the set of all marked nodes constitutes the scope of the abnormal impact.

[0139] Step S380: Integrate the anomaly start time, anomaly development rate, and anomaly impact range with the micro-organizational structure prediction trajectory and the macro-production status prediction trend line to generate production status assessment results.

[0140] The production status assessment result is a comprehensive evaluation of the overall status of the steel production process, which integrates information such as the onset time of the anomaly, the rate of development of the anomaly, the scope of the anomaly's impact, as well as the predicted trajectory of the micro-organism structure and the predicted trend line of the macro-production status.

[0141] When integrating this information, the anomaly's onset time, development rate, and impact range are correlated with the predicted trajectory of the micro-organizational structure and the predicted trend line of the macro-production status. For example, the location of the anomaly's onset time can be marked on the predicted trajectory of the micro-organizational structure and the predicted trend line of the macro-production status, while the anomaly's development rate and impact range are displayed as important supplementary information. In this way, the generated production status assessment results can more clearly reflect the anomalies in the production process and their impact on the micro and macro production status.

[0142] Step S400: Generate production parameter adjustment instructions with timing control logic based on the production status assessment results, and push the production parameter adjustment instructions to the steel production control system to execute the control operation.

[0143] The production status assessment results include abnormal information and trends in production status. Based on this information, production parameter adjustment instructions with time-series control logic are generated, allowing for targeted adjustments to the production process. The time-series control logic considers the temporal sequence of the production process and the order of parameter adjustments, ensuring that the adjustment instructions are executed reasonably and effectively. These production parameter adjustment instructions are then pushed to the steel production control system, enabling the system to adjust production parameters accordingly, thereby correcting abnormalities and ensuring normal production operation.

[0144] When generating production parameter adjustment instructions, the process first analyzes information from the production status assessment results, such as the anomaly's onset time, development rate, and impact range. Based on this information, the production parameters requiring adjustment, as well as the timing and magnitude of the adjustments, are determined. For example, if the anomaly's onset time indicates a temperature anomaly at a certain process node, and the anomaly's development rate is rapid, affecting subsequent process nodes, then it can be determined that parameters such as the heating equipment power at that process node need adjustment. Simultaneously, according to the timing control logic, the order and timing of parameter adjustments are determined. Finally, the specific information of the adjusted parameters (such as parameter name, adjustment value, and execution time) is used to generate the production parameter adjustment instruction, which is then pushed to the steel production control system.

[0145] In one implementation, step S400 may specifically include the following steps S410 to S470: Step S410: Analyze the production status assessment results to extract the anomaly start time, anomaly development rate, and anomaly impact range, and determine the list of target process nodes that need to be controlled based on the anomaly impact range.

[0146] Analyzing production status assessment results involves extracting and analyzing information from the assessment results. By extracting key information such as the anomaly's onset time, development rate, and impact range, the specific details of the anomaly in the production process can be understood. Based on the anomaly's impact range, a list of target process nodes requiring adjustment is determined, clarifying which process nodes' production parameters need to be adjusted to correct the anomaly in a timely manner. When analyzing the production status assessment results, information regarding the anomaly's onset time, development rate, and impact range is extracted. For example, the assessment results indicate that the anomaly started at 10:00 AM, the development rate was 0.3 (units determined based on specific indicators), and the impact range involved process nodes such as the converter, refining furnace, and continuous casting machine. Based on this information, the list of target process nodes requiring adjustment is determined as [converter, refining furnace, continuous casting machine].

[0147] Step S420: Query the list of adjustable parameters for each node in the target process node list. The list of adjustable parameters includes the parameter name, current value and historical adjustment records.

[0148] The adjustable parameter list is a detailed list of parameters that can be adjusted at the target process node, including parameter name, current value, and historical adjustment records. Querying the adjustable parameter list for each node helps to understand which parameters can be adjusted at each process node, as well as the current status and historical adjustment information of these parameters, providing a basis for subsequent parameter adjustments.

[0149] When querying the list of adjustable parameters for each node in the target process node list, relevant information for each node is obtained by accessing the production database or the parameter management module of the control system. For example, for the converter process node, the query yields a list of adjustable parameters including oxygen flow rate, furnace temperature setpoint, etc., with the current value of each parameter recorded in the list. It also includes historical adjustment records for these parameters; for instance, the oxygen flow rate was adjusted three times in the past week, with detailed records of the magnitude and timing of each adjustment.

[0150] Step S430: Calculate the urgency of parameter adjustment based on the abnormal development rate. The urgency of parameter adjustment is positively correlated with the abnormal development rate. The higher the urgency of parameter adjustment, the higher the priority of parameter adjustment.

[0151] Parameter adjustment urgency is an indicator of the degree of urgency for parameter adjustments and is positively correlated with the anomaly development rate. The faster the anomaly development rate, the greater the likelihood of the anomaly worsening, requiring more urgent parameter adjustments. Higher parameter adjustment urgency means higher priority for parameter adjustments, ensuring that the most critical parameters for correcting the anomaly are adjusted first within a limited timeframe. When calculating parameter adjustment urgency based on the anomaly development rate, a functional relationship can be established; for example, if the parameter adjustment urgency is P and the anomaly development rate is V, the functional relationship is P = kV (where k is a constant).

[0152] Step S440: Determine the effective window period based on the time difference between the anomaly start time and the current time. The length of the effective window period is the remaining time from the current time until the anomaly is expected to spread to the target node.

[0153] The effective window for parameter adjustment is a time range within which parameter adjustments can effectively correct anomalies. Based on the time difference between the anomaly's onset and the current time, combined with information such as the anomaly's development rate and impact range, the estimated time for the anomaly to spread to the target node can be predicted. The effective window length is the remaining time from the current time until the anomaly is expected to spread to the target node. Parameter adjustments made within this window can minimize the impact of the anomaly on production.

[0154] When determining the effective window period for parameter adjustment, the time difference between the anomaly's onset time and the current time is first calculated. Then, based on the anomaly's development rate and impact range, the expected time for the anomaly to spread to the target node is predicted. Assuming that the anomaly is expected to spread to the target node at 11:00 AM, the effective window period for parameter adjustment is 30 minutes (from 10:30 AM to 11:00 AM). Within this window period, the parameter adjustment operation needs to be completed as quickly as possible.

[0155] Step S450: Input the list of adjustable parameters, the urgency of parameter adjustment, and the effective window period into the parameter optimizer, and use a multi-objective evolutionary algorithm to solve for the parameter adjustment direction, magnitude, and execution sequence to generate a parameter adjustment scheme.

[0156] The parameter optimizer optimizes parameter tuning schemes, receiving information such as a list of adjustable parameters, the urgency of parameter tuning, and the effective window period as input. In this step, a multi-objective evolutionary algorithm is used to solve for the direction, magnitude, and execution timing of parameter tuning, ensuring that parameter tuning meets the needs of anomaly correction while also considering factors such as production efficiency and cost. The generated parameter tuning scheme clearly defines the tuning direction (increase or decrease), magnitude, and execution time for each parameter, providing specific guidance for adjusting production parameters.

[0157] After inputting the list of adjustable parameters, the urgency of parameter adjustment, and the effective window period into the parameter optimizer, the optimizer uses a multi-objective evolutionary algorithm to solve the problem. For example, the algorithm's objective is to minimize the rate of anomalous development while simultaneously minimizing the total magnitude of parameter adjustments. The algorithm iteratively searches for the optimal parameter adjustment scheme based on the parameter range and constraints in the list of adjustable parameters, as well as the limitations of the parameter adjustment urgency and effective window period. In each iteration, the algorithm generates a set of parameter adjustment schemes, evaluates the fitness of each scheme (i.e., the degree to which it satisfies the objective), and selects the scheme with the highest fitness for retention and evolution. After a predetermined number of generations of evolution, a compromise solution from the Pareto optimal solution set is selected from the final population to determine the adjustment direction, magnitude, and corresponding execution time for each parameter, thus generating the parameter adjustment scheme.

[0158] In one implementation, step S450 may specifically include the following steps S451 to S456: Step S451: Analyze the list of adjustable parameters to determine the optimization variables. Each parameter in the list of adjustable parameters is an optimization variable, and the value range of the optimization variable is the allowable adjustment range of the parameter.

[0159] Analyzing the list of adjustable parameters involves extracting and organizing the parameter information from the list. Each parameter in the list is designated as an optimization variable, thus transforming the parameter adjustment problem into an optimization problem. The range of values ​​for the optimization variables is determined based on the allowable adjustment range of the parameters. This ensures that when solving for parameter adjustment schemes, the adjusted values ​​of the parameters are within a reasonable range, avoiding new problems caused by over-adjustment.

[0160] When analyzing the list of adjustable parameters, each parameter in the list is analyzed. For example, for the oxygen flow rate and furnace temperature setpoint in the list of adjustable parameters for converter process nodes, the oxygen flow rate and furnace temperature setpoint are determined as optimization variables respectively.

[0161] Step S452: Construct a multi-objective optimization function with minimizing the rate of abnormal development as the primary optimization objective and minimizing the total sum of parameter adjustments as the secondary optimization objective. The weight of the objectives changes with the urgency of parameter adjustments; the higher the urgency, the greater the weight of the primary objective.

[0162] The multi-objective optimization function is a function that comprehensively considers multiple optimization objectives. In this step, the primary optimization objective is to minimize the anomaly development rate, and the secondary optimization objective is to minimize the total parameter adjustment magnitude. The objective weights change with the urgency of parameter adjustment; the higher the urgency, the greater the weight of the primary objective. This reflects that when the anomaly develops rapidly, the need to correct the anomaly is given priority.

[0163] When constructing the multi-objective optimization function, let the anomaly development rate be V, the total parameter adjustment magnitude be A, and the weight of the main objective be w.a The secondary objective has a weight of w. b (w) a +w b =1). Multi-objective optimization function F=w a V+w b A. Adjust the target weights based on the urgency of parameter adjustment. For example, when the urgency of parameter adjustment is 0.6, let w... a =0.7, w b =0.3; When the urgency of parameter adjustment is 0.3, let w a =0.5, w b =0.5. In this way, the need to correct anomalies and reduce adjustment magnitude can be reasonably balanced under different abnormal situations.

[0164] Step S453: Determine the time constraints of the optimization algorithm based on the effective window period, and divide the window period into continuous control periods.

[0165] The effective window period is the time range for parameter adjustment. Determining the time constraints of the optimization algorithm based on this window period ensures that the obtained parameter adjustment scheme is executed within the specified time. Dividing the window period into continuous adjustment periods facilitates parameter adjustment by the algorithm in different time periods, better meeting the timing requirements of the production process.

[0166] When determining the time constraints for the optimization algorithm based on the effective window period, assume the effective window period is from 10:30 AM to 11:00 AM, lasting 30 minutes. Divide this window period into consecutive adjustment periods, for example, into three 10-minute intervals. In the optimization algorithm, the execution time for parameter adjustments must be within these adjustment periods to ensure that parameter adjustment instructions are executed in a reasonable time sequence.

[0167] Step S454: Initialize the population for the multi-objective evolutionary algorithm. Each individual in the population represents a set of parameter adjustment schemes, including the adjustment direction, magnitude, and time period allocation of each parameter.

[0168] The population of a multi-objective evolutionary algorithm consists of a set of individuals, each representing a set of parameter adjustment schemes. During population initialization, a set of individuals is randomly generated, each containing the adjustment direction (increase or decrease), adjustment magnitude, and execution time allocation for each parameter. Random initialization covers different parameter adjustment possibilities, providing a broader search space for subsequent evolutionary searches.

[0169] During population initialization, for each individual, the adjustment direction of the adjustable parameters is randomly determined. For example, for the oxygen flow rate parameter, it is randomly determined whether to increase or decrease. Then, the adjustment range is randomly generated within the allowable adjustment range of the parameter. Simultaneously, according to the divided adjustment period, the execution period for each parameter adjustment is randomly assigned. In this way, a population containing multiple individuals is generated, with each individual representing a different set of parameter adjustment schemes.

[0170] Step S455: Evolve the population through selection, crossover, and mutation operations. Each generation of the population calculates the fitness value according to the optimization function, retains the individual with the highest fitness value, and satisfies the time constraint conditions.

[0171] Selection, crossover, and mutation are fundamental operations in multi-objective evolutionary algorithms. These operations continuously evolve the population, gradually bringing individuals closer to the optimal solution. Each generation calculates a fitness value based on the multi-objective optimization function; a higher fitness value indicates a better parameter adjustment scheme for that individual. The individuals with the highest fitness values ​​are retained, while ensuring these individuals meet time constraints. This guarantees that the evolved parameter adjustment scheme both meets the optimization objective and can be executed within the specified time.

[0172] During the selection operation, individuals in the population are ranked according to their fitness values, and individuals with higher fitness values ​​are selected as parents. For example, the top 50% of individuals with the highest fitness values ​​are selected as parents. In the crossover operation, parent individuals are paired up, and some of their genes (i.e., parameter adjustment information) are exchanged to generate new offspring individuals. For example, for two parent individuals, their oxygen flow rate parameter adjustment range information is exchanged. In the mutation operation, some genes in offspring individuals are randomly changed to increase population diversity. For example, the adjustment direction of the furnace temperature setpoint in a certain offspring individual is randomly changed. After calculating the fitness values ​​for each generation, the individuals with the highest fitness values ​​are retained, and it is checked whether these individuals meet time constraints, such as whether the parameter adjustment execution time is within the specified control period.

[0173] Step S456: After a preset number of generations of evolution, select a compromise solution from the Pareto optimal solution set from the final population, determine the adjustment direction, magnitude and corresponding execution time of each parameter, and generate a parameter adjustment scheme.

[0174] The preset number of generations represents the number of iterations in the multi-objective evolutionary algorithm. After this preset number of generations, the individuals in the population gradually converge to the Pareto optimal solution set. The Pareto optimal solution set is a set of solutions for which all objectives cannot be improved simultaneously by further adjusting the parameters. A compromise solution is selected from the Pareto optimal solution set, comprehensively considering both the primary and secondary optimization objectives, determining the adjustment direction, magnitude, and corresponding execution time for each parameter, and generating the final parameter adjustment scheme.

[0175] After a predetermined number of generations of evolution, a final population is obtained. From this final population, a Pareto optimal solution set is identified, for example, by comparing the fitness values ​​of individuals with the objective function values ​​to select individuals that meet the Pareto optimality criteria. Then, a compromise solution is selected from the Pareto optimal solution set. This compromise solution needs to balance minimizing the rate of anomalous development with minimizing the total magnitude of parameter adjustments. For example, a solution is chosen that can significantly reduce the rate of anomalous development while minimizing the magnitude of parameter adjustments. Based on the compromise solution, the adjustment direction, magnitude, and corresponding execution time of each parameter are determined, and this information is compiled into a parameter adjustment scheme.

[0176] Step S460: Convert the parameter adjustment scheme into a production parameter adjustment instruction that conforms to the protocol format of the steel production control system. The production parameter adjustment instruction includes the target node identifier, parameter code, adjustment value and execution timestamp.

[0177] Steel production control systems have their own defined protocol format. Converting parameter adjustment schemes into production parameter adjustment instructions that conform to this protocol format is essential to ensure that the instructions can be correctly received and executed by the system. Production parameter adjustment instructions contain key information such as target node identifiers, parameter codes, adjustment values, and execution timestamps. This information clarifies the target of the instruction, the parameters to be adjusted, the adjusted values, and the execution time.

[0178] When converting parameter adjustment schemes into production parameter adjustment instructions, the target node identifier is first determined, for example, the identifier for the converter process node is 001. For each adjustable parameter, its corresponding parameter code is found, such as the parameter code for oxygen flow rate being 0001. The adjustment value is determined according to the parameter adjustment scheme, such as increasing the oxygen flow rate by 20 cubic meters per hour. Simultaneously, the execution timestamp is determined according to the execution period in the parameter adjustment scheme, such as the execution timestamp for oxygen flow rate adjustment being 10:30 AM. This information is then organized according to the protocol format of the steel production control system to generate production parameter adjustment instructions, for example, the instruction format is [Target node identifier: 001, Parameter code: 0001, Adjustment value: +20 cubic meters per hour, Execution timestamp: 10:30 AM].

[0179] Step S470: The production parameter adjustment command is pushed to the parameter execution unit of the steel production control system according to the execution sequence via the industrial bus, so that the parameter execution unit performs the parameter adjustment operation at the corresponding time according to the execution timestamp in the production parameter adjustment command.

[0180] An industrial bus is a communication line used for data transmission in industrial control systems. Through the industrial bus, production parameter adjustment commands are pushed to the parameter execution unit of the steel production control system according to the execution sequence. The parameter execution unit is responsible for receiving the commands and performing the parameter adjustment operation at the corresponding time according to the execution timestamp in the command, thereby achieving real-time control of the production process.

[0181] In one implementation, step S470 may specifically include the following steps S471 to S476: Step S471: Parse the production parameter adjustment instructions to extract the execution timestamps, and sort them in ascending order by execution timestamps to generate an instruction execution sequence.

[0182] Parsing production parameter adjustment instructions involves extracting and processing information from the instructions. By extracting the execution timestamps and sorting them in ascending order to generate an instruction execution sequence, it can be ensured that the instructions are executed in the correct chronological order. This avoids confusion in parameter adjustments and guarantees the orderly progress of the production process.

[0183] Step S472: Monitor the current load rate of the industrial bus. If the current load rate exceeds the preset load threshold, adjust the push priority according to the urgency of the instruction execution timestamp. Instructions with earlier timestamps have higher priority.

[0184] The current load rate of the industrial bus reflects the bus's workload at any given moment. A preset load threshold is a pre-defined standard value; when the current load rate exceeds this threshold, it indicates that the bus load is too high and may not be able to process all instructions in a timely manner. Adjusting the push priority based on the urgency of the instruction execution timestamp ensures that instructions with earlier timestamps have higher priority. This guarantees that important instructions are pushed and executed first, reducing instruction execution delays caused by excessive bus load.

[0185] Step S473: Encapsulate the instructions in the instruction execution sequence into data frames conforming to the bus protocol in order of priority. The data frame contains the instruction identifier, the target node address, the parameter data, and the checksum.

[0186] Encapsulating instructions sequentially into data frames conforming to the bus protocol, according to priority, ensures correct transmission of instructions on the industrial bus. Each data frame contains an instruction identifier, target node address, parameter data, and a checksum. The instruction identifier uniquely identifies each instruction, the target node address specifies the target node to which the instruction will be executed, the parameter data contains specific parameter adjustment information, and the checksum verifies the accuracy of the data transmission.

[0187] When encapsulating a data frame, each instruction is processed sequentially according to its execution priority. For example, a data frame is generated for the highest priority instruction. The instruction identifier can be a unique number, such as 0001. The target node address is determined based on the target node identifier in the instruction, such as 001 for the converter process node. The parameter data includes the parameter code and adjustment value, such as 0001 for the parameter code and +20 cubic meters per hour for the adjustment value. The checksum is calculated based on the instruction identifier, target node address, and parameter data, and is used to verify the integrity of the data. This information is encapsulated according to the bus protocol format to generate a data frame [Instruction Identifier: 0001, Target Node Address: 001, Parameter Data: 0001 +20 cubic meters per hour, Checksum: XXXX].

[0188] Step S474: Push the data frame to the parameter execution unit of the steel production control system via the industrial bus, so that the execution unit can verify the check code after receiving the data frame. If the verification is successful, the instruction content is parsed and cached in the instruction queue.

[0189] The encapsulated data frame is pushed to the parameter execution unit of the steel production control system via the industrial bus. Upon receiving the data frame, the parameter execution unit first verifies the checksum. Verification of the checksum ensures that no errors occurred during data transmission. If the verification passes, it indicates that the data frame content is complete and accurate. At this point, the instruction content is parsed and cached in the instruction queue, awaiting execution at an appropriate time.

[0190] When a data frame is pushed via the industrial bus, the data frame is transmitted to the parameter execution unit on the bus. After receiving the data frame, the parameter execution unit calculates a checksum according to a pre-set verification algorithm and compares it with the checksum in the data frame. For example, a CRC checksum algorithm is used to calculate the checksum. If the calculated checksum matches the checksum in the data frame, the verification is successful. At this point, the instruction content is parsed, and information such as the target node address, parameter encoding, and adjustment value are extracted and cached in the instruction queue.

[0191] Step S475: The parameter execution unit compares the current system time with the execution timestamp of the instruction in the instruction queue. When the system time reaches the execution timestamp, the corresponding instruction is retrieved from the instruction queue to perform parameter adjustment operations.

[0192] The parameter execution unit compares the current system time with the execution timestamps of instructions in the instruction queue in real time. When the system time reaches the execution timestamp, it indicates that the instruction's execution time has arrived. The corresponding instruction is retrieved from the instruction queue, and the production parameters are adjusted according to the parameter adjustment information in the instruction to achieve real-time control of the production process.

[0193] The parameter execution unit obtains the current system time through the system clock. For example, if the current system time is 10:30 AM, and there is an instruction in the instruction queue with an execution timestamp of 10:30 AM, the parameter execution unit retrieves the instruction from the instruction queue when the system time reaches that timestamp. Based on the target node address and parameter code in the instruction, it locates the corresponding production equipment and parameters. For example, it locates the converter process node based on the target node address 001 and the oxygen flow parameter based on the parameter code 0001. Then, it adjusts the parameters according to the adjustment value in the instruction, increasing the converter's oxygen flow rate by 20 cubic meters per hour.

[0194] Step S476: After execution, an instruction execution result message is generated and returned to the monitoring system via the industrial bus, completing the push and execution process of the production parameter adjustment instruction.

[0195] The instruction execution result message contains information about the execution status of the parameter adjustment operation, such as success or failure. This result message is returned to the monitoring system via the industrial bus, enabling the system to promptly understand the instruction execution status and facilitating further monitoring and management of the production process.

[0196] After the parameter execution unit completes the parameter adjustment operation, it generates an instruction execution result message. For example, if the oxygen flow adjustment operation is successful, the message "[Instruction Identifier: 0001, Execution Result: Success]" is generated. This message is encapsulated into a data frame according to the bus protocol and returned to the monitoring system via the industrial bus. Upon receiving the execution result message, the monitoring system parses it to understand the execution status of the instruction. If execution is successful, the monitoring system can continue to monitor changes in production status; if execution fails, the monitoring system can take appropriate measures, such as resending the instruction or troubleshooting, thereby completing the push and execution process of the production parameter adjustment instruction.

[0197] It is understood that the various algorithms involved in the above descriptions of the embodiments of the present invention can all be obtained from relevant content in the prior art. To save space, they will not be elaborated on in the embodiments of the present invention. In addition, those skilled in the art can supplement the details based on common knowledge in the art when implementing the solutions of the present invention. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set thresholds based on historical data, experience or business scenario requirements, train the model based on a general model training method, set the number of layers in the model structure based on actual needs, select activation functions, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes here.

[0198] Please see Figure 2 , Figure 2This is a schematic diagram of a monitoring system provided in an embodiment of the present invention. The monitoring system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 can be connected via a bus or other means. The processor 101 (or Central Processing Unit, CPU) is the computing and control core of the monitoring system, capable of parsing various instructions and processing various data within the monitoring system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.), and can be used to send and receive data under the control of the processor 101; the communication interface 102 can also be used for data transmission and interaction within the monitoring system. The memory 103 is a memory device in the monitoring system used to store programs and data. It is understood that the memory 103 here can include the built-in memory of the monitoring system, or it can include extended memory supported by the monitoring system. The memory 103 provides storage space, which stores the operating system of the monitoring system; this invention does not limit this.

[0199] In one embodiment, the processor 101 executes the IoT-based steel production monitoring method provided above in the embodiments of the present invention by running a computer program in the memory 103.

Claims

1. A method for monitoring steel production based on the Internet of Things, characterized in that, The method includes: By deploying a multi-type sensor collaborative sensing network at the target process nodes in the steel production process, a multi-physics field collaborative sensing data set including temperature field distribution data, stress field distribution data, magnetic field intensity change data, and acoustic emission signal data is collected. Based on the multi-physics field collaborative sensing data set, a physical field coupling mapping model is constructed. By tracking the evolutionary correlation of cross-physical field features, a nonlinear mapping relationship is established between the evolution trajectory of micro-organizational structure and the trend of macro-production state changes. Using the physical field coupling mapping model, continuous feature correlation analysis is performed on the collected multi-physical field collaborative sensing data set based on the nonlinear mapping relationship to generate production status assessment results. Based on the production status assessment results, a production parameter adjustment instruction with timing control logic is generated, and the production parameter adjustment instruction is pushed to the steel production control system to execute the control operation.

2. The method according to claim 1, characterized in that, The construction of a physical field coupling mapping model based on the multi-physics field collaborative sensing data set establishes a nonlinear mapping relationship between the evolution trajectory of micro-organizational structure and the trend of macro-production state changes by tracking the evolutionary correlation of cross-physical field features, including: Spatiotemporal registration processing is performed on the multi-physics field collaborative sensing data set. The sliding step size of the time window is adjusted according to the sampling frequency fluctuation of different sensors. The temperature field, stress field, magnetic field and acoustic emission signal data are calibrated to a unified spatiotemporal reference system to generate a spatiotemporal registration data stream. The spatiotemporal registration data stream is tracked across physical field features. By using a rolling time window, the diffusion trend of the temperature field gradient, the transmission and attenuation process of the stress field strain, the time-varying characteristics of the eddy current response of the magnetic field strength, and the spectral energy migration trajectory of the acoustic emission signal are obtained to generate a multi-physical field feature evolution sequence. The correlation strength is tracked based on the multi-physics field feature evolution sequence. The correlation degree across physical field features is calculated by the feature covariance change rate within a continuous time segment, and a time-varying curve of correlation strength updated with the production process is generated. The hidden layer connection topology of the neural network is adjusted according to the time-varying curve of the correlation strength, so that the connection weight between network nodes increases or decreases with the fluctuation of the time-varying curve of the correlation strength, and a coupled mapping network framework with topological structure is constructed. A training sample sequence is constructed by synchronously collecting video streams of micro-organizational structure evolution and macro-production status recording curves during historical production processes. The training sample sequence is then input into the coupling mapping network framework segment by segment according to the production time sequence. The network connection weights are corrected through the backpropagation algorithm to generate a preliminary physical field coupling mapping model. The preliminary physical field coupling mapping model is optimized by adjusting the network topology based on the latest acquired multi-physics field collaborative sensing data and the prediction deviation of the model output until the fluctuation amplitude of the prediction deviation is lower than the convergence threshold within a continuous preset time period, and the final physical field coupling mapping model is determined.

3. The method according to claim 2, characterized in that, The process of performing spatiotemporal registration processing on the multi-physics collaborative sensing data set involves adjusting the sliding step size of the time window based on the sampling frequency fluctuations of different sensors, calibrating the temperature field, stress field, magnetic field, and acoustic emission signal data to a unified spatiotemporal reference system, and generating a spatiotemporal registration data stream, including: Extract the original acquisition timestamps and spatial coordinate information of each physical field data in the multi-physics collaborative sensing data set, construct a spatiotemporal coordinate matrix, and mark the process node number corresponding to each data point; Monitor the current sampling frequency of each sensor, calculate the sampling frequency fluctuation value, and when the fluctuation value exceeds the preset fluctuation threshold, adjust the sliding step size of the time window according to the fluctuation ratio so that the window size is positively correlated with the current sampling frequency; For temperature field distribution data, spatiotemporal interpolation is performed on the original data according to the adjusted time window. Kriging interpolation is used to supplement missing coordinate point data in the spatial dimension, and linear interpolation is used to supplement missing time point data in the temporal dimension, generating spatiotemporal continuous temperature field data. Using the same windowing logic as the temperature field, strain propagation path completion is performed on the stress field distribution data, spatiotemporal compensation of eddy current effect is performed on the magnetic field intensity change data, and spatiotemporal alignment of spectrum energy is performed on the acoustic emission signal data to generate spatiotemporal continuous data of each physical field. The spatiotemporal continuous data of each physical field are aligned with the coordinates and timestamps of a unified spatiotemporal reference system to obtain a spatiotemporal registration data stream containing temperature field, stress field, magnetic field and acoustic emission signals. Each spatiotemporal coordinate point in the spatiotemporal registration data stream contains the monitoring values ​​of the four types of physical fields.

4. The method according to claim 2, characterized in that, The step of tracking correlation strength based on the multi-physics feature evolution sequence, calculating the correlation degree across physical field features through the rate of change of feature covariance within continuous time segments, and generating a time-varying curve of correlation strength updated with the production process includes: The multiphysics feature evolution sequence is divided into continuous non-overlapping time segments according to the production process, and the length of each time segment is determined by the adjusted time window sliding step size. Within each time segment, the covariance value of the temperature field gradient diffusion trend and the stress field strain transmission attenuation process is calculated. The covariance value characterizes the degree of synchronous change of the two types of features within the time segment, wherein the two types of features are the temperature field gradient diffusion trend and the stress field strain transmission attenuation process. The covariance values ​​of temperature field and magnetic field, temperature field and acoustic emission signal, stress field and magnetic field, stress field and acoustic emission signal, and magnetic field and acoustic emission signal were calculated using the same method, resulting in six sets of cross-physical field covariance sequences. Perform a difference operation on the covariance values ​​corresponding to adjacent time segments to calculate the covariance change rate, which is the difference between the covariance value of the next segment and the covariance value of the previous segment divided by the length of the time segment. A correlation calculation function is constructed based on the covariance value and the rate of change of covariance. The correlation is obtained by weighted summation of the covariance value and the rate of change. The weight coefficients are determined based on the contribution of the correlation to the early warning of production anomalies in historical data. The correlation of each time segment is arranged according to the production time sequence to generate a time-varying curve set of correlation strength. Each time-varying curve of the correlation strength in the time-varying curve set corresponds to a set of correlation strength changes over time across physical field characteristics.

5. The method according to claim 2, characterized in that, The process involves constructing a training sample sequence using video streams of microscopic organizational structure evolution and macroscopic production status recording curves collected synchronously during historical production processes. This training sample sequence is then input into the coupling mapping network framework segment by segment according to production time sequence. The network connection weights are corrected using a backpropagation algorithm to generate a preliminary physical field coupling mapping model, including: The microstructure evolution video stream and macro production status recording curve corresponding to the same production batch are retrieved from the historical production database. The microstructure evolution video stream contains a sequence of metallographic microscope images collected at preset time intervals, and the macro production status recording curve contains data on the changes in rolling force, cooling rate and plate shape parameters at the same time interval. Image features are extracted from the video stream of the microstructure evolution, the grain boundary movement trajectory is tracked by the edge detection algorithm, and the phase composition ratio change rate is calculated by the grayscale analysis algorithm to generate a microstructure evolution feature sequence. Trend extraction is performed on the macroscopic production state recording curve. High-frequency noise is eliminated by sliding window filtering. The trend of rolling force change, cooling rate change and plate shape parameter change is extracted by trend line fitting algorithm to generate macroscopic production state evolution feature sequence. The micro-organizational structure evolution feature sequence and the macro-production state evolution feature sequence are aligned by timestamp to obtain micro-macro feature pairs, which are arranged in production time sequence to form a training sample sequence. Each sample in the training sample sequence contains multi-physics field feature data of the previous time segment and the micro-macro feature pair corresponding to the current time point. The training sample sequence is divided into continuous training segments according to the production time sequence. Each segment contains a preset number of samples. The training segments are input into the coupled mapping network framework segment by segment. The micro-macro feature prediction values ​​output by the model are calculated through forward propagation. The predicted values ​​of the micro-macro features are compared with the actual micro-macro features to calculate the prediction error. The network connection weights are corrected along the time axis using the backpropagation algorithm. The weights are updated once for each input training segment until all training segments are processed, generating a preliminary physical field coupling mapping model.

6. The method according to claim 1, characterized in that, The production status assessment results include the anomaly initiation time, development rate, and impact range. The production status assessment results are generated by performing continuous feature correlation analysis on the collected multi-physics collaborative sensing data set based on the nonlinear mapping relationship using the physical field coupling mapping model, including: The collected multi-physics field collaborative sensing data set is input into the preprocessing layer of the physical field coupling mapping model, and spatiotemporal registration processing consistent with the training phase is performed to generate a spatiotemporal registration data stream. The latest time-varying curve of correlation strength is loaded by calling the physical field coupling mapping model, and cross-physical field feature coupling is performed on the spatiotemporal registration data stream. The correlation relationship of different physical field features is obtained by adjusting the coupling weight, and the correlation feature stream is generated. The associated feature stream is input into the nonlinear mapping layer of the physical field coupling mapping model, and the micro-organization structure evolution prediction trajectory and macro-production state prediction trend line are calculated through the neural network of the topological structure. The predicted trajectory of microstructural evolution is compared with a preset standard evolution trajectory in a rolling manner, and the trajectory deviation is calculated by a sliding time window, wherein the window size is adjusted with the prediction time. The macro production state prediction trend line is compared with the preset standard trend line in a rolling manner, and the trend deviation is calculated using the same sliding time window to generate a macro trend deviation sequence. The trajectory deviation and macro trend deviation sequences are combined to generate a comprehensive deviation index. When the comprehensive deviation index exceeds a preset warning threshold, the anomaly start time is determined by tracing back the change process of the comprehensive deviation index. The anomaly development rate is calculated based on the rate of change of the comprehensive deviation index after the anomaly initiation time. The range of process nodes that may be affected within a preset time period is predicted by the spatial distribution relationship between the anomaly development rate and the current production process nodes, and the scope of the anomaly impact is determined. The anomaly's onset time, development rate, and impact range are integrated with the micro-organizational structure prediction trajectory and the macro-production status prediction trend line to generate a production status assessment result.

7. The method according to claim 6, characterized in that, The process involves calling the physical field coupling mapping model to load the latest correlation strength time-varying curve, performing cross-physical field feature coupling on the spatiotemporal registration data stream, and obtaining the correlation relationship of different physical field features by adjusting the coupling weights to generate a correlation feature stream, including: The physical field coupling mapping model receives the latest time-varying curve of correlation intensity through the internal weight update module and analyzes the cross-physical field correlation intensity value corresponding to each time point in the time-varying curve of correlation intensity. The spatiotemporal registration data stream is divided into data segments in chronological order, which are matched with the time granularity of the time-varying curve of the correlation intensity. Each data segment contains multiphysics feature data within the current time window. For each data segment, the association strength value at the corresponding time point is extracted from the association strength time-varying curve to construct a coupling weight matrix, and the element values ​​in the coupling weight matrix are updated with the association strength value; The temperature field features, stress field features, magnetic field features, and acoustic emission signal feature vectors in the data segment are weighted and combined with the coupling weight matrix to generate a cross-physical field coupling feature vector. During the coupling process, the rows and columns of the weight matrix correspond to different physical field feature dimensions. The cross-physical field coupling feature vectors of continuous time segments are concatenated in chronological order to obtain a temporally correlated feature stream. The temporal resolution of the correlated feature stream is consistent with that of the spatiotemporal registration data stream, and the dimension of the feature vector at each time point matches the dimension of the coupling weight matrix.

8. The method according to claim 6, characterized in that, The step of comparing the predicted trajectory of microstructural evolution with a preset standard evolution trajectory in a rolling manner, and calculating the trajectory deviation through a sliding time window, includes: Retrieve the standard evolution trajectory of the microstructure corresponding to the current steel grade from the process standard database. The standard evolution trajectory includes grain growth trajectory, phase transformation trajectory and dislocation density change trajectory. The initial sliding time window size is the base window size. As the prediction duration increases, the window size is increased by a preset ratio so that the length of the predicted trajectory covered by the window is positively correlated with the prediction duration. The predicted trajectory of microstructure evolution is input into the trajectory comparison module, which calculates the trajectory coordinate difference at corresponding time points within a sliding time window and generates a coordinate difference sequence. The coordinate difference sequence is weighted and summed, with the weight coefficients increasing in reverse chronological order to generate the trajectory deviation within the window; The sliding time window moves sequentially over time, and the trajectory deviation within each window is repeatedly calculated to generate a trajectory deviation sequence. Each element in the trajectory deviation sequence corresponds to the trajectory deviation value of a time window.

9. The method according to claim 6, characterized in that, The abnormal development rate is calculated based on the rate of change of the comprehensive deviation index after the onset of the abnormality. The range of process nodes that may be affected within a preset time period is predicted by the relationship between the abnormal development rate and the spatial distribution of the current production process nodes, thus determining the scope of the abnormal impact, including: Extract the index data after the abnormal start time from the comprehensive deviation index sequence, calculate the index difference between adjacent time points, and generate a comprehensive deviation change rate sequence. The comprehensive deviation rate of change sequence is filtered by moving average to generate a smoothed rate of change sequence, and the average value of the smoothed rate of change sequence is determined as the abnormal development rate. Obtain the spatial distribution map of the current production process nodes, mark the coordinates of the initial process node corresponding to the start time of the anomaly, and calculate the spatial distance between the initial node and the adjacent process nodes and the material transfer time. The anomaly diffusion coefficient is calculated based on the anomaly development rate and the material transport time. The diffusion coefficient is the product of the anomaly development rate and the material transport time, and represents the probability that the anomaly will spread from the initial node to the adjacent node. The abnormal diffusion coefficient is compared with a preset diffusion threshold. When the diffusion coefficient exceeds the threshold, the corresponding adjacent process node is marked as a potentially affected node. This process is repeated until all potentially affected nodes are marked to obtain the abnormal influence range.

10. A monitoring system, characterized in that, include: A memory, wherein a computer program is stored; A processor for loading the computer program to implement the IoT-based steel production monitoring method as described in any one of claims 1-9.

Images