Ultra-short-term prediction method, system and equipment for steel rolling impact load and medium

Abstract

The invention relates to the field of power system load prediction, in particular to a steel rolling impact load ultra-short-term prediction method, system and device and a medium. According to the method, three key characteristics of amplitude abrupt change rate, impact duration and periodic repetition are extracted by analyzing millisecond-level sampling data of total load of steel rolling. And by utilizing an improved OPTICS clustering algorithm, in combination with mahalanobis distance and dynamic xi parameter adjustment, the features are mapped to typical working condition categories, and working condition changes are accurately captured. And further matching a preset process knowledge base, and calculating a working condition deviation factor to trigger the re-calibration of the prediction model. The method does not need to depend on internal production parameters of an enterprise, effectively overcomes the defects of a traditional method in the aspects of data acquisition, decomposition algorithm and environment coupling, remarkably improves the accuracy and reliability of prediction, and provides an efficient solution for load prediction of the steel industry.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system load forecasting, in particular to a steel rolling impact load ultra-short-term forecasting method, system, device and medium. BACKGROUND

[0002] In the field of power system load forecasting, especially in the ultra-short-term (next day) load forecasting under the steel industry scenario, traditional methods face many challenges. The steel rolling process has high complexity and dynamics, and its load characteristics are influenced by various factors, including production activities such as start-stop of rolling mills, roll changing, billet specification changing, and sudden situations such as environmental protection power limiting and equipment failure. However, the existing technology has significant deficiencies in data acquisition, decomposition algorithm, and environmental coupling.

[0003] Firstly, traditional methods rely on internal information such as production plans and equipment parameters provided by enterprises, but the amount of effective information that can be actually obtained is often less than 30%, and key parameters such as rolling mill roll changing period and billet specification changing information are often missing. This greatly limits the accuracy and reliability of the prediction. For example, in the existing technology, such as the fine prediction method for power system area load disclosed in patent number CN119209495A, although the load forecasting model is optimized through industry labels and user portraits, it still relies on a large amount of historical data and user information, and mainly predicts the overall load of the power system area, without going deep into the ultra-short-term load forecasting of specific industrial scenarios such as steel rolling.

[0004] Secondly, existing wavelet decomposition, empirical mode decomposition (EMD) and other algorithms have the problem of modal aliasing when dealing with millisecond-level impact (peak change rate > 15 MW / s) and random intermittent characteristics (day fluctuation frequency up to 200-500 times) of rolling load, and cannot accurately extract load characteristics. These technical defects make it difficult for traditional methods to meet the high requirements of the steel industry for ultra-short-term load forecasting accuracy and real-time performance.

[0005] In addition, the existing technology fails to fully integrate dynamic factors such as environmental protection power limiting and equipment failure, resulting in a sharp increase in prediction deviation rate of 3-5 times under special working conditions. For example, in the event of sudden situations such as environmental protection power limiting or equipment failure, traditional methods often fail to adjust the prediction model in time, resulting in a significant increase in the deviation of the prediction results.

[0006] In summary, the existing technology has many problems in data acquisition, decomposition algorithm limitations and environmental coupling deficiencies, especially under non-intrusive data acquisition conditions, how to effectively improve the prediction accuracy and cope with the impact of sudden factors has become a technical problem to be solved. SUMMARY

[0007] The application aims to provide a steel rolling impact load ultra-short-term prediction method, system, device and medium to effectively improve the prediction accuracy and cope with the influence of sudden factors under non-invasive data acquisition conditions.

[0008] To achieve the above-mentioned purpose, the following technical solutions are adopted.

[0009] The steel rolling impact load ultra-short-term prediction method comprises the following steps.

[0010] S1, based on the millisecond-level sampling data of the total rolling load, three-dimensional features of amplitude mutation rate, impact duration and periodicity are extracted;

[0011] S2, the improved OPTICS algorithm is used to map the three-dimensional features to typical working condition categories, the feature dimension difference is eliminated by Mahalanobis distance, and the working condition drift is captured by combining the dynamic xi parameter adjustment, wherein the xi parameter is a parameter in the improved OPTICS algorithm, representing the dynamic threshold of the density clustering accessibility criterion;

[0012] S3, based on the clustering results, the load change mode of the preset process knowledge base is matched, and the working condition deviation factor is calculated to trigger the prediction model recalibration.

[0013] Optionally, the three-dimensional feature extraction in step S1 comprises:

[0014] Based on the difference between the load peak value and the valley value in a 10-second time window divided by the average value, the load mutation intensity caused by the rolling mill start-stop event is quantified;

[0015] The time period ratio of the energy ratio of the short-time window and the long-time window exceeding the threshold is calculated to reflect the rolling rhythm continuity;

[0016] The production rhythm stability is evaluated by spectrum analysis, and the reciprocal of the spectrum peak position variance is used as the stability index, wherein the calculation results of the amplitude mutation rate, the impact duration and the periodicity are used as the input data set of step S2.

[0017] Optionally, the improved OPTICS algorithm in step S2 comprises:

[0018] Mahalanobis distance is used to replace Euclidean distance, and feature weight is dynamically updated based on covariance matrix to eliminate the dimension dominant effect of amplitude mutation rate;

[0019] The xi parameter is increased by a preset step every hour to expand the clustering radius and adapt to the gradual change of load characteristics caused by roll wear;

[0020] The detection threshold of the sudden failure class small probability event is set as a fixed proportion of the total sample size; the output clustering label includes the identification results of the rough rolling continuous working condition, the finishing rolling intermittent working condition and the fault shutdown working condition.

[0021] Optionally, the amplitude mutation rate calculation result is directly input into a neighborhood density calculation module of the OPTICS clustering algorithm;

[0022] The impact duration calculation result is used to optimize the reachable distance sorting of the OPTICS algorithm;

[0023] The period repetition degree calculation result participates in constructing the core object screening condition of the OPTICS algorithm; wherein the output of the clustering module provides the working condition classification label and the cluster center distance data for step S3.

[0024] Optionally, the working condition inversion in step S3 includes:

[0025] Integrating the standard operation process of the hot rolling production line, defining the load change mapping rules of 12 types of working conditions to construct a process knowledge base;

[0026] When the amplitude mutation rate suddenly rises and the period repetition degree is zero in the clustering result, it is marked as a roll changing event; when the impact duration suddenly drops and the time exceeding the threshold value lasts for more than 2 hours, it is marked as an environmental protection production limiting event;

[0027] Taking the ratio of the current cluster center distance to the historical cluster radius as a working condition deviation factor, when the factor exceeds a preset threshold value, the prediction model parameter recalibration is triggered.

[0028] Optionally, the prediction model recalibration includes:

[0029] When the working condition deviation factor is greater than 1.2, the parameter update instruction is activated;

[0030] Based on the real-time load data, the LSTM prediction network is trained in a rolling manner, and the fully connected layer weight matrix in the original model is replaced;

[0031] The updated prediction model is applied to the prediction of subsequent load sequences, and the next day's load curve is output.

[0032] A steel rolling impact load prediction system, comprising:

[0033] A feature extraction module for calculating the amplitude mutation rate, impact duration and period repetition degree, and outputting a three-dimensional feature vector;

[0034] A working condition clustering module connected to the feature extraction module, for executing an improved OPTICS algorithm, including a Mahalanobis distance calculation unit, an xi parameter dynamic adjustment unit and a minimum cluster size screening unit, and outputting a working condition classification label;

[0035] A correction execution module connected to the working condition clustering module, including a process knowledge base matching unit and a deviation factor calculation unit, and when the deviation factor exceeds a threshold value, an LSTM model weight update instruction is triggered;

[0036] A data pipeline adopts a message queue architecture to transmit load sampling data, feature vectors and working condition labels in real time.

[0037] Optionally, the feature extraction module comprises:

[0038] A mutation rate calculation unit is configured to calculate the mutation rate based on a peak-valley difference to mean value ratio of a 10-second window;

[0039] A duration statistics unit is configured to calculate the long-time window energy ratio threshold overrun duration proportion based on a short-time window / long-time window energy ratio;

[0040] The working condition clustering module comprises:

[0041] A covariance matrix updating unit is configured to refresh the feature weight parameters every hour;

[0042] A core object screening unit is configured to separate the fault event samples according to a minimum cluster size threshold;

[0043] The correction execution module comprises:

[0044] A rule matching engine is configured to call 12 working condition mapping rules of a process knowledge base;

[0045] A rolling training unit is configured to update the LSTM model weight based on real-time data increment.

[0046] An electronic device comprises a processor and a memory, and the processor is configured to execute a computer program stored in the memory to implement the steel rolling impact load ultra-short-term prediction method.

[0047] A computer readable storage medium stores at least one instruction, and the at least one instruction is executed by a processor to implement the steel rolling impact load ultra-short-term prediction method.

[0048] Compared with the prior art, the present application has the following beneficial effects,

[0049] This invention proposes an ultra-short-term prediction method for impact loads in steel rolling mills based on unsupervised feature learning technology. By constructing a three-dimensional feature space and employing an improved OPTICS clustering algorithm, it achieves the inversion from non-intrusive load data to production processes, effectively solving the problems of insufficient data acquisition, limitations of decomposition algorithms, and lack of environmental coupling in traditional methods. Under non-intrusive data acquisition conditions, it effectively improves prediction accuracy and addresses the impact of sudden factors. First, by extracting three core feature dimensions—amplitude mutation rate, impact duration, and cycle repeatability—this invention can accurately identify key events and changes in operating conditions during the rolling process without relying on internal production parameters, thus significantly improving the accuracy and reliability of predictions. Second, the improved OPTICS clustering algorithm further improves the accuracy and adaptability of operating condition identification by using Mahalanobis distance to eliminate differences in feature dimensions and combining dynamic xi parameter adjustment to capture operating condition drift. Furthermore, based on the matching of clustering results with a pre-set process knowledge base, this invention can calculate the operating condition deviation factor in real time and trigger the recalibration of the prediction model, thereby effectively addressing the impact of sudden factors on the prediction results. In specific embodiments, under different scenarios such as normal rolling, environmental power rationing, and equipment failure, the present invention reduces the mean absolute percentage error (MAPE) by 44.7%, 51.4%, and 52.7% respectively compared to traditional methods, fully demonstrating its superior technical effect. Furthermore, claims 2 to 6 provide detailed specifications for key technical aspects such as feature extraction, clustering algorithm optimization, and operating condition inversion, further enhancing the technical advantages of the present invention. For example, the specific calculation method for three-dimensional features ensures the accuracy and efficiency of feature extraction; the improved OPTICS algorithm is optimized, improving the stability and reliability of clustering results; and the adaptiveness and accuracy of the prediction model are further enhanced by constructing a process knowledge base and a dynamic correction mechanism. Finally, applying the method of the present invention to specific system architectures and electronic devices achieves a complete transformation from theory to practice, providing an efficient and reliable solution for ultra-short-term load forecasting in the steel industry. Attached Figure Description

[0050] Figure 1 This is a schematic flowchart illustrating the steps of an embodiment of the steel rolling impact load ultra-short-term prediction method according to the present invention.

[0051] Figure 2 This is a schematic diagram of a module in an embodiment of the steel rolling impact load ultra-short-term prediction system according to the present invention.

[0052] Figure 3 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0053] The present application will be described in detail below with reference to the drawings and in conjunction with embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0054] The following detailed description is exemplary and is intended to provide further details of the present application. Unless otherwise specified, all technical terms used in the present application have the same meanings as understood by those skilled in the art to which the present application belongs. The terms used in the present application are only for the purpose of describing the specific embodiments and are not intended to limit the exemplary embodiments according to the present application.

[0055] Embodiment 1

[0056] As Figure 1 shown, the present application provides a steel rolling impact load ultra-short-term prediction method, which realizes high-precision prediction of steel rolling impact load under non-invasive data acquisition conditions by using unsupervised feature learning technology, combining improved OPTICS clustering algorithm and dynamic correction mechanism. The specific implementation steps and technical details of the method will be described in detail below.

[0057] In the steel rolling production process, the rolling load has strong randomness and impact, and the millisecond-level load change poses a great challenge to the stable operation of the power system. The traditional load prediction method relies on the production plan and equipment parameters of the enterprise, and the data acquisition is insufficient, and it cannot effectively handle the high-frequency impact characteristics of the rolling load, so the prediction accuracy is difficult to meet the actual demand. In addition, the existing method will significantly increase the prediction deviation rate when facing sudden working conditions (such as environmental protection power limitation, equipment failure). In view of these problems, the present application proposes an ultra-short-term prediction method based on unsupervised feature learning, which realizes accurate prediction and dynamic correction of rolling load by constructing a three-dimensional feature space and an improved clustering algorithm.

[0058] The present application first extracts three key features from the millisecond-level sampling data of the total rolling load: amplitude mutation rate, impact duration and periodicity. These features can effectively reflect the dynamic changes in the rolling process, and do not need to rely on detailed production parameters within the enterprise, thereby solving the problem of insufficient data acquisition.

[0059] Amplitude mutation rate: This feature is used to quantify the mutation intensity of the load caused by rolling mill start-stop, roll change and other events. The specific calculation method is: in a 10-second time window, calculate the difference between the peak and valley values of the load, and divide it by the average load value in the window. For example, when changing the roll in the cold rolling line, the amplitude mutation rate may exceed 1.2, which indicates that the load has undergone significant mutation. In this way, the amplitude mutation rate can effectively capture the load changes caused by equipment operation in the rolling process.

[0060] Calculation method:

[0061] Physical meaning: Quantify the intensity of load mutation caused by events such as mill start-stop and roll change (e.g., R mutation > 1.2 when changing rolls in a cold rolling line).

[0062] Data source: Total rolling load sampling data within a 10-second D5000 (power grid dispatching control system) window (sampling rate 1000 Hz).

[0063] Impact duration: This feature is used to reflect the continuity of rolling rhythm. It is calculated by counting the proportion of time periods where the energy ratio between short-time window (e.g., 5 seconds) and long-time window (e.g., 50 seconds) exceeds a threshold. When the rolling process is continuous, the proportion of impact duration will be higher, such as more than 85%. This feature can effectively identify intermittent impacts in the rolling process, providing an important basis for subsequent working condition identification.

[0064] Calculation method:

[0065] Physical meaning: Count the proportion of time periods where STA / LTA threshold is exceeded, reflecting the continuity of rolling rhythm (Timpact>85% for continuous rolling).

[0066] Parameter setting: Short-time window m=5s, long-time window n=50s, dynamically adjusted for rough rolling / finishing rolling.

[0067] Cycle repetition degree: This feature is used to evaluate the stability of production rhythm. Through spectral analysis, the reciprocal of the variance of spectral peak position is calculated as a stability indicator. In normal rolling process, the cycle repetition degree is usually high, such as more than 0.8. This feature can effectively identify periodic changes in the rolling process, further enriching the dimensionality of the feature space.

[0068] Calculation method:

[0069] Physical meaning: Evaluate the stability of production rhythm (Speriod>0.8 for normal rolling).

[0070] Through the extraction of the above three features, the invention constructs a three-dimensional feature space that can comprehensively reflect the dynamic characteristics of rolling load. These features not only effectively capture key events in the rolling process, but also provide a solid foundation for subsequent working condition identification and prediction.

[0071] After extracting the three-dimensional features, the application uses an improved OPTICS clustering algorithm to map these features to typical working condition categories. Traditional clustering algorithms (such as K-means) have limitations when dealing with high-dimensional data, especially when the feature dimension differences are large, which can easily lead to biased clustering results. To this end, the application uses Mahalanobis distance instead of Euclidean distance, and dynamically updates the feature weight through the covariance matrix, effectively eliminating the influence of feature dimension differences on clustering results.

[0072] Mahalanobis distance takes into account the correlation between features, adjusting the feature weight through the inverse of the covariance matrix. In the application, the covariance matrix is dynamically updated every hour to adapt to changes in load characteristics during the rolling process. For example, during the rolling process, load characteristics gradually change. By dynamically updating the covariance matrix, this gradual change can be effectively captured, thereby improving the accuracy of clustering.

[0073] To adapt to the gradual change of load characteristics during the rolling process, the application introduces a dynamic xi parameter adjustment mechanism. The xi parameter increases by a preset step size, such as 0.01, every hour. This adjustment mechanism can effectively expand the clustering radius, thereby better adapting to the changes in load characteristics caused by roll wear. The xi parameter is the ξ parameter in the improved OPTICS algorithm: a dynamic threshold representing the density clustering reachability criterion, initially set to 0.05 and automatically incremented by 0.01 every hour. By expanding the clustering radius, it adapts to the gradual change in load characteristic distribution caused by roll wear (such as a 15% radius expansion in the later stages of wear). The matching relationship between the parameter increment and the equipment aging rate is achieved through dynamic updating of the covariance matrix. For example, in the later stages of roll wear, the clustering radius may expand by 15%. By dynamically adjusting the xi parameter, the accuracy and stability of the clustering results can be ensured.

[0074] In the steel rolling process, sudden failures and other low-probability events have a significant impact on load prediction. Therefore, the application sets the detection threshold for sudden failure and other low-probability events to a fixed proportion of the total sample size, such as 5%. In this way, sudden failures and other low-probability events can be effectively detected, providing a basis for subsequent working condition correction.

[0075] Through the improved OPTICS clustering algorithm, the application can effectively map three-dimensional features to typical working condition categories, including rough rolling continuous working conditions, finishing rolling intermittent working conditions, and fault shutdown working conditions. For example, when the amplitude mutation rate is greater than 0.8 and the impact duration is greater than 80%, the clustering result is identified as a rough rolling continuous working condition; when the periodicity repetition degree is greater than 0.9 and the impact duration is less than 50%, it is identified as a finishing rolling intermittent working condition; when the amplitude mutation rate is less than 0.1 and the impact duration is 0, it is identified as a fault shutdown working condition. These clustering results provide an important basis for subsequent working condition inversion and prediction model correction.

[0076] Improved OPTICS clustering algorithm

[0077] Objective: Map high-dimensional features to typical working condition categories (e.g. rough rolling continuous, equipment failure), and realize unsupervised working condition recognition.

[0078] Algorithm optimization points:

[0079] Distance metric: Use Mahalanobis distance instead of Euclidean distance to eliminate the over-sensitivity of feature dimension difference to amplitude mutation rate:

[0080]

[0081] Where ∑ is the covariance matrix, and the dynamic update period is 1 hour.

[0082] After feature extraction and clustering, the invention maps the clustering results to the actual production scene through working condition inversion and dynamic correction mechanism, and provides working condition correction factors for the prediction model. This process is the key link for the invention to realize high-precision prediction.

[0083] Parameter dynamic adjustment:

[0084] xi parameter: increase by 0.01 per hour to capture the load characteristic drift caused by roll wear (e.g. 15% expansion of cluster radius in the later wear stage).

[0085] Minimum cluster size: min_cluster_size = 0.05, suitable for small probability event (e.g. sudden failure) detection.

[0086] Output results:

[0087] Cluster label Typical operating condition Characteristic range C1 Continuous roughing R mutation > 0.8, T impact > 80% C2 Intermittent finishing S period > 0.9, T impact < 50% C3 Failure stop R mutation < 0.1, T impact = 0

[0088] The invention integrates the steel process knowledge base, defines 12 types of working condition load change patterns, and constructs an expert rule base. These rules cover load change characteristics under various working conditions from normal rolling to sudden failure. For example, roll changing events are usually accompanied by a sharp rise in amplitude mutation rate (more than 1.5) and zero periodicity; environmental protection production limit shows a sudden drop in impact duration (more than 30%) and a time duration exceeding the threshold for more than 2 hours. Through these rules, the invention can accurately map the clustering results to the actual production scene.

[0089] In the working condition inversion process, the invention calculates the working condition deviation factor to evaluate the deviation degree of the current working condition from the historical working condition. The specific calculation method is: take the ratio of the current cluster center distance to the historical cluster radius as the working condition deviation factor. When this factor exceeds the preset threshold (e.g. 1.2), trigger the parameter recalibration of the prediction model. This mechanism can effectively deal with the impact of sudden working conditions on the prediction results, ensuring the real-time and accuracy of the prediction model.

[0090] When the operating condition deviation factor exceeds the threshold value, the application activates the parameter update instruction, and trains the long short-term memory (LSTM) prediction model based on real-time load data. By replacing the fully connected layer weight matrix in the original model, the updated prediction model can better adapt to the current operating condition changes. Finally, the updated prediction model is applied to the prediction of subsequent load sequences, and the next day's load curve is output.

[0091] Operating condition inversion and correction mechanism

[0092] Objective: Map the clustering results to the actual production scene and provide operating condition correction factors for the prediction model.

[0093] Key technologies:

[0094] Expert rule base:

[0095] Integrate steel process knowledge base (such as hot rolling production line standard operation procedure), define 12 types of load change patterns of operating conditions (for example: roll change event: accompanied by R mutation sudden rise (>1.5) and S period zero.

[0096] Environmental protection production limit: represented by T impact sudden drop (>30%) and sustained for more than 2 hours.

[0097] Dynamic correction feedback:

[0098] Correction factor calculation:

[0099] When the alpha operating condition is greater than 1.2, the prediction model parameter recalibration is triggered.

[0100] Through the above operating condition inversion and dynamic correction mechanism, the application can realize high-precision prediction of steel rolling impact load under non-invasive data acquisition conditions, and effectively cope with the impact of sudden operating conditions. In actual application, the prediction mean absolute percentage error (MAPE) of the application under normal rolling, environmental protection power limitation and equipment failure in different scenes is reduced by 44.7%, 51.4% and 52.7% respectively, fully proving its superior technical effect.

[0101] Technical advantages and embodiments

[0102] Information compensation capability: only relying on total load data can identify 83% of rolling process changes (such as rough rolling→finishing rolling switching), reducing the dependence on production plan data.

[0103] Real-time: clustering calculation time <200ms, meeting the 15-minute level prediction period requirement.

[0104] Noise resistance: Mahalanobis distance measurement makes the algorithm maintain more than 90% operating condition recognition accuracy when the load fluctuation is ±20%.

[0105] Example data (January 2025, Wu'an rolling load measurement):

[0106]

[0107] Example 2

[0108] As Figure 2 shown, the present application also provides a steel rolling impact load prediction system, which is based on the above prediction method and realizes the full-process automatic processing from data acquisition to prediction output through modular design. The specific composition of the system and its technical details will be described in detail below.

[0109] The prediction system of the present application includes the following key modules: feature extraction module, working condition clustering module, correction execution module and data pipeline. These modules realize real-time prediction and dynamic correction of steel rolling impact load through efficient data transmission and processing mechanism.

[0110] Feature extraction module: This module is responsible for extracting three key features of amplitude mutation rate, impact duration and periodicity from the millisecond-level sampling data of rolling total load. The specific implementation is as follows:

[0111] Mutation rate calculation unit: Based on the difference between the peak and valley values of the load within a 10-second time window divided by the average value, the amplitude mutation rate is calculated. This unit can effectively capture the load mutation caused by mill start-stop events.

[0112] Duration statistics unit: By calculating the proportion of time periods where the energy ratio of short-time window to long-time window exceeds the threshold, the impact duration is calculated. This unit can reflect the continuity of rolling rhythm.

[0113] Periodicity calculation unit: Through spectral analysis, the reciprocal of the variance of the spectral peak position is calculated as the periodicity. This unit can evaluate the stability of the production rhythm.

[0114] These calculation units work together to output a three-dimensional feature vector, providing basic data for subsequent working condition clustering.

[0115] Working condition clustering module: This module connects the feature extraction module and is responsible for executing the improved OPTICS clustering algorithm. The specific implementation is as follows:

[0116] Mahalanobis distance calculation unit: Mahalanobis distance is used instead of Euclidean distance, and feature weights are dynamically updated based on covariance matrix. This unit can effectively eliminate the influence of feature dimension difference on clustering results.

[0117] xi parameter dynamic adjustment unit: increase xi parameter preset step length, for example, 0.01, per hour. This unit can adapt to the gradual change of load characteristics caused by roll wear.

[0118] Minimum cluster size screening unit: set the detection threshold of small probability events such as sudden failure to a fixed proportion of the total sample size, for example, 5%. This unit can effectively detect small probability events such as sudden failure.

[0119] Through the cooperative work of these units, the working condition clustering module outputs the working condition classification label, including the identification results of rough rolling continuous working condition, finishing rolling intermittent working condition and fault shutdown working condition.

[0120] Correction execution module: this module is connected with the working condition clustering module, responsible for mapping the clustering results to the actual production scene, and providing the working condition correction factor for the prediction model. The specific implementation is as follows:

[0121] Rule matching engine: integrate the steel process knowledge base, define 12 types of load change patterns. When the amplitude mutation rate in the clustering result suddenly rises and the cycle repetition degree is zero, it is marked as roll changing event; when the impact duration suddenly drops and the time exceeding the threshold lasts more than 2 hours, it is marked as environmental protection production limiting event. This unit can accurately map the clustering results to the actual production scene.

[0122] Deviation factor calculation unit: take the ratio of the current cluster center distance to the historical cluster radius as the working condition deviation factor. When the factor exceeds the preset threshold (for example, 1.2), trigger the parameter re-calibration of the prediction model. This unit can effectively deal with the influence of sudden working condition on the prediction result.

[0123] Rolling training unit: based on real-time load data, rolling training LSTM prediction network, replacing the fully connected layer weight matrix in the original model. This unit can dynamically update the prediction model, ensuring its real-time performance and accuracy.

[0124] Data pipeline: use message queue architecture to transmit load sampling data, feature vector and working condition label in real time. This module ensures the efficient transmission and processing of data, and provides guarantee for the real-time performance of the system.

[0125] System running process

[0126] In actual operation process, the prediction system of the application works according to the following process:

[0127] Data acquisition: the system acquires millisecond level sampling data of total rolling load in real time through the data pipeline.

[0128] Feature extraction: the feature extraction module calculates the amplitude mutation rate, impact duration and cycle repetition degree from the collected data, and outputs a three-dimensional feature vector.

[0129] Working condition clustering: The working condition clustering module receives the feature vector and outputs the working condition classification label through the improved OPTICS clustering algorithm.

[0130] Working condition inversion and correction: The correction execution module matches the clustering results with the process knowledge base, calculates the working condition deviation factor, and triggers the dynamic correction of the prediction model when necessary.

[0131] Prediction output: The updated prediction model outputs the next day's load curve based on real-time data, providing a basis for power system scheduling.

[0132] The prediction system of the present application can achieve high-precision prediction of steel rolling impact load under non-invasive data acquisition conditions and effectively cope with the impact of sudden working conditions. The system not only improves the accuracy and real-time performance of the prediction, but also provides an intelligent power scheduling solution for enterprises.

[0133] The present application solves the problems of insufficient data acquisition, limitations of decomposition algorithms and missing environmental coupling in traditional steel rolling impact load prediction methods by constructing a three-dimensional feature space, an improved OPTICS clustering algorithm and a dynamic correction mechanism. Through detailed feature extraction, clustering analysis and working condition inversion, the present application not only improves the prediction accuracy, but also enhances the adaptability of the prediction model through the dynamic correction mechanism. In addition, the system architecture design of the present application realizes the full-process automation from data acquisition to prediction output, providing an efficient and reliable power scheduling solution for enterprises.

[0134] Example 3

[0135] An electronic device comprising a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement a steel rolling impact load prediction method.

[0136] The application further provides an electronic device 100 for implementing the steel rolling impact load prediction method of the above-mentioned embodiments; the electronic device 100 comprises a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104. The memory 101 can be used to store the computer program 103, the processor 102 can implement the steps of the steel rolling impact load prediction method of the embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 can mainly comprise a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program (such as a sound playing function, an image playing function, etc.) required by a function, etc.; the data storage area can store data (such as audio data) created according to the use of the electronic device 100, etc. In addition, the memory 101 can comprise a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. The at least one processor 102 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 102 can be a microprocessor or can also be any conventional processor, etc. The processor 102 is the control center of the electronic device 100, and connects all parts of the electronic device 100 by using various interfaces and lines. The memory 101 in the electronic device 100 stores a plurality of instructions to implement the steel rolling impact load prediction method.

[0137] Embodiment 4

[0138] A computer readable storage medium, the computer readable storage medium stores at least one instruction, the at least one instruction is executed by a processor to implement a steel rolling impact load prediction method.

[0139] The modules / units integrated in the electronic device 100, if implemented in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, and read-only memory (ROM). Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer readable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program code. The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocks Figure 1 one or more flows and / or blocksFigure 1 the steps of the functions specified in the one or more blocks.

[0140] It is apparent that the application can be carried out in series of ways, and utilizing equivalent raw materials and / or method steps without departing from the spirit and essential characteristics of the application. Accordingly, all modifications and equivalents thereof are included in the scope of the application.

Claims

1. A method for ultra-short term prediction of impact loads in steel rolling, characterized in that, Comprising, S1, based on the rolling total load millisecond level sampling data, extracting amplitude mutation rate, impact duration and periodicity three-dimensional features; S2, using improved OPTICS algorithm to map three-dimensional features to typical working condition categories, eliminating feature dimension difference by Mahalanobis distance, combining with dynamic xi parameter adjustment to capture working condition drift, wherein xi parameter is the parameter in improved OPTICS algorithm, representing dynamic threshold of density clustering accessibility criterion; S3, based on the clustering results, matching the load change mode of the preset process knowledge base, calculating the working condition deviation factor to trigger the prediction model re-calibration.

2. The steel rolling impact load ultra-short-term prediction method according to claim 1, characterized by, The three-dimensional feature extraction in step S1 includes: Based on the difference between the load peak value and the valley value in the 10-second time window divided by the average value, the load mutation intensity caused by the rolling mill start-stop event is quantified; The time period ratio of the energy ratio of the short-time window to the long-time window exceeding the threshold is calculated to reflect the continuity of the rolling rhythm; The production rhythm stability is evaluated by spectrum analysis, and the reciprocal of the spectrum peak position variance is taken as the stability index, wherein the calculation results of the amplitude mutation rate, the impact duration and the periodicity are taken as the input data set of step S2.

3. The steel rolling impact load ultra-short-term prediction method according to claim 1, characterized by, The improved OPTICS algorithm in step S2 includes: Mahalanobis distance is used to replace Euclidean distance, and feature weight is dynamically updated based on covariance matrix to eliminate the dimension dominant effect of amplitude mutation rate; The xi parameter is increased by a preset step every hour to expand the clustering radius and adapt to the gradual change of load characteristics caused by roll wear; The detection threshold of the small probability event of sudden failure class is set as a fixed proportion of the total sample size; the output clustering label includes the recognition results of rough rolling continuous working condition, finishing rolling intermittent working condition and fault shutdown working condition.

4. The steel rolling impact load ultra-short-term prediction method according to claim 2 or 3, characterized in that: The calculation result of the amplitude mutation rate is directly input into the neighborhood density calculation module of the OPTICS clustering algorithm; The calculation result of the impact duration is used to optimize the reachable distance sorting of the OPTICS algorithm; The calculation result of the periodicity is used to construct the core object screening condition of the OPTICS algorithm; wherein the output of the clustering module provides the working condition classification label and the cluster center distance data for step S3.

5. The steel rolling impact load ultra-short-term prediction method according to claim 1, characterized by, The working condition inversion in step S3 includes: Integrating the standard operation process of hot rolling production line, defining the load change mapping rules of 12 types of working conditions to build the process knowledge base; When the amplitude mutation rate in the clustering result suddenly rises and the periodicity is zero, it is marked as a roll change event; when the impact duration suddenly drops and the time exceeding the threshold is more than 2 hours, it is marked as an environmental protection production limiting event; The ratio of the current cluster center distance to the historical cluster radius is taken as the working condition deviation factor, and when the factor exceeds the preset threshold, the prediction model parameter re-calibration is triggered.

6. The steel rolling impact load ultra-short-term prediction method according to claim 5, characterized by, The prediction model re-calibration includes: When the working condition deviation factor is greater than 1.2, the parameter update instruction is activated; Based on the real-time load data, the LSTM prediction network is trained rolling, and the full connection layer weight matrix in the original model is replaced; The updated prediction model is applied to the prediction of subsequent load sequence, and the next day load curve is output.

7. A steel rolling impact load prediction system based on the steel rolling impact load prediction method according to any one of claims 1-6, characterized in that, Comprising: A feature extraction module for calculating amplitude mutation rate, impact duration and periodicity, and outputting a three-dimensional feature vector; The working condition clustering module is connected with the feature extraction module and is configured to execute an improved OPTICS algorithm, including a Mahalanobis distance calculation unit, an xi parameter dynamic adjustment unit, and a minimum cluster size screening unit, and outputs a working condition classification label; The correction execution module is connected with the working condition clustering module and includes a process knowledge base matching unit and a deviation factor calculation unit, and triggers an LSTM model weight update instruction when the deviation factor exceeds a threshold value; The data pipeline adopts a message queue architecture to transmit load sampling data, feature vectors, and working condition labels in real time.

8. The system of claim 7, wherein: The feature extraction module includes: a mutation rate calculation unit configured to calculate a mutation rate based on a peak-valley difference to mean value ratio of a 10-second window; a duration statistics unit configured to calculate a long-time window energy ratio threshold value exceeding duration proportion based on a short-time window / long-time window energy ratio; The working condition clustering module includes: a covariance matrix updating unit configured to refresh feature weight parameters every hour; a core object screening unit configured to separate fault event samples according to a minimum cluster size threshold value; The correction execution module includes: a rule matching engine configured to call 12 types of working condition mapping rules of a process knowledge base; a rolling training unit configured to update LSTM model weights based on real-time data increments.

9. An electronic device, comprising: The processor is configured to execute a computer program stored in the memory to implement the steel rolling impact load ultra-short-term prediction method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction, and the at least one instruction is executed by the processor to implement the steel rolling impact load ultra-short-term prediction method according to any one of claims 1 to 6.

Images