Continuous casting production process time series data segmentation method and device, computer equipment

Abstract

The application relates to the technical field of data processing, and discloses a continuous casting production process time sequence data segmentation method and device and computer equipment. The method comprises the following steps: collecting time sequence data in a continuous casting production process, covering a pulling speed, a stopper position, a crystallizer liquid level and the like. Any kind of time sequence data collected in a preset data segmentation period is denoised to obtain denoised time sequence data. The denoised time sequence data is grouped based on a preset unit time span, a dynamic time warping distance of a reference denoised time sequence data group corresponding to a plurality of preset key events of each group is calculated, and a reference group is determined. When the reference group exists, the number of preset unit time spans covering the preset unit time span is determined according to the comparison between the preset duration of the key event corresponding to the reference group and the preset unit time span, and then an actual data segment is determined and marked. The denoised time sequence data is flexibly and reasonably segmented according to different situations, and more suitable data units can be provided for subsequent data analysis and processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a continuous casting production process time series data segmentation method and device and computer equipment. BACKGROUND

[0002] The continuous casting process is a metallurgical process of forming a steel billet by continuously cooling and solidifying liquid steel. As a core production process in the modern steel industry, it belongs to the typical high-temperature process manufacturing field. As a key link between the steelmaking and rolling processes, the stability of the production process not only directly affects the geometric size precision and internal metallurgical quality of the continuous casting billet, but also has a decisive influence on the yield, microstructure and surface defects of the final steel product through genetic effects. Therefore, establishing a real-time monitoring system and an abnormal early warning mechanism for the continuous casting production process has important engineering practical value for improving the quality of steel products.

[0003] At present, the steel industry in China is in a key stage of intelligent manufacturing transformation and upgrading. With the improvement of digital infrastructure such as multi-source heterogeneous sensor networks, industrial internet of things (IIoT, Industrial Internet of Things) platforms and manufacturing execution systems (MES, Manufacturing Execution System), the process data available in the production process has been significantly improved in terms of sampling frequency, dimension scale and time continuity, and its data category covers process operation parameters, equipment operating status, product quality indicators and other modalities. However, although the existing data acquisition system has achieved the ability to acquire massive industrial data, there is still a significant technical gap in the depth of data value mining and intelligent application.

[0004] At present, the production process optimization of steel enterprises mainly relies on the structured data sets provided by the process control system (PCS, Process Control System) and the manufacturing execution system, and uses traditional statistical process control (SPC, Statistical Process Control) or shallow machine learning algorithms for analysis. Through practical verification, the existing technical solutions have the following inherent defects:

[0005] Low data value density: for high-frequency sampled multi-dimensional time series data streams (including but not limited to key process variables such as casting speed, stopper opening degree, mold vibration parameters), the existing system only realizes the basic trend display and static threshold alarm function, and fails to effectively establish a deep mapping relationship between process parameters and billet quality indicators;

[0006] Model robustness is insufficient: when traditional machine learning algorithms such as support vector machine (SVM), random forest (RF) and the like directly process original time series signals, there are problems such as low feature extraction efficiency, high noise sensitivity and exponential growth of computational complexity, which are difficult to meet the real-time demand of millisecond-level response in industrial field;

[0007] Process interpretability is missing: existing data analysis methods lack deep integration with continuous casting metallurgical mechanisms (including solidification heat transfer dynamics, molten steel flow control theory, etc.), making it difficult to convert analysis results into executable process adjustment strategies.

[0008] It is particularly pointed out that the key time series parameters such as the withdrawal speed, the stopper position, and the mold liquid level constitute a direct representation of the stability of the continuous casting production process. Specifically, the non-steady state fluctuation of the withdrawal speed has a strong correlation with the stability of the meniscus in the mold, which may induce defects such as protective slag entrapment and surface transverse cracks; the deviation of the stopper positioning will destroy the mass flow balance from the ladle to the tundish, increasing the probability of center segregation and internal crack formation.

[0009] However, the above process time series data has typical characteristics such as high sampling frequency, significant non-stationary characteristics and low signal-to-noise ratio, and traditional time domain analysis methods have inherent limitations in feature extraction efficiency and engineering applicability, which seriously restricts the quality prediction and process optimization capability based on data-driven. SUMMARY

[0010] Therefore, the present application provides a continuous casting production process time series data segmentation method and device and computer equipment, which can provide more suitable data units for subsequent data analysis and processing by flexibly and reasonably segmenting the denoising time series data according to different situations.

[0011] According to one aspect of the present application, a continuous casting production process time series data segmentation method is provided, which comprises:

[0012] Real-time acquisition of time series data generated in the continuous casting production process, wherein the time series data includes at least one of withdrawal speed, stopper position and mold liquid level;

[0013] For any kind of time series data collected within a preset data segmentation period, the time series data is denoised to obtain denoising time series data;

[0014] The de-noised time series data is sequentially divided into multiple groups based on a preset unit time span. For any de-noised time series data group, a reference de-noised time series data group corresponding to the de-noised time series data group is determined in the reference de-noised time series data group based on a dynamic time warping distance between the reference de-noised time series data group corresponding to each of a plurality of preset key events in the continuous casting production process and the de-noised time series data group, wherein the preset key events include at least one of nozzle replacement, start of a casting pass, and end of a casting pass, and the preset key event corresponds to a reference de-noised time series data group and a preset duration;

[0015] The number of preset unit time spans that the de-noised time series data group should actually cover is determined according to a comparison relationship between the preset duration of the preset key event corresponding to the reference de-noised time series data group and the preset unit time span.

[0016] Based on the position of the de-noised time series data group in the de-noised time series data and the number of preset unit time spans that should actually be covered, a corresponding actual de-noised time series data segment of the de-noised time series data group in the de-noised time series data is determined, and the determined actual de-noised time series data segment is marked as the preset key event corresponding to the reference de-noised time series data group.

[0017] According to another aspect of the present application, a continuous casting production process time series data segmentation device is provided, which comprises:

[0018] A data acquisition module is configured to acquire time series data generated in a continuous casting production process in real time, wherein the time series data includes at least one of a withdrawal speed, a stopper position, and a mold liquid level.

[0019] A data de-noising module is configured to de-noise any time series data acquired within a preset data segmentation period to obtain de-noised time series data.

[0020] A data reference module is configured to sequentially divide the de-noised time series data into multiple groups based on a preset unit time span. For any de-noised time series data group, a reference de-noised time series data group corresponding to the de-noised time series data group is determined in the reference de-noised time series data group based on a dynamic time warping distance between the reference de-noised time series data group corresponding to each of a plurality of preset key events in the continuous casting production process and the de-noised time series data group, wherein the preset key events include at least one of nozzle replacement, start of a casting pass, and end of a casting pass, and the preset key event corresponds to a reference de-noised time series data group and a preset duration.

[0021] a segment strategy determination module configured to determine the number of preset unit time spans that the denoised time series data set should actually cover according to a comparison relationship between a preset duration of a preset key event corresponding to the reference denoised time series data set and the preset unit time span;

[0022] a data segment module configured to determine an actual denoised time series data segment corresponding to the denoised time series data set in the denoised time series data based on a position of the denoised time series data set in the denoised time series data and the number of preset unit time spans that the denoised time series data set should actually cover, and mark the determined actual denoised time series data segment as the preset key event corresponding to the reference denoised time series data set.

[0023] According to still another aspect of the present application, a computer device is provided, which comprises a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, and the processor implements the above-mentioned continuous casting production process time series data segmentation method when executing the program.

[0024] According to the above technical solution, the continuous casting production process time series data segmentation method and device, and the computer device provided by the present application collect time series data in the continuous casting production process, including casting speed, stopper position, and crystallizer liquid level. Any kind of time series data collected in a preset data segmentation period is denoised to obtain denoised time series data. The denoised time series data is grouped based on a preset unit time span, and the dynamic time warping distance of a reference denoised time series data set corresponding to a plurality of preset key events is calculated to determine a reference group. When there is a reference group, the number of preset unit time spans covered is determined according to a comparison between a preset duration of a key event corresponding to the reference group and the preset unit time span, and then the actual data segment is determined and marked. The denoised time series data is flexibly and reasonably segmented according to different situations, which can provide more suitable data units for subsequent data analysis and processing.

[0025] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0026] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0027] Figure 1 FIG. 1 shows a flowchart of a continuous casting production process time series data segmentation method provided by an embodiment of the present application;

[0028] Figure 2 A flowchart of a self-adaptive wavelet denoising method provided by an embodiment of the present application is shown.

[0029] Figure 3 A structure diagram of a continuous casting production process time series data segmentation device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0030] The present application will be described in detail below with reference to the accompanying 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.

[0031] A continuous casting production process time series data segmentation method is provided in the present embodiment, as shown in the figure, the method comprises: Figure 1

[0032] Step 101, real-time collection of time series data generated in the continuous casting production process, wherein the time series data comprises at least one of the following: casting speed, stopper position, and mold liquid level.

[0033] In the above embodiments of the present application, time series data such as casting speed, stopper position, and mold liquid level in the continuous casting production process is collected in real time, which can be carried out based on the sensor network and the industrial Internet of Things (IIoT) platform widely deployed in the continuous casting production site.

[0034] Specifically, the sensor network is like the “nerve perception end” of the continuous casting production, which is distributed in various key positions of the continuous casting production line. For example, in the casting speed control area, a high-precision casting speed sensor is installed, which can accurately measure the casting speed of the continuous casting billet at a certain frequency (such as ten or one hundred times per second), and convert the physical casting speed information into an electrical signal in real time. For the stopper position, the corresponding position sensor can continuously monitor the lifting state of the stopper and accurately capture the subtle changes in its position. The mold liquid level sensor is always concerned about the fluctuation of the liquid level in the mold, and timely feedbacks the liquid level data.

[0035] The industrial Internet of Things (IIoT) platform plays the role of “data aggregation hub”. Various data generated by the sensor are transmitted quickly and stably to the IIoT platform through wired or wireless communication methods such as industrial Ethernet, 5G wireless communication, etc. The platform has strong data receiving and processing capability, and can preliminarily arrange and check the massive data from numerous sensors to ensure the accuracy and integrity of the data.

[0036] ​In particular, according to the periodic fluctuation characteristics of continuous casting production, the IIoT platform can adaptively segment the original time series data. For example, in a complete casting cycle of continuous casting production, the change rules of different stages of casting speed, stopper position and crystallizer liquid level have their own characteristics. The platform will accurately segment the continuous time series data into data segments with independent meaning according to these internal rules.

[0037] Finally, the processed and segmented data is stored in a structured database of the local computing device. The structured database has a standardized data storage format, which facilitates efficient querying, analysis and mining of data for subsequent monitoring, optimization and fault diagnosis of the continuous casting production process, and provides solid data support.

[0038] Step 102, for any time series data collected within a predetermined data segmentation period, the time series data is denoised to obtain denoised time series data.

[0039] Next, for any time series data collected within a predetermined data segmentation period (e.g. 50 hours), the aforementioned time series data is denoised to obtain denoised time series data, which is prepared for further data processing.

[0040] Optionally, as shown in Figure 2 Step 102, the time series data is denoised to obtain denoised time series data, which specifically includes:

[0041] Step 1021, adaptively selecting the optimal wavelet basis and the optimal decomposition level corresponding to the time series data.

[0042] Step 1022, based on the selected optimal wavelet basis and optimal decomposition level, the time series data is wavelet-decomposed, threshold-processed, and then reconstructed for signal denoising to obtain the denoised time series data corresponding to the time series data.

[0043] In the above embodiments of the present application, the casting speed and stopper position data in the continuous casting process are usually disturbed by noise, such as equipment vibration, sensor error or instability of the steel liquid flow. The traditional wavelet denoising method relies on fixed wavelet basis (such as Daubechies-4) and decomposition level (such as level=2), which cannot adapt to the noise characteristics of different casting or process stages, and may cause insufficient denoising or signal distortion. Therefore, through the "adaptive wavelet denoising method" of the above embodiments of the present application, i.e. by automatically selecting the optimal wavelet basis and the optimal decomposition level, the denoising effect of the continuous casting process data can be optimized. The core steps are as follows:

[0044] 1. Construct a candidate wavelet basis set:

[0045] A set of multiple wavelet bases is constructed, which can be denoted as W, such as W = {db4, sym5, coif2}. These wavelet bases have good locality and smoothness in time series data processing, and are suitable for use in continuous casting processes. In W, db4, also known as Daubechies 4 wavelet, is a wavelet base proposed by the famous wavelet analysis scholar Ingrid Daubechies, and "4" represents a specific parameter related feature such as the order of vanishing moments. sym5, also known as Symlet 5 wavelet, is a wavelet base. Symlet wavelets are an improvement over Daubechies wavelets, with the property of approximate symmetry, and "5" is also related to the characteristics of the wavelet. coif2, also known as Coiflet 2 wavelet. The Chinese name is Coiflet 2 wavelet. Coiflet wavelets have unique properties such as higher vanishing moments, and "2" is used to distinguish wavelet bases with different parameter settings in the wavelet system.

[0046] 2. Multi-level decomposition:

[0047] For the input time series data, for each wavelet base, the maximum decomposition level is determined by the wavelet calculator, and wavelet decomposition is performed for each level to obtain the approximation coefficients and detail coefficients.

[0048] 3. Adaptive threshold processing:

[0049] The soft threshold processing method is used for the detail coefficients. The threshold is calculated according to the signal length and noise characteristics, specifically the standard deviation of the detail coefficients multiplied by a product related to the logarithm of the signal length. The soft threshold processing formula is: compare the absolute value of the detail coefficient with the threshold, take the larger value, and then assign the corresponding sign according to the positive or negative of the original detail coefficient.

[0050] 4. Denoising effect evaluation and optimal parameter selection:

[0051] For each combination of wavelet base and decomposition level, the denoised signal is reconstructed. The residual variance of the denoised signal and the original signal is calculated. The combination with the smallest residual variance is selected as the optimal wavelet base and the optimal decomposition level, and the corresponding denoised signal is returned. If the signal length does not meet the requirements, the original signal is returned directly.

[0052] 5. Signal reconstruction and denoising:

[0053] Based on the selected optimal wavelet base and optimal decomposition level, the time series data is decomposed by wavelet, threshold processed, and then the signal is reconstructed to obtain the denoised time series data (denoised time series data).

[0054] Therefore, by the above steps, the optimal wavelet basis and decomposition level can be automatically selected according to the noise characteristics of the time series data in the continuous casting process, effectively removing the noise while retaining the key features of the signal, and having higher robustness and adaptability compared to the traditional fixed parameter wavelet denoising method, which can significantly improve the accuracy of subsequent analysis.

[0055] In step 103, the denoised time series data is sequentially divided into multiple groups based on a preset unit time span. For any denoised time series data group, based on the dynamic time warping distance between the reference denoised time series data group corresponding to each of a plurality of preset key events in the continuous casting production process and the denoised time series data group, the reference denoised time series data group corresponding to the denoised time series data group is determined in the reference denoised time series data group, wherein the preset key events include at least one of nozzle replacement, start of pouring and end of pouring, and the preset key events correspond to a reference denoised time series data group and a preset duration.

[0056] Then, in the continuous casting production process, the occurrence of key events such as nozzle replacement, start of pouring and end of pouring will have a specific impact on time series data (such as casting speed, stopper position and mold liquid level, etc.). By calculating the dynamic time warping distance between the denoised time series data group and the reference denoised time series data group corresponding to each preset key event, the most similar reference denoised time series data group can be accurately found out, so as to accurately identify the key event corresponding to the current data segment, which helps to timely grasp the important nodes in the production process.

[0057] The denoised time series data is grouped by a preset unit time span (for example, 1 hour), and the most similar key event data group (reference denoised time series data group) is determined according to the dynamic time warping distance. This method fully considers the duration of the key event. Combined with the preset duration of each preset key event, the actual coverage range of the data grouping can be more reasonably determined, so that the data segmentation is more consistent with the key event process in the actual production, avoiding the situation that the key event data may be segmented or incomplete due to fixed length segmentation, and improving the quality and practicality of data segmentation.

[0058] After accurately identifying the key events and reasonably segmenting the data, the changes of the time series data before and after each key event can be more clearly observed and analyzed. For example, in the nozzle replacement event, the change rule of the data such as casting speed and stopper position can be analyzed in detail to find out possible problems or optimization points, providing more valuable information for the monitoring and analysis of the production process, and helping to timely discover potential production abnormalities and take corresponding measures.

[0059] Based on accurate identification of key events and reasonable segmentation of data, the continuous casting production process can be analyzed and evaluated more deeply. By comparing and studying the time series data under different key events, production experience is summarized, and production process parameters and production plans are optimized. For example, according to the data characteristics at the beginning and end of the casting, the supply of raw materials and the production rhythm are reasonably arranged, the production efficiency and product quality are improved, and the production cost is reduced.

[0060] The embodiments of the present application can also be integrated into the automatic control system of continuous casting production to realize automatic identification of key events and automatic segmentation of data. Reducing manual intervention, improving the degree of automation and intelligent level of production, making the production process more stable and reliable, and also reducing the work burden of operators.

[0061] In particular, regarding the preset duration of each of the nozzle replacement, the start of the casting, and the end of the casting, for example:

[0062] For nozzle replacement:

[0063] For small continuous casting machines, if the equipment is relatively small and the operation space is limited, but the overall process is relatively simple, the nozzle replacement operation may be relatively fast, and the preset duration can be 10 or 20 minutes. For example, some small experimental continuous casting equipment, workers can complete the nozzle disassembly, new nozzle installation and debugging and other work relatively quickly after being familiar with the operation process.

[0064] For large continuous casting machines, large continuous casting equipment has complex structure, and nozzle replacement may involve more auxiliary operations, such as partial shutdown of equipment, implementation of safety protection measures, etc. The preset duration can be 30 or 60 minutes or even longer. For example, on a large slab continuous casting machine, the nozzle replacement needs to be carried out strictly according to the operation procedures to ensure the safety and stable operation of the equipment, so it takes a long time.

[0065] For the start of the casting:

[0066] For single-flow continuous casting machines, the start of the casting mainly involves preparation work before casting, such as butt joint of the ladle, preheating of the tundish, and molten steel injection, etc. If the equipment and process are well prepared, the preset duration can be 15 or 30 minutes. For example, some simple billet continuous casting production lines can quickly complete the operations of the start of the casting after the molten steel arrives.

[0067] For multi-flow continuous casting machines, multiple flows are cast simultaneously, and the start of the casting needs to coordinate the operations of multiple flows to ensure the supply of molten steel, temperature control, etc. The preset duration can be 30 or 60 minutes. For example, a large multi-flow slab continuous casting machine needs to be checked and debugged comprehensively to ensure that each flow can start casting normally, so the time is relatively long.

[0068] For the end of casting, the end of casting is divided into two types: normal end of casting and abnormal end of casting.

[0069] For the normal end of casting, if the casting is normally ended according to the plan, for the small continuous casting machine, it is only necessary to complete the treatment of the remaining molten steel, and the equipment is simply cleaned and inspected, and the preset duration can be 20 or 30 minutes. For example, some small-scale alloy steel continuous casting production, after completing the predetermined casting amount, the tail blank is quickly treated and the equipment is cleaned.

[0070] For the abnormal end of casting, if the casting is ended due to abnormal conditions such as equipment failure, molten steel quality problem, etc., more detailed equipment inspection, fault elimination and quality analysis are required. For large continuous casting machines, the preset duration can be 1 or 2 hours or even longer. For example, in the process of continuous casting production, serious faults such as mold water leakage occur, and the damaged equipment needs to be checked comprehensively and the repair plan needs to be developed, so the duration of the end of casting will be greatly increased.

[0071] Optionally, in step 103, before determining the benchmark reference denoising time series data set corresponding to the denoising time series data set in the benchmark denoising time series data set based on the dynamic time warping distance between the denoising time series data set and the respective benchmark denoising time series data set of the plurality of preset key events in the continuous casting production process, it further comprises:

[0072] Step 106, selecting a benchmark denoising time series data set corresponding to any one of the preset key events, calculating the dynamic time warping distance between the denoising time series data set and the selected benchmark denoising time series data set based on the dynamic time warping distance calculation formula, wherein the dynamic time warping distance calculation formula is:

[0073] ,

[0074] denotes the dynamic time warping distance measuring the similarity between the denoising time series data set and the selected benchmark denoising time series data set , is the alignment path, is used to describe how to align and match the data points in and , until the alignment mode that makes the difference between and minimum is found, denotes the data point pair in the alignment path , i is the data point index in , and j is the data point index in , is used to measure the difference between the i-th data point in and the j-th data point in data points with in the data points the degree of difference between, denotes the path that makes the minimum is found by searching all possible alignment paths , as the dynamic time warping distance between and .

[0075] In the above embodiment of the application, taking the key event of "beginning of pouring" as an example, it is necessary to select the denoising time series data of the casting speed, stopper position or crystallizer liquid level recorded at the normal beginning of pouring as the reference denoising time series data set corresponding to the "beginning of pouring". Assuming that the casting speed data is selected as the analysis object, the reference denoising time series data set is a series of denoising processing recorded at the previous normal beginning of pouring, for example, denoted as:

[0076] ,

[0077] wherein, is the number of reference denoising time series data in the reference denoising time series data set.

[0078] For example, when the real-time monitoring of the continuous casting production process is performed, a denoising casting speed time series data is obtained, and n is the number of casting speed time series data. If it is necessary to judge whether the data corresponds to the "beginning of pouring" event, it is necessary to perform DTW (Dynamic Time Warping) calculation on the data and the reference denoising time series data set of the "beginning of pouring" selected in the foregoing.

[0079] Assuming , .

[0080] Firstly, all possible alignment paths are listed, for example:

[0081] , and the like.

[0082] For each alignment path, the is calculated. Taking as an example:

[0083] Then, the is calculated. For , that is, .

[0084] The above calculation is performed for all possible alignment paths, and the path that results in the smallest path is found, and the value corresponding to this path is .

[0085] Therefore, the dynamic time warping distance between the to-be-analyzed denoised time series data set and the reference denoised time series data set can be obtained, so as to judge the similarity between the two, and further identify the key events in the continuous casting production process.

[0086] In step 104, according to the comparison relationship between the preset duration of the preset key event corresponding to the reference denoised time series data set and the preset unit time span, the number of preset unit time spans that the denoised time series data set should actually cover is determined.

[0087] Then, in the continuous casting production, the key events such as nozzle replacement, casting start and casting end have their specific duration process. By comparing the preset duration with the preset unit time span to determine the number of covers, the range of key events in the time series data can be more accurately defined. For example, the nozzle replacement event may involve a series of complex operation steps, and its actual duration may span multiple preset unit time spans. Accurate calculation of the number of covers can completely capture the period of the event's influence on the time series data, avoiding data truncation and incomplete analysis of event characteristics.

[0088] The preset unit time span (which can be set to 1 hour) is the basic unit of data segmentation, but the duration of the key event is often different. According to the comparison relationship to determine the actual number of covers, the data segmentation can be more in line with the actual development process of the key event. For example, the casting start event may have related time series data changes gradually unfold within a specific time period. Reasonable determination of the number of covers can ensure that the data segment containing the complete characteristics of the event is accurately divided out, improving the quality of data segmentation and providing a more reasonable data basis for subsequent data analysis.

[0089] Accurate determination of the number of preset unit time spans that the denoised time series data set should actually cover helps to more accurately analyze the influence of key events on time series data. Taking the mold level data as an example, in the nozzle replacement event, the liquid level may fluctuate due to changes in steel flow and other factors. A data segment that completely contains the event's influence period can more realistically reflect the variation of the liquid level, thereby more accurately analyzing the relationship between the event and the data changes, and providing a reliable basis for the optimization of production processes.

[0090] In the continuous casting production process, it is crucial to monitor the occurrence and impact of key events in real time. By determining the coverage number, the location and range of key events in time series data can be identified in a timely and accurate manner, allowing production monitoring personnel to more intuitively observe the changes in various parameters when the event occurs. For example, for the end of the casting event, monitoring personnel can accurately divide the data segment to timely discover abnormal data changes related to the end of the casting, take appropriate measures, and ensure the stable operation of the production process.

[0091] Based on accurate data segmentation and key event analysis results, strong support can be provided for production decision-making. After understanding the duration and impact range of key events, production managers can reasonably arrange production plans and optimize resource allocation. For example, based on the coverage number and impact period of the nozzle replacement event, spare nozzles and related tools can be prepared in advance to reduce replacement time and improve production efficiency; or based on the start and end times of the casting, the steel supply rhythm can be adjusted to avoid production interruptions or resource waste.

[0092] Optionally, in step 104, the number of preset unit time spans that the denoised time series data set should actually cover is determined according to the comparison relationship between the preset duration of the preset key event corresponding to the benchmark denoised time series data set and the preset unit time span, specifically including:

[0093] Step 1041, if the determined preset duration of the preset key event corresponding to the benchmark denoised time series data set is less than or equal to the preset unit time span, the number of preset unit time spans that the denoised time series data set should actually cover is 1.

[0094] Step 1042, if the determined preset duration of the preset key event corresponding to the benchmark denoised time series data set is greater than the preset unit time span, the preset duration is converted into a multiple of the preset unit time span and rounded up to obtain the number of preset unit time spans that the denoised time series data set should actually cover.

[0095] In the above embodiments of the present application, there are two cases:

[0096] Case 1: The preset duration is less than or equal to the preset unit time span:

[0097] The preset unit time span is set to be 1 hour for example. In case one, the preset duration of the preset key event corresponding to the identified benchmark reference denoising time series data set is less than or equal to 1 hour, for example, the preset duration is 30 minutes. Since the duration of the key event does not exceed the preset 1 hour unit time span, from the perspective of data segmentation and covering the complete impact of the key event, a data segment of one preset unit time span (1 hour) is sufficient to contain the impact of the key event on the time series data. Therefore, the number of preset unit time spans that the aforementioned denoising time series data set should actually cover is 1. That is, during data segmentation, based on the position of the current denoising time series data set, a data segment of 1 hour in length can completely capture the data characteristics related to the key event.

[0098] In case two, the preset duration is greater than the preset unit time span:

[0099] Similarly, the preset unit time span is 1 hour. If the preset duration of the preset key event corresponding to the identified benchmark reference denoising time series data set is greater than 1 hour, for example, the preset duration is 1 hour and 20 minutes.

[0100] First, the preset duration of 1 hour and 20 minutes is converted into a multiple of 1 hour. After conversion, 1 hour and 20 minutes is approximately 1.33 times the preset unit time span. Then, the upward rounding operation is performed. Even if the remaining time is less than one complete preset unit time span, it is still part of the key event duration. In order to ensure that the data segment can completely cover the impact of the key event, it needs to be rounded up. After rounding up 1.33, we get 2. Therefore, the number of preset unit time spans that the aforementioned denoising time series data set should actually cover is 2. This means that during data segmentation, a data segment of 2 hours in length is needed to completely contain the impact of the key event on the time series data, so as to facilitate subsequent accurate analysis and processing.

[0101] Optionally, in step 104, based on the dynamic time warping distance between the plurality of benchmark reference denoising time series data sets corresponding to the plurality of preset key events in the continuous casting production process and the denoising time series data set, the benchmark reference denoising time series data set corresponding to the denoising time series data set is determined in the benchmark reference denoising time series data set. Specifically, it includes:

[0102] Step 1043, if , the selected benchmark reference denoising time series data set is the most similar benchmark reference denoising time series data set to the denoising time series data set, wherein is a preset similarity threshold for judging the similarity between and , is a preset threshold coefficient, is a set of historical calculated dynamic time warping distances, represents the maximum value in

[0103] In the above embodiments of the present application, if the dynamic time warping distance is less than a preset similarity threshold δ (i.e. , it is considered that there exists a reference denoised time series data group most similar to the current denoised time series data group.

[0104] In particular, if the dynamic time warping distance is greater than or equal to the preset similarity threshold δ (i.e. ), it means that the current denoised time series data group is not similar to the reference denoised time series data group . In this case, it is possible to continue searching for other possible reference data groups (reference denoised time series data groups) in the set of reference denoised time series data groups to determine whether there exists a reference denoised time series data group more similar to the current denoised time series data group. Specifically, the following steps can be taken:

[0105] 1. Traverse all reference data groups (reference denoised time series data groups): traverse all reference denoised time series data groups corresponding to each of the plurality of preset key events in the continuous casting production process.

[0106] 2. Calculate the dynamic time warping distance: calculate the DTW distance between the current denoised time series data group and each reference denoised time series data group.

[0107] 3. Compare with the threshold value: compare the calculated DTW distance with the preset similarity threshold δ.

[0108] 4. Determine the reference: if there exists any reference denoised time series data group with a DTW distance less than δ from the current denoised time series data group, it is determined as the reference denoised time series data group; if the DTW distances of all reference data groups are greater than or equal to δ, it may be necessary to re-examine the threshold setting or consider other factors to determine the reference (i.e. to re-determine δ).

[0109] Therefore, when the DTW is greater than or equal to the preset similarity threshold δ, it is possible to continue searching until a more suitable reference denoised time series data group is found.

[0110] Step 105, based on the position of the denoised time series data set in the denoised time series data, and the number of preset unit time spans that should be covered in actuality, determine the corresponding actual denoised time series data segment of the denoised time series data set in the denoised time series data, and mark the determined actual denoised time series data segment as the preset key event corresponding to the reference denoised time series data set.

[0111] In the above embodiments of the present application, for example in the continuous casting production process, the preset unit time span is 1 hour. The denoised time series data is the drawing speed data recorded in chronological order, and the total duration is 5 hours, with each hour as a basic data grouping unit (preset unit time span), and a total of 5 basic data groups (denoised time series data sets).

[0112] Next, the position of the denoised time series data set is determined. Assuming that in the analysis process, it is determined that the 3rd basic data group, i.e. the denoised time series data set (i.e. the data set from the 2nd hour to the 3rd hour) is the most similar to the reference denoised time series data set corresponding to the water nozzle replacement preset key event. Then the position of the denoised time series data set in the denoised time series data is the 3rd position.

[0113] Next, the number of preset unit time spans that should be covered in actuality is determined. The preset duration of the water nozzle replacement preset key event is, for example, 1 hour and 20 minutes. Since the preset duration is greater than the preset unit time span (1 hour), the preset duration is converted into multiples of the preset unit time span and rounded up. After conversion, 1 hour and 20 minutes is approximately 1.33 times 1 hour, and rounding up gives the number of preset unit time spans that should be covered in actuality as 2. According to the position of the denoised time series data set (3rd position) and the number of preset unit time spans that should be covered in actuality (2), since 2 preset unit time spans are to be covered and the data set starts at the 2nd hour, the actual denoised time series data segment is from the 2nd hour, covering the data from the 2nd hour to the 4th hour (the 2nd-3rd hour is the original data set, plus the adjacent next 1 hour of data to meet the requirement of covering 2 unit time spans).

[0114] Next, the preset key event is marked, and the determined actual denoised time series data segment from the 2nd hour to the 4th hour is marked as the water nozzle replacement preset key event. In this way, in subsequent analysis of the continuous casting production time series data, the data segment corresponding to the water nozzle replacement event can be clearly identified, and the impact of the event on the drawing speed and other data can be analyzed, providing a basis for monitoring and optimizing the production process.

[0115] Through the above steps, the actual data segment (actual denoising time series data segment) corresponding to the key event can be accurately determined from the denoising time series data, and is marked, which helps to deeply analyze the relationship between various events in the continuous casting production process and data changes.

[0116] Optionally, the continuous casting production process time series data segmentation method further comprises:

[0117] In step 107, when no reference denoising time series data group is determined, the divided denoising time series data group is directly used as an actual denoising time series data segment.

[0118] In the above embodiments of the present application, in the continuous casting production process, the time series data is continuous and has inherent logical correlation. When no reference denoising time series data group is determined, the divided denoising time series data group is directly used as an actual denoising time series data segment, avoiding data truncation or segmentation that may be caused by forcibly matching key events. For example, the change of the pulling speed data in different time periods is continuous. If the data group is segmented in order to match a certain key event, the change trend and integrity of the data may be damaged. However, directly using the divided data group can completely retain the characteristics of the data in a specific time period, and provide a more real and reliable basis for subsequent data analysis.

[0119] If the actual data segment is determined in the manner of matching the key event without determining the reference denoising time series data group, the complexity and computational amount of data processing will be increased. Directly using the divided data group as an actual data segment (actual denoising time series data segment) reduces the additional calculation steps and judgment logic, and makes the data processing procedure more simple and efficient. For example, the calculation and comparison of dynamic time warping distance to determine the similarity with the key event are not needed, which saves the calculation resources and time cost.

[0120] For most time series data in the continuous casting production process, it may not be directly related to the preset key event. Directly using these divided data groups as actual data segments facilitates the regular data analysis, such as statistical average, variance, trend analysis, etc. Taking the mold level data as an example, directly using the divided data segment can more conveniently analyze the fluctuation of the mold level in different time periods, and timely find the abnormal points or rules in the data, without considering the influence of the key event, so that the data analysis is more focused on the characteristics of the data itself.

[0121] The continuous casting production process is complex and diverse, and various unexpected situations and events can occur. Directly using the divided data set as an actual data segment (actual denoising time series data segment) can adapt to such diverse production situations. Even if no matching key event is found, the data can still be effectively segmented and analyzed to timely discover potential problems or changes in the production process. For example, some temporary, unexpected fluctuations or abnormal situations occur during production. By directly using the divided data segment, these information can be better captured to provide a reference for production adjustment and optimization.

[0122] Directly determining the actual data segment according to the division manner can better maintain the time sequence of the data. This is very important for analyzing the time sequence relationship of the data in the continuous casting production process, such as analyzing the relationship between the change of the pulling speed over time and the product quality. If the time sequence of the data is disturbed due to matching the key event, the analysis result can be biased. Directly using the divided data segment can ensure the time sequence integrity of the data, making the analysis result more accurate and reliable.

[0123] Optionally, the continuous casting production process time series data segmentation method further includes:

[0124] In step 108, the denoising time series data in the actual denoising time series data segment is processed by dimension reduction and symbolic mapping to obtain string data converted based on the actual denoising time series data segment.

[0125] In the above embodiments of the present application, the denoising time series data in the actual denoising time series data segment usually contains a large number of data points, which occupies more storage resources. By dimension reduction processing, the high-dimensional time series data is mapped to a low-dimensional discrete symbol space, which can significantly reduce the data amount. For example, an actual denoising time series data segment originally requiring storage of tens of thousands of data points can only need to store a few dozen or even fewer symbols after processing, greatly saving storage space and reducing storage cost.

[0126] Optionally, in step 108, the denoising time series data in the actual denoising time series data segment is processed by dimension reduction and symbolic mapping, specifically including:

[0127] In step 1081, the denoising time series data in the actual denoising time series data segment is processed by dimension reduction using a piecewise aggregate approximation method until the denoising time series data in the actual denoising time series data segment is mapped to a discrete symbol space.

[0128] In the above embodiment of the present application, for example, in the continuous casting production process, an actual denoising time series data segment is determined through the preceding steps, which records the withdrawal speed data in a certain time period, the time length is 3 hours, and the data acquisition frequency can be once per minute, so the data segment contains 3x60=180 data points, and suppose the sequence composed of these data points is .

[0129] First, determine how many paragraphs the data segment will be divided into. For example, decide to divide the 180 data points into 18 paragraphs, each containing 180÷18=10 data points.

[0130] For each paragraph, calculate the mean of the data points therein. Through calculation, the mean sequence of the 18 paragraphs is obtained , so the preliminary dimension reduction is completed, reducing the data from 180 dimensions to 18 dimensions.

[0131] Next, set the symbol set, and pre-set a discrete symbol set, which can determine the mapping rule according to the distribution of the data. For example, divide the value range of the data into 4 intervals, each interval corresponding to a symbol. According to the above mapping rule, each element in the mean sequence Y is mapped to the corresponding symbol, and finally a string data composed of symbols is obtained. For example, the string obtained after mapping is , and this string is the string data converted from the actual denoising time series data segment.

[0132] Through the above steps, the withdrawal speed data in the actual denoising time series data segment is processed by the PAA (Piecewise Aggregate Approximation) method, and is mapped to the discrete symbol space, obtaining string data convenient for subsequent analysis and processing.

[0133] By applying the technical solution of the embodiment, the complete processing flow of time series data in the continuous casting production process is covered, from data acquisition, denoising, key event identification to data segmentation, forming an organic whole. Through fine processing and analysis of the data, abnormal conditions in the production process can be found in time, production process parameters can be optimized, product quality and production efficiency can be improved, and production cost can be reduced, providing strong support for stable operation and continuous improvement of continuous casting production.

[0134] Further, as a specific implementation of the Figure 1 method, the embodiment of the present application provides a continuous casting production process time series data segmentation device, as shown in Figure 3 , which comprises:

[0135] The data acquisition module 201 is configured to acquire time series data generated in the continuous casting production process in real time, wherein the time series data comprises at least one of a withdrawal speed, a stopper position and a mold liquid level height.

[0136] The data denoising module 202 is configured to perform denoising processing on any kind of time series data acquired within a preset data segmentation period to obtain denoised time series data.

[0137] The data benchmarking module 203 is configured to divide the denoised time series data into groups in sequence based on a preset unit time span, and for any denoised time series data group, determine a benchmarking reference denoised time series data group corresponding to the denoised time series data group from reference denoised time series data groups corresponding to a plurality of preset key events in the continuous casting production process based on dynamic time warping distances between the reference denoised time series data groups and the denoised time series data group, wherein the preset key events comprise at least one of nozzle replacement, start of a pouring and end of a pouring, and the preset key event corresponds to a reference denoised time series data group and a preset duration.

[0138] The segmentation strategy determination module 204 is configured to determine a number of preset unit time spans that should actually be covered by the denoised time series data group according to a comparison relationship between the preset duration of the preset key event corresponding to the benchmarking reference denoised time series data group and the preset unit time span.

[0139] The data segmentation module 205 is configured to determine actual denoised time series data segments corresponding to the denoised time series data group in the denoised time series data based on a position of the denoised time series data group in the denoised time series data and the number of preset unit time spans that should actually be covered, and mark the determined actual denoised time series data segments as the preset key event corresponding to the benchmarking reference denoised time series data group.

[0140] It should be noted that other corresponding descriptions of the functions of the device for segmenting time series data in the continuous casting production process provided in the embodiments of the present application can be referred to the corresponding descriptions in the method, which will not be described here. Figures 1 to 2 The method, and the virtual device embodiment shown in

[0141] Based on the above method as Figures 1 to 2 indicated, and Figure 3 the virtual device embodiment, in order to achieve the above purpose, the embodiments of the present application also provide a computer device, which can be a personal computer, a server, a network device, etc., the computer device comprises a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the continuous casting production process time series data segmentation method as Figures 1 to 2 indicated above.

[0142] Optionally, the computer device can further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, and the like. The user interface can include a display, an input unit such as a keyboard, and the like. Optionally, the user interface can further include a USB interface, a card reader interface, and the like. The network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), and the like.

[0143] Those skilled in the art can understand that the computer device structure provided by the embodiment does not constitute a limitation on the computer device, and can include more or fewer components, or combine certain components, or different component arrangements.

[0144] The storage medium can further include an operating system, a network communication module. The operating system is a program for managing and saving computer device hardware and software resources, supporting the running of information processing programs and other software and / or programs. The network communication module is used to realize the communication between the components in the storage medium and the communication with other hardware and software in the entity device.

[0145] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware platform, or by hardware to collect time series data in continuous casting production process, including casting speed, stopper position and crystallizer liquid level, etc. Any kind of time series data collected in the preset data segmentation period is denoised to obtain denoised time series data. Based on the preset unit time span, it is grouped, and the dynamic time warping distance of each group and the reference denoised time series data group corresponding to a plurality of preset key events is calculated to determine the reference group. When there is a reference group, the preset duration corresponding to the key event and the preset unit time span are compared to determine the number of preset unit time spans, and then the actual data segment is determined and marked. By flexibly and reasonably segmenting the denoised time series data according to different situations, more suitable data units can be provided for subsequent data analysis and processing.

[0146] Those skilled in the art can understand that the drawings are only a schematic diagram of a preferred implementation scenario, and the modules or flows in the drawings are not necessarily necessary for implementing the present application. Those skilled in the art can understand that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the description of the implementation scenario, or can be correspondingly changed and located in one or more devices different from the implementation scenario. The modules of the above implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0147] The above application number is only for description, and does not represent the advantages and disadvantages of the implementation scene. The above disclosure is only some specific implementation scenes of the application, but the application is not limited thereto, and any changes made by those skilled in the art shall fall within the protection scope of the application.

Claims

1. A method for segmenting time series data in a continuous casting production process, characterized in that, The continuous casting production process time series data segmentation method comprises: Real-time acquisition of time series data generated in the continuous casting production process, wherein the time series data comprises at least one of a withdrawal speed, a stopper position and a crystallizer liquid level; For any kind of time series data collected within a preset data segmentation period, the time series data is subjected to denoising processing to obtain denoised time series data; Based on a preset unit time span, the denoised time series data is sequentially divided into multiple groups, for any denoised time series data group, based on the dynamic time warping distance between the reference denoised time series data group corresponding to each of a plurality of preset key events in the continuous casting production process and the denoised time series data group, a target reference denoised time series data group corresponding to the denoised time series data group is determined in the reference denoised time series data group, wherein the preset key events comprise at least one of a nozzle replacement, a pouring start and a pouring end, and the preset key event corresponds to a reference denoised time series data group and a preset duration; According to the comparison relationship between the preset duration of the preset key event corresponding to the target reference denoised time series data group and the preset unit time span, the number of preset unit time spans that the denoised time series data group should actually cover is determined; Based on the position of the denoised time series data group in the denoised time series data and the number of preset unit time spans that should actually be covered, the actual denoised time series data segment corresponding to the denoised time series data group in the denoised time series data is determined, and the determined actual denoised time series data segment is marked as the preset key event corresponding to the target reference denoised time series data group.

2. The method of claim 1, wherein, The determination of the number of preset unit time spans that the denoised time series data group should actually cover according to the comparison relationship between the preset duration of the preset key event corresponding to the target reference denoised time series data group and the preset unit time span comprises: If the preset duration of the preset key event corresponding to the target reference denoised time series data group determined is less than or equal to the preset unit time span, the number of preset unit time spans that the denoised time series data group should actually cover is 1; If the preset duration of the preset key event corresponding to the target reference denoised time series data group determined is greater than the preset unit time span, the preset duration is converted into a multiple of the preset unit time span and is rounded up to obtain the number of preset unit time spans that the denoised time series data group should actually cover.

3. The method of claim 1, wherein the method further comprises: The continuous casting production process time series data segmentation method further comprises: When no target reference denoised time series data group is determined, the divided denoised time series data group is directly used as an actual denoised time series data segment.

4. The method of claim 1, wherein, The denoising processing of the time series data to obtain denoised time series data comprises: Adaptive selection of an optimal wavelet basis and an optimal decomposition level corresponding to the time series data; The time series data is decomposed by a wavelet, is processed by a threshold, and is reconstructed to obtain denoised time series data corresponding to the time series data.

5. The method of claim 1, wherein, The dynamic time warping distance between the denoised time series data group and the reference denoised time series data group corresponding to each of the plurality of preset key events in the continuous casting production process is calculated, and the reference denoised time series data group corresponding to the denoised time series data group is determined in the reference denoised time series data group. The dynamic time warping distance between the denoised time series data group and the reference denoised time series data group corresponding to each of the plurality of preset key events in the continuous casting production process is calculated, and the reference denoised time series data group corresponding to the denoised time series data group is determined in the reference denoised time series data group. ， denotes the similarity between the denoised time series data set and the selected reference denoised time series data set , the dynamic time warping distance between , the alignment path, for describing how to align match the data points in and until the alignment that makes the difference between and minimum is found, denotes the pair of data points in the alignment path , i is the index of data points in , j is the index of data points in , for measuring the difference between the i-th data point in and the j-th data point in , , , denotes the path that makes the difference between minimum is found by searching all possible alignment paths , as the dynamic time warping distance between and .​​ 6. The method of claim 5, wherein, The continuous casting production process time series data segmentation method further comprises: If , the selected reference denoised time series data set is the reference denoised time series data set most similar to the denoised time series data set, wherein is a preset similarity threshold for judging the similarity between and , is a preset threshold coefficient, is a set of dynamically time warping distances calculated historically, represents the maximum value in .

7. The method according to any one of claims 1 to 6, wherein, The denoised time series data in the actual denoised time series data segment is processed by dimension reduction and symbolic mapping to obtain string data converted from the actual denoised time series data segment. The denoised time series data in the actual denoised time series data segment is processed by dimension reduction and symbolic mapping to obtain string data converted from the actual denoised time series data segment.

8. The method of claim 7, wherein, The denoised time series data in the actual denoised time series data segment is processed by dimension reduction and symbolic mapping to obtain string data converted from the actual denoised time series data segment. The continuous casting production process time series data segmentation device comprises:

9. A continuous casting production process time series data segmentation apparatus characterized by comprising: The data acquisition module is configured to acquire time series data generated in the continuous casting production process in real time, wherein the time series data comprises at least one of a withdrawal speed, a stopper position, and a mold liquid level; The data denoising module is configured to denoise any time series data acquired within a preset data segmentation period to obtain denoised time series data; The data reference module is configured to divide the denoised time series data into a plurality of groups based on a preset unit time span, and for any denoised time series data group, to determine a reference denoised time series data group corresponding to the denoised time series data group in reference denoised time series data groups corresponding to each of a plurality of preset key events in the continuous casting production process based on a dynamic time warping distance between the denoised time series data group and the reference denoised time series data group, wherein the preset key events comprise at least one of a nozzle replacement, a pouring start, and a pouring end, and each preset key event corresponds to a reference denoised time series data group and a preset duration. ​ A segment strategy determination module is configured to determine the number of preset unit time spans that the denoised time series data set should actually cover according to a comparison relationship between a preset duration of a preset key event corresponding to the reference denoised time series data set and a preset unit time span. A data segment module is configured to determine an actual denoised time series data segment corresponding to the denoised time series data set in the denoised time series data based on a position of the denoised time series data set in the denoised time series data and the number of preset unit time spans that the denoised time series data set should actually cover, and mark the determined actual denoised time series data segment as the preset key event corresponding to the reference denoised time series data set.

10. A computer device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, The processor implements the computer program to realize the continuous casting production process time series data segmenting method in any one of claims 1 to 8.

Images