Continuous casting crystallizer liquid level abnormal fluctuation prediction method, tracing factor analysis method and system

Abstract

The invention provides a continuous casting crystallizer liquid level abnormal fluctuation prediction method, a factor tracing analysis method and a system, and belongs to the field of continuous casting processes. According to the liquid level abnormal fluctuation prediction method, a liquid level difference threshold value and a duration threshold value are set respectively, a three-dimensional liquid level abnormal fluctuation matrix composed of all abnormal fluctuation sets is constructed, a crystallizer liquid level fluctuation abnormal prediction model is constructed based on the liquid level abnormal matrix at the t moment, and the input of the model is historical continuous casting parameters sampled at the t moment. The output is a liquid level anomaly matrix; and after training, inputting continuous casting parameters at a to-be-predicted moment tf, outputting a liquid level anomaly matrix of the tf, solving an average value of N * M predicted values in the matrix at the tf moment, taking the average value as a liquid level anomaly index at the tf moment, judging whether liquid level fluctuation anomaly exists at the to-be-predicted moment, and further performing tracing analysis. According to the invention, the prediction of the fluctuation abnormity of the liquid level is extended to be within the preset duration range, and the accuracy of the fluctuation abnormity prediction is improved.

Description

Technical field:

[0001] The present invention belongs to the field of continuous casting technology, and in particular relates to a method for predicting abnormal fluctuations in the liquid level of a continuous casting crystallizer, a method for tracing the cause of the fluctuations, and a system for the tracing cause of the fluctuations. Background technology:

[0002] During the continuous casting process, mold level stability is crucial to slab quality. Mold level fluctuations not only degrade slab quality but can also cause slag entrainment and even molten steel leaks. The causes of mold level fluctuation (MLF) are complex and include submerged nozzle blockage, varying casting speeds, unstable argon flow, mold vibration, and roll compression of molten steel. Predicting and analyzing the causes of mold level anomalies in continuous casting are crucial for reducing the accident rate and improving product quality.

[0003] In the prior art, data analysis and simulation are generally used to identify or predict mold level fluctuations based on big data. For example, Chinese patent No. 202210767436.8 discloses a method and system for identifying abnormal mold level fluctuations. The method uses mold level fluctuation data and a method based on a combination of fast Fourier transform and wavelet entropy to comprehensively analyze the fluctuation of molten steel in the mold over a period of time (different heats or different castings), accurately locate the time when abnormal level fluctuations occur, and quickly trace the cause of the fluctuations. This method can be applied to both offline historical data analysis and online assessment of mold level fluctuations, thereby reducing the impact of mold level fluctuations on the quality of the ingot, reducing ingot quality losses, and improving continuous casting production efficiency. However, the prior art, such as the above-mentioned method, generally only focuses on the prediction of the actual liquid level value. The method of directly predicting the actual liquid level value is more accurate in a short period of time, but as time goes on, its prediction accuracy will decrease significantly, and it cannot effectively reflect the possibility of abnormal fluctuations occurring in the next minute. In addition, only predicting the actual liquid level value cannot directly reflect the severity of the abnormal liquid level fluctuations that may occur in the future. Summary of the invention:

[0004] In order to solve the above problems, an embodiment of the present invention provides a method for predicting abnormal liquid level fluctuations in a continuous casting crystallizer, a causal analysis method and a system. By defining a liquid level abnormal fluctuation index to reflect the probability and severity of abnormal liquid level fluctuations in a future period of time, and using a model to predict the index, the possibility of future abnormal liquid level fluctuations and their potential severity can be more directly reflected. At the same time, the SHAP (SHapley Additive exPlanations) analysis method is adopted to calculate the influence of each input variable on the liquid level abnormality by excluding the liquid level value itself, and the results are displayed in the form of images, so that it can be intuitively seen which factors are related to the liquid level fluctuation at a specific moment.

[0005] In order to achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows:

[0006] In a first aspect, an embodiment of the present invention provides a method for predicting abnormal fluctuations in liquid level in a continuous casting mold, the method comprising the following steps:

[0007] Step S1, obtain the continuous casting parameters of historical time 1-T and the time to be predicted t f Continuous casting parameters; construct historical data sets based on historical continuous casting parameters;

[0008] Step S2, set the set K = {k i ,i=1,2,…,N}, where k i The threshold value of the difference between the preset liquid level values ​​of the continuous casting crystallizer; if the difference between the liquid level value at time t and the liquid level value within T0 seconds before exceeds the threshold value k i , then it is determined that there is a threshold k at time t i Abnormal liquid level fluctuations All moments 1-T in the historical data set are based on the threshold k i The fluctuation of For each threshold k i Calculate and obtain the time series of N abnormal fluctuation sets

[0009] Step S3, set the set Δ={δ j ,j=1,2,…,M}; where δ j Represents the preset time threshold; if δ after time t j Occurs within seconds based on threshold k i If the liquid level fluctuates abnormally, it is determined that δ j There is a threshold k in the time period i Abnormal liquid level fluctuations δ of all time points 1-T in the historical data set jThe fluctuations within the period form a time series, recorded as

[0010] Step S4: for each abnormal fluctuation set MLF ki , then based on the threshold δ j The calculation is performed to obtain the time series of M abnormal fluctuation sets, and the N×M×T liquid level abnormal fluctuation matrix MLF∈R is obtained for all abnormal fluctuation sets. N×M×T ;

[0011] Step S5: Based on the liquid level anomaly matrix MLF(t) at time t, a mold liquid level fluctuation anomaly prediction model is constructed. The input of the model is the historical continuous casting parameters sampled at time t, and the output is the liquid level anomaly matrix at time t.

[0012] Step S6: training the fluctuation anomaly prediction model based on the historical data set to obtain a mature fluctuation anomaly prediction model;

[0013] Step S7: The current time to be predicted t f Continuous casting parameters X(t f ) is input into the mature fluctuation anomaly prediction model, and the output is t f The liquid level abnormal matrix MLF(t f )’s estimated output(t f )∈R N×M ; Find the output of the model (t f ) is used as the average value of the predicted values ​​of all matrix elements in the liquid level anomaly index mlf(t f )

[0014] Step S8: Based on the predicted value of the liquid level abnormality index Assess the risk of abnormal liquid level fluctuations occurring after the predicted time.

[0015] As a preferred embodiment of the present invention, the abnormal fluctuation of the liquid level in step S2 The calculation formula is as follows:

[0016]

[0017] In formula (1), L(t) represents the crystallizer liquid level value at time t.

[0018] As a preferred embodiment of the present invention, the abnormal fluctuation of the liquid level in step S3 The calculation formula is as follows:

[0019]

[0020] As a preferred embodiment of the present invention, in step S5, the input of the model is the historical continuous casting parameters sampled at time t. For two-dimensional data, T1 represents the length of the time series and D represents the feature dimension;

[0021] The continuous casting parameters are subjected to mean downsampling processing, with a sampling interval of t s , the length of the sequence after sampling is w, and the formula for mean downsampling is:

[0022]

[0023] In formula (5), mean represents the mean of the time series in the time dimension;

[0024] The input of the model at time t is input(t)∈R w×D is obtained by s ) to t are downsampled to obtain:

[0025]

[0026] As a preferred embodiment of the present invention, D, which represents the characteristic dimension, includes stopper rod opening, pulling speed, and width adjustment.

[0027] As a preferred embodiment of the present invention, in step S7, the average value of the liquid level abnormality index prediction value is calculated using the following formula:

[0028]

[0029] In formula (6), Output(t f ) matrix.

[0030] As a preferred embodiment of the present invention, step S8 further includes: if If the value is greater than or equal to the preset threshold, it is determined that there is an abnormal fluctuation risk; if If the value is less than the preset threshold but greater than the value at the previous moment, it is judged as an increased risk; if If it is less than the preset threshold and less than the value at the previous moment, it is determined to be risk-free.

[0031] In a second aspect, an embodiment of the present invention further provides a continuous casting mold liquid level abnormal fluctuation prediction system, the system comprising: a data acquisition module, a liquid level threshold k setting module, a k-based fluctuation sequence calculation module, a time threshold δ setting module, a δ-based fluctuation sequence calculation module, a three-dimensional fluctuation matrix construction module, a prediction model construction module, a prediction model training module, an abnormal index prediction module and a prediction result output module; wherein,

[0032] The data acquisition module is used to obtain the continuous casting parameters at the historical time 1-T and the time to be predicted t f Continuous casting parameters; build historical data sets based on historical continuous casting parameters;

[0033] The liquid level threshold k setting module is used to set the set K={k i ,i=1,2,…,N}, where k i A threshold value representing a difference between preset liquid level values ​​of a continuous casting mold;

[0034] The k-based fluctuation sequence calculation module is used to calculate the value of the liquid level at time t when the difference between the liquid level value at time t and the liquid level value within the previous T0 seconds exceeds the threshold k. i , determine whether there is a threshold k at time t i Abnormal liquid level fluctuations And calculate all the moments 1-T in the historical data set based on the threshold k i The fluctuation of For each threshold k i Calculate and obtain the time series of N abnormal fluctuation sets

[0035] The time threshold δ setting module is used to set the set Δ={δ j ,j=1,2,…,M}; where δ j Indicates the preset duration threshold;

[0036] The δ-based fluctuation sequence calculation module is used when δ after time t j Occurs within seconds based on threshold k i When the liquid level fluctuates abnormally, determine the δ j There is a threshold k in the time period i Abnormal liquid level fluctuations δ of all time points 1-T in the historical data set j The fluctuations within the period form a time series, recorded as

[0037] The three-dimensional fluctuation matrix construction module is used for each abnormal fluctuation set MLF ki , then based on the threshold δ j The calculation is performed to obtain the time series of M abnormal fluctuation sets, and the N×M×T liquid level abnormal fluctuation matrix MLF∈R is obtained for all abnormal fluctuation sets. N×M×T ;

[0038] The prediction model building module is used to build a mold liquid level fluctuation abnormality prediction model based on the liquid level abnormality matrix MLF(t) at time t. The input of the model is the historical continuous casting parameters sampled at time t, and the output is the liquid level abnormality matrix at time t;

[0039] The prediction model training module is used to train the fluctuation anomaly prediction model based on the historical data set to obtain a mature fluctuation anomaly prediction model;

[0040] The abnormal index prediction module is used to calculate the current predicted time t f Continuous casting parameters X(t f ) is input into the mature fluctuation anomaly prediction model, and the output is t f The liquid level abnormal matrix MLF(t f )’s estimated output(t f )∈R N×M ; Find the output of the model (t f ) is used as the average value of the predicted values ​​of all matrix elements in the liquid level anomaly index mlf(t f )

[0041] The prediction result output module is used to output the predicted value of the liquid level abnormality index. Assess the risk of abnormal liquid level fluctuations occurring after the predicted time.

[0042] In a third aspect, an embodiment of the present invention further provides a method for tracing the cause of abnormal fluctuations in the liquid level of a continuous casting mold, the method comprising steps S1 to S8 as described above; and further comprising:

[0043] Step S9, use the GradientExplainer to calculate input (t f )right The additive interpretation of the Shapley SHAP value and obtain a value that is consistent with the input(t f ) SHAP matrix shap(t f )∈R w×dim , each element in the matrix represents input(t f ) in the corresponding position variable;

[0044] Step S10, accumulating the SHAP values ​​in the time dimension to obtain the SHAP value of each feature;

[0045] Step S11: The shape of all features at each moment d (t f) are aggregated into time series to visually display and analyze the impact of different features on liquid level prediction fluctuations, as well as the changes in the impact of each feature on abnormal fluctuations over time.

[0046] In a fourth aspect, an embodiment of the present invention further provides a causal analysis system for abnormal fluctuations in the liquid level of a continuous casting crystallizer, the system comprising all the modules in the prediction system described above, and further comprising: a SHAP matrix calculation module, a characteristic SHAP value calculation module and a causal analysis module; wherein,

[0047] The SHAP matrix calculation module is used to calculate the input (t f )right The additive explanation of Shapley (SHAP) value and get a value that is the same as the input (t f ) SHAP matrix shap(t f )∈R w×dim , each element in the matrix represents input(t f ) in the corresponding position variable;

[0048] The feature SHAP value calculation module is used to accumulate the SHAP values ​​in the time dimension to obtain the SHAP value of each feature;

[0049] The abductive analysis module is used to transform all features into shapes at each moment. d (t f ) are aggregated into time series to visually display and analyze the impact of different features on liquid level prediction fluctuations, as well as the changes in the impact of each feature on abnormal fluctuations over time.

[0050] The solution of the embodiment of the present invention has the following beneficial effects:

[0051] The embodiments of the present invention provide a method for predicting abnormal fluctuations in the liquid level of a continuous casting crystallizer, a method for causal analysis, and a system, and a threshold-based method for calculating the abnormal liquid level fluctuation index. A time series classification model based on deep learning is constructed, and a neural network is used to learn time series data to predict the liquid level abnormality index. The SHAP analysis method is used to calculate the impact of various factors on the liquid level abnormality, extending the prediction of abnormal liquid level fluctuations to a predetermined time range, thereby improving the accuracy of the fluctuation abnormality prediction.

[0052] Of course, it is not necessary to achieve all of the advantages described above simultaneously in order to implement any product or method of the present invention. Description of the drawings:

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 This is a flow chart of a method for predicting abnormal fluctuations in the liquid level of a continuous casting mold according to an embodiment of the present invention;

[0055] Figure 2 It is a schematic diagram of the principle of the method for tracing the cause of abnormal fluctuation of liquid level in a continuous casting crystallizer according to an embodiment of the present invention. Specific implementation method:

[0056] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations. It should be noted that the embodiments of the present invention and the features in the embodiments can also be combined with each other in the absence of conflict.

[0057] It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures. In the description of the present invention, the terms "first," "second," "third," "fourth," etc. are used only to distinguish the description and are not to be understood as indicating or implying relative importance.

[0058] In order to solve the problem of the short time range of the existing prediction of liquid level fluctuation in the crystallizer, the embodiment of the present invention provides a method for predicting abnormal liquid level fluctuation in a continuous casting crystallizer, a causal analysis method and a system, and defines an abnormal liquid level fluctuation index to reflect the probability and severity of abnormal liquid level fluctuation in the future. A prediction model is constructed based on the defined abnormal liquid level fluctuation index, and the liquid level value itself is excluded from the input of the model. Variables other than the liquid level are used, including stopper opening, pulling speed, width adjustment, etc. Based on these input variables, the model predicts the abnormal liquid level index, thereby providing an early warning of the probability of occurrence and severity of abnormal liquid level, and more directly reflecting the possibility of abnormal liquid level fluctuation in the future and its potential severity. At the same time, the SHAP analysis method is used to calculate the influence of each input variable excluding the liquid level value itself on the liquid level anomaly, and the results are displayed in the form of images, so that it can be intuitively seen which factors are related to the liquid level fluctuation at a specific moment.

[0059] like Figure 1As shown, the method for predicting abnormal fluctuations in the liquid level of a continuous casting mold comprises the following steps:

[0060] Step S1, obtain the continuous casting parameters of historical time 1-T and the time to be predicted t f Continuous casting parameters; build a historical data set based on historical continuous casting parameters.

[0061] In this step, the continuous casting parameters include acquisition time, liquid level value, etc.

[0062] Step S2, set the set K = {k i ,i=1,2,…,N}, where k i The threshold value of the difference between the preset liquid level values ​​of the continuous casting crystallizer; if the difference between the liquid level value at time t and the liquid level value within T0 seconds before exceeds the threshold value k i , then it is determined that there is a threshold k at time t i Abnormal liquid level fluctuations The calculation formula is as follows:

[0063]

[0064] In formula (1), L(t) represents the crystallizer liquid level value at time t.

[0065] All moments 1-T in the historical data set are based on the threshold k i The fluctuation of For each threshold k i Calculate and obtain the time series of N abnormal fluctuation sets

[0066] The threshold k in this step reflects the severity of abnormal liquid level fluctuations. The larger k is, the more severe the abnormal liquid level fluctuations screened out will be.

[0067] Step S3, set the set Δ={δ j ,j=1,2,…,M}; where δ j Represents the preset time threshold; if δ after time t j Occurs within seconds based on threshold k i If the liquid level fluctuates abnormally, it is determined that δ j There is a threshold k in the time period i Abnormal liquid level fluctuations The calculation formula is as follows:

[0068]

[0069] δ of all time points 1-T in the historical data set j The fluctuations within the period form a time series, recorded as

[0070] In this step, the threshold δ j Reflects the time from the current moment to the occurrence of abnormal liquid level fluctuation, δ j The larger the value is, the wider the time range from the moment when the value is 1 to the occurrence of abnormal liquid level fluctuations.

[0071] Step S4: for each abnormal fluctuation set MLF ki , then based on the threshold δ j The calculation is performed to obtain the time series of M abnormal fluctuation sets, and the N×M×T liquid level abnormal fluctuation matrix MLF∈R is obtained for all abnormal fluctuation sets. N×M×T , expressed as:

[0072]

[0073] And the liquid level abnormality matrix at time t is expressed as MLF(t)∈R N×M .

[0074] Step S5: Based on the liquid level anomaly matrix MLF(t) at time t, a mold liquid level fluctuation anomaly prediction model is constructed. The input of the model is the historical continuous casting parameters sampled at time t, and the output is the liquid level anomaly matrix at time t.

[0075] In this step, the continuous casting parameters It is two-dimensional data, T1 represents the length of the time series, and D represents the feature dimension, including stopper opening, pulling speed, width adjustment, etc. The linear interpolation method is used to unify the sampling frequency between different dimensions to one data point per second.

[0076] Since the sampling frequency of data in the continuous casting process is high, one data per second, in order to enable the model to process data with a longer time span, the original data is processed by mean downsampling, and the sampling interval is t s , the length of the sequence after sampling is w, and the formula for mean downsampling is:

[0077]

[0078] In formula (4), mean represents the mean of the time series in the time dimension;

[0079] The input of the model at time t is input(t)∈R w×D is obtained by s ) to t are downsampled to obtain:

[0080]

[0081] In formula (5), MeanSampling() represents mean downsampling.

[0082] The output of the model at time t is output(t)∈R N×M , the expected output is MLF(t).

[0083] The volatility anomaly prediction model constructed in this step addresses the time series classification problem. Its input is time series data, and its output is a binary label. Binary cross-entropy is used as the model's loss function. The informer model is used as the time series classification model. It optimizes computational efficiency through the ProbSparse self-attention mechanism and self-attention distillation mechanism, while also employing a one-step decoding mechanism to improve prediction speed. It can efficiently process long sequences of data, delivering high accuracy and fast predictions. It is suitable for various time series prediction scenarios, such as stock prices, weather data, and industrial indicators.

[0084] Step S6: training the fluctuation anomaly prediction model based on the historical data set to obtain a mature fluctuation anomaly prediction model.

[0085] Step S7: The current time to be predicted t f Continuous casting parameters X(t f ) is input into the mature fluctuation anomaly prediction model, and the output is t f The liquid level abnormal matrix MLF(t f )’s estimated output(t f )∈R N×M ; Find the output of the model (t f ) is used as the average value of the predicted values ​​of all matrix elements in the liquid level anomaly index mlf(t f )

[0086] In this step, the average value of the liquid level abnormality index prediction value is calculated using the following formula:

[0087]

[0088] Step S8: Based on the predicted value of the liquid level abnormality index Assess the risk of abnormal level fluctuation after the predicted time; if If the value is greater than or equal to the preset threshold, it is determined that there is an abnormal fluctuation risk; if If the value is less than the preset threshold but greater than the value at the previous moment, it is judged as an increased risk; if If it is less than the preset threshold and less than the value at the previous moment, it is determined to be risk-free.

[0089] Based on the above abnormal fluctuation prediction method, the embodiment of the present invention also provides a method for tracing the cause of abnormal fluctuation of liquid level in continuous casting mold. Figure 2As shown, the fluctuation traceability analysis method includes not only the step of fluctuation anomaly prediction, but also the following steps:

[0090] Step S9, use the GradientExplainer to calculate input (t f )right The additive explanation of Shapley (SHAP) value and get a value that is the same as the input (t f ) SHAP matrix shap(t f )∈R w×dim , each element in the matrix represents input(t f ) in the corresponding position variable.

[0091] Step S10, in order to measure each feature (including stopper opening, pulling speed, width adjustment, etc.) The contribution of SHAP value is accumulated in the time dimension to obtain the SHAP value of each feature.

[0092] (7)

[0093] In formula (8), shape d (t f ) indicates that feature d is in t f SHAP value at the moment.

[0094] Step S11: The shape of all features at each moment d (t f ) are aggregated into time series to visually display and analyze the impact of different features on liquid level prediction fluctuations, as well as the changes in the impact of each feature on abnormal fluctuations over time.

[0095] Specifically, the higher the SHAP value of feature d, the greater the impact of feature d on the abnormal liquid level fluctuation; if the SHAP value of feature d increases or decreases over time, it means that the impact of feature d on the abnormal liquid level fluctuation increases or decreases as the continuous casting progresses.

[0096] Based on the same idea, an embodiment of the present invention further provides a continuous casting mold liquid level abnormal fluctuation prediction and continuous casting mold liquid level abnormal fluctuation tracing analysis system, including: a data acquisition module, a liquid level threshold k setting module, a k-based fluctuation sequence calculation module, a time threshold δ setting module, a δ-based fluctuation sequence calculation module, a three-dimensional fluctuation matrix construction module, a prediction model construction module, a prediction model training module, an abnormal index prediction module and a prediction result output module; wherein,

[0097] The data acquisition module is used to obtain the continuous casting parameters of the historical time 1-T and the time to be predicted t f Continuous casting parameters; construct historical data sets based on historical continuous casting parameters;

[0098] The liquid level threshold k setting module is used to set the set K={k i ,i=1,2,…,N}, where k i A threshold value representing a difference between preset liquid level values ​​of a continuous casting mold;

[0099] The k-based fluctuation sequence calculation module is used to calculate the value of the liquid level at time t when the difference between the liquid level value at time t and the liquid level value within the previous T0 seconds exceeds the threshold k. i , determine whether there is a threshold k at time t i Abnormal liquid level fluctuations And calculate all the time 1-T in the historical data set based on the threshold k i The fluctuation of For each threshold k i Calculate and obtain the time series of N abnormal fluctuation sets

[0100] The time threshold δ setting module is used to set the set Δ={δ j ,j=1,2,…,M}; where δ j Indicates the preset duration threshold;

[0101] The δ-based fluctuation sequence calculation module is used when δ after time t j Occurs within seconds based on threshold k i When the liquid level fluctuates abnormally, determine the δ j There is a threshold k in the time period i Abnormal liquid level fluctuations δ of all time points 1-T in the historical data set j The fluctuations within the period form a time series, recorded as

[0102] The three-dimensional fluctuation matrix construction module is used for each abnormal fluctuation set MLF ki , then based on the threshold δ j The calculation is performed to obtain the time series of M abnormal fluctuation sets, and the N×M×T liquid level abnormal fluctuation matrix MLF∈R is obtained for all abnormal fluctuation sets. N×M×T ;

[0103] The prediction model building module is used to build a mold liquid level fluctuation abnormality prediction model based on the liquid level abnormality matrix MLF(t) at time t. The input of the model is the historical continuous casting parameters sampled at time t, and the output is the liquid level abnormality matrix at time t;

[0104] The prediction model training module is used to train the fluctuation anomaly prediction model based on the historical data set to obtain a mature fluctuation anomaly prediction model;

[0105] The abnormal index prediction module is used to calculate the current predicted time t f Continuous casting parameters X(t f ) is input into the mature fluctuation anomaly prediction model, and the output is t f The liquid level abnormal matrix MLF(t f )’s estimated output(t f )∈R N×M ; Find the output of the model (t f ) is used as the average value of the predicted values ​​of all matrix elements in the liquid level anomaly index mlf(t f )

[0106] The prediction result output module is used to output the predicted value of the liquid level abnormality index. Assess the risk of abnormal level fluctuation after the predicted time; if If the value is greater than or equal to the preset threshold, it is determined that there is an abnormal fluctuation risk; if If the value is less than the preset threshold but greater than the value at the previous moment, it is judged as an increased risk; if If it is less than the preset threshold and less than the value at the previous moment, it is determined to be risk-free.

[0107] The embodiment of the present invention further provides a traceability analysis system for abnormal fluctuations in the liquid level of a continuous casting mold, the system comprising all the modules in the prediction system described above, and further comprising: a SHAP matrix calculation module, a characteristic SHAP value calculation module and a traceability analysis module; wherein,

[0108] The SHAP matrix calculation module is used to calculate the input (t f )right The additive explanation of Shapley (SHAP) value and get a value that is the same as the input (t f ) SHAP matrix shap(t f )∈R w×dim , each element in the matrix represents input(t f ) in the corresponding position variable;

[0109] The feature SHAP value calculation module is used to accumulate the SHAP values ​​in the time dimension to obtain the SHAP value of each feature;

[0110] The abductive analysis module is used to transform all features into shapes at each moment. d (t f ) are aggregated into time series to visually display and analyze the impact of different features on liquid level prediction fluctuations, as well as the changes in the impact of each feature on abnormal fluctuations over time.

[0111] In this embodiment, each module is implemented by a processor, and a memory is appropriately added when storage is required. The processor may be, but is not limited to, a microprocessor MPU, a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), other programmable logic devices, discrete gates, transistor logic devices, discrete hardware components, etc. The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk storage. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0112] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode.

[0113] It should also be noted that the system and the method described in this embodiment correspond to each other, and the description and limitation of the method are also applicable to the system and will not be repeated here.

[0114] It can be seen that the embodiments of the present invention provide a method for predicting abnormal fluctuations in the liquid level of a continuous casting crystallizer, a method for causal analysis, and a system, and a threshold-based method for calculating the abnormal liquid level fluctuation index. This method constructs a time series classification model based on deep learning, and uses a neural network to learn time series data to predict the liquid level abnormality index. The SHAP analysis method is used to calculate the impact of various factors on the liquid level abnormality, extending the prediction of abnormal liquid level fluctuations to a predetermined time range, thereby improving the accuracy of the fluctuation abnormality prediction.

[0115] The above description is only a preferred embodiment of the present invention and an explanation of the technical principles used. It is not intended to limit the scope of the invention to be protected, but merely represents a preferred embodiment of the present invention. Those skilled in the art should understand that the scope of the invention involved in the present invention is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the inventive concept. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative work shall fall within the scope of protection of the present invention.

Claims

1. A method for predicting abnormal fluctuations in liquid level in a continuous casting mold, characterized in that: The method comprises the following steps: Step S1, obtain the continuous casting parameters of historical time 1-T and the time to be predicted t f Continuous casting parameters; construct historical data sets based on historical continuous casting parameters; Step S2, set the set K = {k i ,i=1,2,…,N}, where k i The threshold value of the difference between the preset liquid level values ​​of the continuous casting crystallizer; if the difference between the liquid level value at time t and the liquid level value within T0 seconds before exceeds the threshold value k i , then it is determined that there is a threshold k at time t i Abnormal liquid level fluctuations All moments 1-T in the historical data set are based on the threshold k i The fluctuation of For each threshold k i Calculate and obtain the time series of N abnormal fluctuation sets Step S3, set the set Δ={δ j ,j=1,2,…,M}; where δ j Represents the preset time threshold; if δ after time t j Occurs within seconds based on threshold k i If the liquid level fluctuates abnormally, it is determined that δ j There is a threshold k in the time period i Abnormal liquid level fluctuations δ of all time points 1-T in the historical data set j The fluctuations within the period form a time series, recorded as Step S4: for each abnormal fluctuation set MLF ki , then based on the threshold δ j The calculation is performed to obtain the time series of M abnormal fluctuation sets, and the N×M×T liquid level abnormal fluctuation matrix MLF∈R is obtained for all abnormal fluctuation sets. N×M×T ; Step S5: Based on the liquid level anomaly matrix MLF(t) at time t, a mold liquid level fluctuation anomaly prediction model is constructed. The input of the model is the historical continuous casting parameters sampled at time t, and the output is the liquid level anomaly matrix at time t. Step S6: training the fluctuation anomaly prediction model based on the historical data set to obtain a mature fluctuation anomaly prediction model; Step S7: The current time to be predicted t f Continuous casting parameters X(t f ) is input into the mature fluctuation anomaly prediction model, and the output is t f The liquid level abnormal matrix MLF(t f )’s estimated output(t f )∈R N×M ; Find the output of the model (t f ) is used as the average value of the predicted values ​​of all matrix elements in the liquid level anomaly index mlf(t f ) Step S8: Based on the predicted value of the liquid level abnormality index Assess the risk of abnormal liquid level fluctuations occurring after the predicted time.

2. The method for predicting abnormal fluctuations in liquid level in a continuous casting mold according to claim 1, characterized in that: Abnormal liquid level fluctuation in step S2 The calculation formula is as follows: In formula (1), L(t) represents the crystallizer liquid level value at time t.

3. The method for predicting abnormal fluctuation of liquid level in continuous casting mold according to claim 2, characterized in that: Abnormal liquid level fluctuation in step S3 The calculation formula is as follows:

4. The method for predicting abnormal fluctuation of liquid level in continuous casting mold according to claim 1, characterized in that: In step S5, the input of the model is the historical continuous casting parameters sampled at time t. For two-dimensional data, T1 represents the length of the time series and D represents the feature dimension; The continuous casting parameters are subjected to mean downsampling processing, with a sampling interval of t s , the length of the sequence after sampling is w, and the formula for mean downsampling is: In formula (5), mean represents the mean of the time series in the time dimension; The input of the model at time t is input(t)∈R w×D is obtained by s ) to t are downsampled to obtain:

5. The method for predicting abnormal fluctuation of liquid level in continuous casting mold according to claim 4, characterized in that: D represents the characteristic dimension, including stopper opening, pulling speed, and width adjustment.

6. The method for predicting abnormal fluctuation of liquid level in continuous casting mold according to claim 1, characterized in that: In step S7, the average value of the liquid level abnormality index prediction value is calculated using the following formula: In formula (6), Output(t f ) matrix.

7. The method for predicting abnormal fluctuation of liquid level in continuous casting mold according to claim 5, characterized in that: Step S8 further includes: if If the value is greater than or equal to the preset threshold, it is determined that there is an abnormal fluctuation risk; if If the value is less than the preset threshold but greater than the value at the previous moment, it is judged as an increased risk; if If it is less than the preset threshold and less than the value at the previous moment, it is determined to be risk-free.

8. A continuous casting mold liquid level abnormal fluctuation prediction system, characterized in that: The system includes: a data acquisition module, a liquid level threshold k setting module, a k-based fluctuation sequence calculation module, a time threshold δ setting module, a δ-based fluctuation sequence calculation module, a three-dimensional fluctuation matrix construction module, a prediction model construction module, a prediction model training module, an abnormal index prediction module and a prediction result output module; wherein, The data acquisition module is used to obtain the continuous casting parameters at the historical time 1-T and the time to be predicted t f Continuous casting parameters; build historical data sets based on historical continuous casting parameters; The liquid level threshold k setting module is used to set the set K={k i ,i=1,2,…,N}, where k i A threshold value representing a difference between preset liquid level values ​​of a continuous casting mold; The k-based fluctuation sequence calculation module is used to calculate the value of the liquid level at time t when the difference between the liquid level value at time t and the liquid level value within the previous T0 seconds exceeds the threshold k. i , determine whether there is a threshold k at time t i Abnormal liquid level fluctuations And calculate all the moments 1-T in the historical data set based on the threshold k i The fluctuation of For each threshold k i Calculate and obtain the time series of N abnormal fluctuation sets The time threshold δ setting module is used to set the set Δ={δ j ,j=1,2,…,M}; where δ j Indicates the preset duration threshold; The δ-based fluctuation sequence calculation module is used when δ after time t j Occurs within seconds based on threshold k i When the liquid level fluctuates abnormally, determine the δ j There is a threshold k in the time period i Abnormal liquid level fluctuations δ of all time points 1-T in the historical data set j The fluctuations within the period form a time series, recorded as The three-dimensional fluctuation matrix construction module is used for each abnormal fluctuation set MLF ki , then based on the threshold δ j The calculation is performed to obtain the time series of M abnormal fluctuation sets, and the N×M×T liquid level abnormal fluctuation matrix MLF∈R is obtained for all abnormal fluctuation sets. N×M×T ; The prediction model building module is used to build a mold liquid level fluctuation abnormality prediction model based on the liquid level abnormality matrix MLF(t) at time t. The input of the model is the historical continuous casting parameters sampled at time t, and the output is the liquid level abnormality matrix at time t; The prediction model training module is used to train the fluctuation anomaly prediction model based on the historical data set to obtain a mature fluctuation anomaly prediction model; The abnormal index prediction module is used to calculate the current predicted time t f Continuous casting parameters X(t f ) is input into the mature fluctuation anomaly prediction model, and the output is t f The liquid level abnormal matrix MLF(t f )’s estimated output(t f )∈R N×M ; Find the output of the model (t f ) is used as the average value of the predicted values ​​of all matrix elements in the liquid level anomaly index mlf(t f ) The prediction result output module is used to output the predicted value of the liquid level abnormality index. Assess the risk of abnormal liquid level fluctuations occurring after the predicted time.

9. A method for tracing the cause of abnormal fluctuations in the liquid level of a continuous casting mold, characterized in that: The method comprises steps S1 to S8 according to any one of claims 1 to 7; and further comprises: Step S9, use the GradientExplainer to calculate input (t f )right The additive interpretation of the Shapley SHAP value and obtain a value that is consistent with the input(t f ) SHAP matrix shap(t f )∈R w×dim , each element in the matrix represents input(t f ) in the corresponding position variable; Step S10, accumulating the SHAP values ​​in the time dimension to obtain the SHAP value of each feature; Step S11: The shape of all features at each moment d (t f ) are aggregated into time series to visually display and analyze the impact of different features on liquid level prediction fluctuations, as well as the changes in the impact of each feature on abnormal fluctuations over time.

10. A continuous casting mold liquid level abnormal fluctuation tracing analysis system, characterized in that: The system includes all the modules as shown in claim 8, and further includes: a SHAP matrix calculation module, a characteristic SHAP value calculation module and a causal analysis module; wherein, The SHAP matrix calculation module is used to calculate the input (t f )right The additive explanation of Shapley (SHAP) value and get a value that is the same as the input (t f ) SHAP matrix shap(t f )∈R w×dim , each element in the matrix represents input(t f ) in the corresponding position variable; The feature SHAP value calculation module is used to accumulate the SHAP values ​​in the time dimension to obtain the SHAP value of each feature; The abductive analysis module is used to transform all features into shapes at each moment. d (t f ) are aggregated into time series to visually display and analyze the impact of different features on liquid level prediction fluctuations, as well as the changes in the impact of each feature on abnormal fluctuations over time.

Images