Flexible dynamic prediction method and system for discontinuous gas supply in converter steelmaking

Abstract

The invention discloses a flexible dynamic prediction method and system for discontinuous gas supply in converter steelmaking. CO concentration data and real-time total flow data of converter gas and start-stop signals of an air feeder are obtained; separating out independent gas supply flow data of each gas channel; performing staged modeling on the gas supply flow data of each gas channel to form three staged models, namely an ascending stage model, a stationary stage model and a descending stage model; flexible dynamic prediction is carried out based on a staged model; and inputting a test set, obtaining predicted values in a segmented manner, comparing the predicted values with actual values to obtain prediction errors, and analyzing prediction effects of different channel models. The defects that current converter steelmaking discontinuous gas supply flow data modeling is difficult, pressurization workshop detection data cannot depend on artificial experience, the gas recovery amount cannot be quantitatively predicted, subjectivity is high, the gas energy utilization efficiency of a pressurization workshop is affected, and safety risks exist can be overcome.

Description

Technical Field

[0001] The present application relates to the technical field of converter steelmaking gas control. Specifically, it relates to a flexible dynamic prediction method and system for discontinuous gas supply in converter steelmaking. Background Art

[0002] As the recipient of converter gas, a by-product of the converter, the pressurizing workshop needs to store, pressurize, match the calorific value, and transport the gas in the middle. In order to achieve the goals of green energy management and energy conservation and emission reduction, the modern gas pressurization production process needs to improve the gas utilization efficiency and reduce gas emissions on the premise of ensuring safe storage and transportation. This requires achieving a dynamic balance among the gas recovery amount or converter gas supply amount, gas storage amount, and gas transportation amount in the pressurizing workshop. This technological requirement poses a demand for the dynamic prediction of converter gas supply volume.

[0003] Converter steelmaking is currently the most common steelmaking method. Due to the characteristics of its production process flow, its production process is discontinuous. The converter steelmaking production process is complex, and the production environment is extremely harsh. The molten iron temperature is as high as 1600°C, and it is difficult to comprehensively monitor process parameters online in real time. At the same time, there are differences in production conditions for different converters and different heats of the same converter, which will also cause changes in the converter steelmaking process. All the above reasons make it difficult to accurately predict the gas flow rate generated by converter steelmaking, that is, the gas supply flow rate to the pressurizing workshop. At the same time, it is inconvenient for the pressurizing workshop to obtain real-time detection data of converter steelmaking production, and only a small amount of data information related to the gas produced by the converter can be collected, including the CO concentration of converter gas, the real-time total flow rate of multi-converter gas supply, the blower switch information, etc. This brings greater difficulties to the prediction of converter gas supply volume and cannot achieve continuous prediction and accurate prediction. Currently, in the pressurizing workshop, mainly based on whether the blower between the steelmaking workshop and the pressurizing workshop is turned on, the number of furnaces for gas supply is judged manually, and then according to experience, the gas cabinet recovery amount and gas output amount are controlled. This method cannot quantitatively predict the gas recovery amount and is highly subjective, which not only affects the gas energy utilization efficiency of the pressurizing workshop but also poses safety risks. Summary of the Invention

[0004] The technical problem to be solved by the present application is: to provide a flexible dynamic prediction method and system for discontinuous gas supply in converter steelmaking, so as to improve the current defects that it is difficult to model discontinuous gas supply flow data in converter steelmaking, the detection data in the pressurizing workshop cannot be utilized and depends on manual experience, it is impossible to quantitatively predict the gas recovery amount and is highly subjective, which not only affects the gas energy utilization efficiency of the pressurizing workshop but also poses safety risks.

[0005] To solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A discontinuous gas supply flexible dynamic prediction method for converter steelmaking, characterized by comprising the following steps:

[0007] Step S1: Obtain the CO concentration data, real-time total flow data of converter gas, and the start-stop signals of the air blowers.

[0008] Step S2: According to the start-stop signals of the air blowers and the CO concentration data, separate the individual gas supply flow data of each gas channel.

[0009] Step S3: Conduct phased modeling on the gas supply flow data of each gas channel to form three phased models: an ascending stage model, a stable stage model, and a descending stage model.

[0010] Step S4: Based on the phased models, perform flexible dynamic prediction, and adjust the three phased models in real time to adapt to the dynamic changes in the production process.

[0011] Step S5: Input the test set, obtain the predicted values in segments, compare the predicted values with the actual values to obtain the prediction error, and evaluate whether the prediction error reaches the set value; analyze the prediction effects of different channel models.

[0012] In the above technical solution, separating the individual gas supply flow data of each gas channel includes:

[0013] Obtain the segmented data of the total gas flow of multiple channels transported by a single steel plant, and remove the data during the time periods when all air blowers are not working.

[0014] According to the start-stop signals of the air blowers in each channel and the CO content data, extract the flow data samples during single-channel operation.

[0015] In the above technical solution, the phased modeling includes: dividing the gas supply flow data into three stages: ascending, stable, and descending; respectively using the linear fitting method to establish prediction models for each stage to form three phased models: an ascending stage model, a stable stage model, and a descending stage model.

[0016] In the above technical solution, the phased modeling includes: from the individual gas supply flow data of each gas channel in Step S2, respectively select a certain proportion of data samples and their corresponding time vectors from at least two gas supply channel data samples to form the training data sets of the two gas supply channels, and the remaining unselected data is used as the test data set.

[0017] In the above technical solution, the phased modeling includes dividing the training set: when the CO content increases at two consecutive moments and the norm of the CO content increase value exceeds the set rising threshold, it is considered that the air supply volume enters the rising stage; when the norm of the difference remains within the lower stable threshold for two consecutive sampling moments, it is determined that the air supply enters the stable air supply stage; and when the CO content decreases at two consecutive moments and the norm of the difference exceeds the preset decreasing threshold again, it indicates that the air supply enters the stage of decreasing air supply volume.

[0018] In the above technical solution, the phased modeling further includes: setting the rising threshold, stable threshold, and decreasing threshold, and determining that the air supply flow enters different stages by monitoring the change of the CO content.

[0019] In the above technical solution, the flexible dynamic prediction includes:

[0020] In the rising stage of the air supply volume, use the rising stage model for prediction;

[0021] In the stable air supply stage, use the stable stage model for prediction;

[0022] In the stage of decreasing air supply volume, use the decreasing stage model for prediction.

[0023] In the above technical solution, the flexible dynamic prediction further includes: dynamically adjusting the parameters of the prediction model according to the change of the CO content monitored in real time.

[0024] In the above technical solution, it also includes evaluating the prediction effect of the prediction model, and evaluating the prediction error by comparing the difference between the trapezoidal area of the prediction and the area enclosed by the actual data. That is, it is carried out by comparing the difference between the trapezoidal area of the prediction and the area enclosed by the actual data and the x-axis.

[0025] In the above technical solution, the flexible dynamic prediction method for discontinuous air supply in converter steelmaking is applicable to the air supply flow prediction of multiple converter steelmaking workshops.

[0026] In the above technical solution, the flexible dynamic prediction method for discontinuous air supply in converter steelmaking is set to perform periodic sliding time domain update according to the time lapse in the production process to ensure that the model accurately tracks the equipment and production status.

[0027] In the above technical solution, the flexible dynamic prediction method for discontinuous air supply in converter steelmaking can perform real-time prediction and correction of the gas supply volume based on the prediction model, so as to realize the dynamic flexible prediction of the air supply volume.

[0028] The present invention also provides a modeling and flexible dynamic prediction system for the discontinuous air supply flow in converter steelmaking based on the above method, which is characterized in that it includes:

[0029] A data acquisition module for obtaining the CO concentration data, real-time total flow data of converter gas, and the start-stop signal of the air blower;

[0030] A data processing module for separating the individual flow data of each channel according to the start-stop signal of the air blower and the CO concentration data;

[0031] A modeling module for performing phased modeling on the gas supply flow data of each channel, including the gas supply volume increase stage, stable gas supply stage, and gas supply volume decrease stage;

[0032] A prediction module for performing flexible dynamic prediction based on the phased model and adjusting the prediction model in real time to adapt to the dynamic changes in the production process.

[0033] Correspondingly, a computer-readable storage program can also be provided, which is used to implement the above method when the program is executed.

[0034] Compared with the prior art, the present application has at least the following beneficial effects:

[0035] Differences in data attributes:

[0036] At present, the modeling and prediction analysis of converter steelmaking carried out at home and abroad are mainly based on the real-time process data of converter steelmaking to predict the carbon content and temperature at the end of steelmaking, as well as the oxygen consumption during the steelmaking process. They are all predictions of continuous measurable quantities based on continuous real-time process data. However, the present invention is based on discontinuous flow data, which is more suitable for the actual production process.

[0037] Differences in whether to use the data of the pressurization workshop:

[0038] The prior art basically ignores the data of the pressurization workshop. From the perspective of the pressurization workshop, the present invention conducts research on the modeling and dynamic flexible prediction of the discontinuous gas flow generated by converter steelmaking based on a small amount of process data. It ensures safety and is closer to the actual production process.

[0039] Again, in terms of prediction accuracy:

[0040] The prior art usually relies on manual experience for the prediction of gas flow, with strong subjectivity and a lack of precise quantitative prediction methods. It is difficult to achieve continuous and accurate prediction of gas flow, resulting in a large deviation between the prediction result and the actual value.

[0041] The present application realizes the accurate prediction of gas flow by establishing a mathematical model based on CO concentration, real-time total flow, and the start-stop signal of the air blower.

[0042] The present application adopts phased modeling and flexible dynamic prediction methods, which can adjust the prediction model in real time and improve the accuracy of prediction.

[0043] Thirdly, in terms of the gas utilization efficiency:

[0044] Due to inaccurate prediction in the prior art, the utilization efficiency of gas energy is low, resulting in significant energy waste. It is difficult to achieve dynamic balance management of gas, affecting the usage efficiency of gas.

[0045] The present invention: By accurately predicting the gas flow rate, the dynamic balance management of gas is optimized, and the usage efficiency of gas is improved. The gas emission is reduced, the energy waste is decreased, and energy conservation and emission reduction are achieved.

[0046] In terms of safety:

[0047] Due to inaccurate prediction of gas flow rate in the prior art, there are safety risks, which may lead to accidents such as gas leakage or explosion. The lack of real-time monitoring and warning mechanism makes it difficult to detect and handle abnormal gas flow rate in a timely manner.

[0048] The present invention: By real-time monitoring and predicting the gas flow rate, the safety of the production process is improved. It can timely detect and handle abnormal gas flow rate, reducing the safety risks.

[0049] In terms of adaptability:

[0050] The prior art is difficult to adapt to the dynamic changes in the production process, and the prediction model lacks flexibility. It is difficult to make dynamic adjustments according to different production conditions, with poor adaptability.

[0051] The present invention: By adopting the fixed-period sliding time-domain update and flexible prediction method, it can track the production status in real time and timely adjust the prediction model. It has strong adaptability and can make dynamic adjustments according to different production conditions.

[0052] In terms of real-time performance:

[0053] The prior art: It is not convenient to collect data in the pressurization workshop, so manual estimation is used. It lacks real-time prediction and adjustment capabilities, and it is difficult to respond to the changes in the production process in a timely manner. The timeliness of the prediction results is poor and it is difficult to meet the actual on-site requirements.

[0054] The present invention: By adopting the real-time monitoring and prediction method, it can respond to the changes in the production process in a timely manner. The timeliness of the prediction results is strong and it can meet the actual on-site requirements.

[0055] In summary, by establishing an accurate mathematical model and adopting an advanced prediction method, the present invention significantly improves the accuracy of gas flow rate prediction, gas utilization efficiency, safety, adaptability and real-time performance, and has significant beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] To more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0057] Figure 1 It is a flowchart of the converter steelmaking discontinuous gas supply flexible dynamic prediction method implemented in the embodiments of the present application.

[0058] Figure 2 It is a schematic diagram of three steel plants in the embodiments of the present application for transporting converter gas to the pressurization workshop.

[0059] Figure 3 It is a flowchart of segmented modeling prediction in the embodiments of the present application.

[0060] Figure 4 It is a segmented data sample diagram of the total gas flow in the embodiments of the present application.

[0061] Figure 5 It is a segmented data sample of multi-channel CO content in the embodiments of the present application.

[0062] Figure 6 It is a single-channel flow data sample set in the embodiments of the present application.

[0063] Figure 7 It is a fitting result diagram of channel 1 in the embodiments of the present application.

[0064] Figure 8 It is a fitting result diagram of channel 2 in the embodiments of the present application.

[0065] Figure 9 It is the prediction result of channel 1 in the embodiments of the present application.

[0066] Figure 10 It is the prediction result of channel 2 in the embodiments of the present application. Detailed implementation manners

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Usually, the components of the embodiments of the present application described and shown in the accompanying drawings here can be arranged and designed in various different configurations.

[0068] Accordingly, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application.

[0069] The features and performance of the present application will be further described in detail below in conjunction with embodiments.

[0070] Embodiment 1

[0071] The flexible dynamic prediction method for non - continuous gas supply in converter steelmaking implemented according to the present invention, as Figure 1 shown, includes the following steps:

[0072] Step S1: Obtain data: Collect and obtain necessary input data from the system, including CO concentration, real - time total flow rate, and start - stop signals of the air blower.

[0073] Step S2: Separate single - channel flow data: According to the start - stop signals of the air blower and CO concentration data, separate the individual flow data of each channel from the total flow rate.

[0074] Step S3: Model in stages: Model the flow data of each channel in stages, including the stage of increasing gas supply volume, the stage of stable gas supply, and the stage of decreasing gas supply volume.

[0075] Step S4: Flexible dynamic prediction: Based on the staged model, perform real - time prediction, dynamically adjust the prediction model according to real - time data, and dynamically adjust the parameters of the prediction model according to the change in the CO content monitored in real time to adapt to the dynamic changes in the production process.

[0076] Step S5: Evaluate the prediction effect: Evaluate the accuracy of the prediction model by calculating the prediction error and comparing the predicted value with the actual value.

[0077] In the above - mentioned technical solution, separating the individual gas supply flow data of each gas channel includes:

[0078] Obtain the segmented data of the total flow rate of multi - channel gas transported by a single steel plant, and remove the data during the period when all air blowers are not working; preferably, the total gas supply flow rate data, the CO concentration data of multiple gas supply channels, and the start - stop signals of the air blowers of each channel can be obtained from the historical database.

[0079] According to the start - stop signals of the air blowers of each channel and the CO content data, extract the flow data samples during single - channel operation.

[0080] That is, from the real-time total gas volume data of multiple channels of each steel plant, extract the flow data samples from the start of gas supply to the end of gas supply, remove the data segments where all the blowers of multiple channels are not working, and record the time from the start to the end of these samples.

[0081] In the above technical solution, the phased modeling includes: dividing the gas supply flow data into three stages: rising, stable, and falling; respectively using the linear fitting method to establish prediction models for each stage, forming three phased models: the rising stage model, the stable stage model, and the falling stage model.

[0082] Adopt the method of fixed-period dynamic modeling to independently analyze the gas supply flow of each channel. Through linear piecewise fitting, establish a piecewise prediction model for the gas supply flow of each channel:

[0083] Taking Channel 1 and Channel 2 as examples to illustrate the linear piecewise fitting algorithm, the specific steps are as follows:

[0084] Step 1: Construction of the training dataset and test dataset of the prediction model

[0085] Select a part of the data from the data sample set as the training dataset for the model, and at the same time retain the remaining part of the data as the prediction test set. The training dataset is used for model learning and parameter estimation, while the prediction test set is used to evaluate the prediction performance of the model. Select 70% of the data samples and their corresponding time vectors from the data samples X1 and X2 of Channel 1 and Channel 2 respectively to form the training datasets of Channel 1 and Channel 2 T i represents the time series corresponding to the i-th segment of data samples, and its specific form is T i =[t1,t2,...,t n =[0,Δt,2Δt,…,(n - 1)Δt], where Δt represents the sampling interval time, and n is the total number of data points in this segment of data samples.

[0086] Step 2: Stage division of the training set data

[0087] In the process of modeling the non - continuous gas supply flow in converter steelmaking, in order to accurately reflect the dynamic change characteristics of the gas supply flow, the training dataset D is subdivided into three stages for independent fitting: the rising stage of gas supply volume D r , the stable gas supply stage D s and the falling stage of gas supply volume D d . The transition of the gas supply state is determined by monitoring the dynamic change of the CO content. Specifically, at two consecutive sampling moments, compare the magnitudes of the CO content detection values, calculate the norm of the CO content change amount, and compare it with a preset threshold.

[0088] When the CO content increases at two consecutive moments and the norm of the CO content increase value exceeds the set rising threshold, it is considered that the air supply volume enters the rising stage; when the norm of the difference remains within the lower stable threshold for two consecutive sampling moments, it is judged that the air supply enters the stable air supply stage; and when the CO content decreases at two consecutive moments and the norm of the difference exceeds the preset falling threshold again, it indicates that the air supply enters the stage of decreasing air supply volume. This phased processing strategy helps to improve the accuracy of the prediction model and makes it better adapt to the actual production process. The three-stage division process of the training set D is as follows:

[0089] Rising stage of air supply volume D r : When it is detected that the air supply flow starts to increase continuously, the air supply volume enters the rising stage. This stage represents the process of the gas flow gradually increasing after the air supply fan starts.

[0090] Set the rising threshold as τ r , if the CO content increases significantly at two consecutive sampling moments, that is, the CO content at the current j moment is detected higher than the previous moment and higher than and at the same time the norm of the difference between two adjacent CO content detections exceeds the rising threshold τ r When this happens, as shown in formula (1), the flow data of the rising stage of the air supply volume will be recorded starting from the moment when the air supply volume starts to rise significantly, that is, the j - 2 moment. The data of this stage will be used to fit the prediction model for the rising stage of the air supply flow.

[0091]

[0092] Stable air supply stage D s : After the rising stage, if the change in the CO content tends to be stable, the air supply volume enters the stable stage. In this stage, the air supply flow reaches and maintains at a relatively stable level.

[0093] Set the stable threshold as τ s . Monitor the change in the CO content. If the norm of the difference between the CO content at two adjacent moments is less than the set stable threshold τ s (τ s <τ r ) as shown in formula (2), it is judged that starting from the sampling moment when the air supply volume starts to become stable, that is, the j - 2 moment, the air supply volume enters the stable air supply stage.

[0094] The data of this stage will be used to fit the prediction model for the stable air supply stage.

[0095]

[0096] Falling stage of air supply volume Dd : When the CO content starts to decrease significantly, it is determined that the gas supply volume enters the decreasing stage. In this stage, the gas flow rate sent out by the steelmaking furnace gradually decreases.

[0097] Set the decreasing threshold τ d . Monitor the change of the CO content. When the CO content starts to decrease significantly, that is, the CO content at the current j-th moment is detected lower than the previous moment and lower than at the same time, the norm of the difference between two adjacent detections of the CO content exceeds the threshold τ d , as shown in formula (3), then it is determined that from the sampling moment when the gas supply volume starts to decrease significantly, that is, the j - 2 moment, it enters the gas supply volume decreasing stage. The data in this stage will be used to fit the prediction model for the gas supply volume decreasing stage.

[0098]

[0099] During the process of stage division, it should be noted especially that τ r τ s 、τ d needs to be adjusted according to different production conditions and sampling periods. This method of fitting in stages not only improves the accuracy of the model but also enhances the adaptability of the model to the discontinuity and dynamic changes in the actual production process.

[0100] According to the above method, the training sample data set D can be segmented into D r ={(T r , X r )}, D s ={(T s , X s )} and D d ={(T d , X d )}.

[0101] Step 3: Segmentally fit the gas supply volume prediction model

[0102] To further improve the prediction accuracy, the present application introduces the concept of fixed - period sliding - time - domain modeling. The core of this method lies in using the latest single - channel data for modeling, that is, every fixed period of time, the latest data sample set is re - selected to update the prediction model.

[0103] Linearly fit the data of the three stages in the latest period of the channel respectively.

[0104] 1) Fit D r ={(T r , X r )}

[0105] During the fitting process, first, for the training set D in the stage of increasing air delivery volume r ={(T r , X r )}, a linear model is constructed as shown in Equation (4)

[0106]

[0107] where is the actual flow value corresponding to the moment in the increasing stage , and β0, β1 are the model parameters to be solved. To find the optimal β0 and β1, a loss function is defined to measure the difference between the model prediction value and the actual value. In linear fitting, the commonly used loss function is the Mean Squared Error (MSE), as shown in Equation (5)

[0108]

[0109] To find β0 and β1 that minimize the loss function J r , the partial derivatives of J r with respect to β0 and β1 are taken respectively, and these partial derivatives are set to zero, thus obtaining a set of equations. Solving this set of equations can obtain β0 and β1. Specifically, the expression of β1 is as shown in Equation (6)

[0110]

[0111] where n is the number of samples in this stage. Once β1 is determined, β0 can be calculated through Equation (7)

[0112]

[0113] 2) D s ={(T s , X s )} is fitted

[0114] According to the training set D of the data in the stable air delivery stage s ={(T s , X s )}, the model in the stable stage is constructed as shown in Equation (8).

[0115]

[0116] That is, the ideal situation in the stable air delivery stage is that the air delivery flow is a fixed value. Where is the actual flow value in the stable stage. To solve the model parameter φ, the loss function is defined as the Mean Squared Error (MSE) between the actual value and the predicted value in the data set of the stable air delivery stage, as shown in Equation (9):

[0117]

[0118] Among them, n is the number of data points in the steady stage, is the actual gas supply flow rate value of the i-th data point, and φ is the model prediction value. To find the φ that minimizes the MSE, we can take the partial derivative of the MSE with respect to φ and set it to zero. The expression of φ is shown in formula (10)

[0119]

[0120] 3) D d ={(T d , X d )} fitting

[0121] For the training data set D d ={(T d , X d )} of the gas supply volume decline stage, fitting can be carried out to obtain the prediction model of the decline stage. The model construction method and parameter solution of the decline stage are the same as those of the rise stage, and the gas supply volume decline stage model is obtained by the same method as the rise stage as shown in formula (11).

[0122]

[0123] Among them is the actual flow rate value corresponding to the moment of the decline stage, and α0 and α1 are the parameters of the decline stage model.

[0124] This phased modeling method conforms to the actual situation on site, where the gas supply flow rate first gradually increases to a certain amount and remains basically stable, and then slowly decreases after the fan stops running. By separately constructing the models of the gas supply volume rise stage, steady gas supply stage, and gas supply volume decline stage, and solving the corresponding model parameters, the dynamic change process of the gas supply flow rate can be described more accurately.

[0125] In the above technical solution, the phased modeling further includes: setting a rise threshold, a steady threshold, and a decline threshold, and determining that the gas supply flow rate enters different stages by monitoring the change of the CO content.

[0126] That is, adding step 4: If at the current moment j, it is detected that the change of the CO content satisfies formula (3), it is determined that the steady gas supply stage ends and the gas supply volume decline stage begins. Starting from the j-th moment, the prediction model in formula (14) is used for real-time flow prediction. At the same time, the predicted value of the gas supply content at the j-1 moment is corrected according to formula (14).

[0127]

[0128] Among them, t i represents the current actual time, and i is the predicted value of the flow rate during the descending stage at time t.

[0129] In the above technical solution, the flexible dynamic prediction includes:

[0130] During the ascending stage of the gas supply volume, use the ascending stage model for prediction;

[0131] During the stable gas supply stage, use the stable stage model for prediction;

[0132] During the descending stage of the gas supply volume, use the descending stage model for prediction.

[0133] In the above technical solution, the flexible dynamic prediction further includes: dynamically adjusting the parameters of the prediction model according to the change of the CO content monitored in real time.

[0134] In the above technical solution, it also includes evaluating the prediction effect of the prediction model, and evaluating the prediction error by comparing the difference between the predicted trapezoidal area and the area enclosed by the actual data.

[0135] In the above technical solution, the flexible dynamic prediction method for discontinuous gas supply in converter steelmaking is applicable to the gas supply flow prediction of multiple converter steelmaking workshops.

[0136] In the above technical solution, the flexible dynamic prediction method for discontinuous gas supply in converter steelmaking is set to perform fixed-period sliding time-domain update according to the time progression in the production process to ensure that the model accurately tracks the equipment and production status.

[0137] In the above technical solution, the flexible dynamic prediction method for discontinuous gas supply in converter steelmaking can perform real-time prediction and correction of the gas supply volume based on the prediction model, so as to achieve dynamic flexible prediction of the gas supply volume.

[0138] Based on the above method, the present invention also provides a modeling and flexible dynamic prediction system for the discontinuous gas supply flow in converter steelmaking, which is characterized by including:

[0139] A data acquisition module for obtaining the CO concentration data of converter gas, the real-time total flow data, and the start-stop signal of the blower;

[0140] A data processing module for separating the individual flow data of each channel according to the start-stop signal of the blower and the CO concentration data;

[0141] A modeling module for performing phased modeling on the gas supply flow data of each channel, including the ascending stage of the gas supply volume, the stable gas supply stage, and the descending stage of the gas supply volume;

[0142] A prediction module for flexible dynamic prediction based on a phased model, and real-time adjustment of the prediction model to adapt to dynamic changes in the production process.

[0143] It is also possible to provide a computer-readable storage program that, when executed, is used to implement the above method.

[0144] Embodiment 2

[0145] This embodiment takes a certain iron and steel company as an example, and focuses on solving the problems of modeling and flexible dynamic prediction of the non-continuous gas supply flow in converter steelmaking.

[0146] This company has a total of three steel mills transporting converter gas to the pressurization workshop, which are simply referred to as No. 1 Steel Mill, No. 2 Steel Mill, and No. 3 Steel Mill. No. 1 Steel Mill has three 100-ton converters and three air blowers. The gas recovery time for a single converter is 7 to 8 minutes, and the gas recovery volume for each converter is between 13,500 and 14,000 cubic meters. No. 2 Steel Mill has two 140-ton converters and three air blowers. The gas recovery time for a single converter is 10 to 11 minutes, and the gas recovery volume for each converter is between 17,500 and 18,000 cubic meters. No. 3 Steel Mill has two 140-ton converters and three air blowers. The gas recovery volume for each converter is between 17,500 and 18,000 cubic meters. Due to the relatively large oxygen lance of the converter, the gas recovery time for a single converter is relatively short, which is 9 to 10 minutes. At the same time, at most seven air blowers are working, that is, three air blowers in No. 1 Steel Mill are working, and two air blowers in No. 2 Steel Mill and No. 3 Steel Mill are working respectively. Figure 2 It is a schematic diagram of transporting converter gas from three steel mills to the pressurization workshop.

[0147] From the perspective of the pressurization workshop, according to the CO concentration of the converter gas, the real-time total gas supply flow of each steel mill, and the on-off information of each air blower, a method for modeling and flexible dynamic prediction of the single-furnace gas supply volume of the converter gas is proposed, and an analysis and explanation are carried out based on the above steel mill examples.

[0148] I. Modeling of the gas transportation volume of each gas supply channel

[0149] During the converter steelmaking process, the amount of converter gas generated is directly affected by the converter tonnage and the actual production conditions, and there are significant differences in the air supply characteristics of different channels. Therefore, in this application, combined with the actual characteristics of converter steelmaking, the gas supply flow of each channel is independently analyzed, modeled, and predicted. Since the prediction methods for the gas supply flow of different channels in each steel mill are the same, this application takes three channels in No. 1 Steel Mill as an example to elaborate on the modeling and flexible dynamic prediction scheme in detail.

[0150] First, obtain the total gas supply flow data of No. 1 Steel Mill, the CO concentration data of the three gas supply channels, and the start-stop signals of the air blowers in each channel from the historical database.

[0151] Since it is impossible to obtain the gas supply flow data of each air blower, only the total real-time gas supply flow of the steel plant, that is, the total gas supply flow of the three channels, can be obtained. First, it is necessary to preprocess the gas supply flow data. Subsequently, for the gas supply flow data of each channel, linear piecewise fitting is adopted to establish a gas supply flow prediction model for each channel. After the model is established, the prediction effects of different channel models are analyzed. The specific algorithm process is as Figure 3 shown.

[0152] 1. Obtain the gas supply flow data samples of a single channel

[0153] The production process of the steelmaking plant is batch production by furnace, resulting in significant discontinuous characteristics in the generation of converter gas. In order to accurately analyze the change law of the gas production flow and construct an effective prediction model, it is necessary to extract the data samples within the complete gas supply cycle from the multi-channel total gas volume data. This process includes identifying the complete time period from the start to the stop of each air blower to ensure that the extracted data samples can comprehensively reflect the dynamic changes of the gas supply flow.

[0154] The pressurization workshop can only obtain the total converter gas volume transported by each steel plant, that is, the total gas volume of the three channels. However, in order to ensure the production safety and production efficiency of the pressurization workshop, it is necessary to accurately predict the gas supply flow of each channel. Therefore, first, it is necessary to separate the individual flow data of each channel from the total steel plant transport flow data. The following are the specific steps to split the individual flow data of the three channels:

[0155] Step 1: Obtain the segmented data of the multi-channel total gas volume transported by a single steel plant

[0156] Considering the discontinuous characteristics of the gas production in the steelmaking plant, it is necessary to preprocess the original data. According to the switch information of each air blower, the data in the time period when all three air blowers are not working is removed.

[0157] From the multi-channel real-time total gas volume data of each steel plant, extract the flow data samples X from the start of gas supply to the end of gas supply, remove the data segments where all three air blowers are not working, and record the time T from the start to the end of these samples. As Figure 3 shown, a total of 120 segments of flow data samples are extracted, that is, X = {X1, X2,... X 120}; the corresponding time vector is recorded as T = {T1, T2,... T 120}. Figure 4 This is the segmented data of the total gas production flow of a steelmaking plant.

[0158] Step 2: Obtain the single-channel flow data

[0159] In the production monitoring system of a steelmaking plant, the start-stop signals of the air blowers in multiple channels and the CO content data in the corresponding channels are usually recorded. These data are crucial for identifying the on-off states of the air blowers in the channels. According to the start-stop signals of the blowers in three given channels and the CO content data of each channel, a graph showing the change in CO content in the three channels during the working time T corresponding to the sample X is drawn, as Figure 5 shown. From the graph of the CO content change, samples of the individual channels are identified, and then the flow data samples when only the air blower in one channel is working are extracted from the sample X, as Figure 6 shown.

[0160] The data samples when the three channels are operating separately are represented by X1, X2, and X3 respectively. Specifically, the data samples of section 110 of channel 1 are extracted, X1 = {X4, X 19 , X 33 , X 40 , X 42 , X 51 , X 53 , X 55 , X 57 , X 115}; A total of 14 samples are extracted from channel 2, and its specific form is X2 = {X 23 , X 29 , X 34 , X 36 , X 44 , X 75 , X 79 , X 96 , X 107 , X 109 , X 111 , X 114 , X 117 , X 119}; 2 samples are extracted from channel 3, X3 = {X 86 , X 88}.

[0161] 2. Single-channel gas supply flow modeling

[0162] To adapt to the actual working conditions during the process of recovering converter gas, this application adopts the method of fixed-period dynamic modeling to independently analyze the gas supply flow of each channel. Through linear piecewise fitting, a piecewise prediction model of the gas supply flow for each channel is established, so as to achieve accurate prediction of the gas supply flow of each channel.

[0163] Perform fitting analysis on the data sample sets X1, X2, and X3 of the three channels. During the air supply process, the change in the air supply flow rate first increases from 0, then approaches a steady state, and finally decreases to 0. According to the characteristics of the air supply process, this paper divides the data samples into three stages: rising, steady, and falling. Then, the linear fitting method is used to fit the data samples of these three stages respectively, making the fitting result curve approximately trapezoidal, and this fitting curve conforms to the actual gas production process. Since the data samples of the three channels are few, this paper takes channels 1 and 2 as examples to illustrate the fitting algorithm.

[0164] The specific steps of the fitting algorithm are the same as those in Embodiment 1 and will not be elaborated here.

[0165] This phased modeling method conforms to the actual situation on site, where the air supply flow rate first gradually increases to a certain amount and remains basically stable, and then slowly decreases after the fan stops running. By separately constructing models for the rising stage of the air supply volume, the steady air supply stage, and the falling stage of the air supply volume, and solving the corresponding model parameters, the dynamic change process of the air supply flow rate can be described more accurately. This trapezoidal model intuitively reflects the change trend of the air supply flow rate, as Figure 7 and Figure 8 shown.

[0166] 3. Real-time flexible prediction of the air supply flow rate in a single channel by stages

[0167] To meet the actual needs on the steelmaking plant site, this paper uses the flexible prediction method to predict the air supply flow rate. Flexible prediction is a prediction method for dealing with uncertainty and change, and is applicable to scenarios that need to flexibly respond to various possible situations. It provides an effective prediction range rather than an exact prediction value by considering multiple possible future scenarios to help decision-makers better cope with future uncertainties.

[0168] The prediction model obtained based on piecewise fitting shows that the slope of the air supply flow rate model of each channel is relatively fixed during the rising stage in the first stage, then enters the steady stage, and also has a fixed slope in the final falling stage. In practical applications, the difference in the prediction of the air supply flow rate in a single channel is mainly reflected in the length of time to maintain the steady state, and this length of time can be determined according to the size and change of the CO content in the channel. The specific steps of the real-time flow rate prediction by stages are as follows.

[0169] Step 1: When the air supply fan start signal is received and the CO content is greater than zero, start real-time analysis and comparison of the change in the CO content.

[0170] Step 2: If the change in the CO content satisfies formula (1), it is determined that the rising stage of the air supply volume begins. During this stage, the real-time prediction of the air supply volume is carried out using the prediction model in formula (12) until the change in the CO content satisfies formula (2).

[0171]

[0172] Among them, t i represents the current actual time, is the predicted value of the flow rate during the descending stage at time t i of the moment.

[0173] Step 3: If at the current moment j, it is detected that the change in CO content satisfies formula (2), then it is determined that the ascending stage of the gas supply volume ends and the stable gas supply stage begins. Starting from the moment j, the prediction model in formula (13) is used for real-time flow rate prediction. At the same time, the predicted value of the gas supply content at the moment j - 1 is corrected according to formula (13).

[0174]

[0175] Among them, is the predicted value of the flow rate during the stable stage.

[0176] Step 4: If at the current moment j, it is detected that the change in CO content satisfies formula (3), then it is determined that the stable gas supply stage ends and the descending stage of the gas supply volume begins. Starting from the moment j, the prediction model in formula (14) is used for real-time flow rate prediction. At the same time, the predicted value of the gas supply content at the moment j - 1 is corrected according to formula (14).

[0177]

[0178] Among them, t i represents the current actual time, is i the predicted value of the flow rate during the descending stage at time t

[0179] 4. Evaluation of the prediction effect of the segmented prediction model

[0180] Step 1: Construction of the prediction test set

[0181] The remaining 30% of the samples of Channel 1 and Channel 2 are respectively combined to form a test set to test the prediction effect of the model.

[0182] Step 2: Segmented prediction

[0183] Based on the test samples in the test set, the gas supply volume is predicted according to the detailed steps mentioned in the subsequent "Real-time flexible prediction of single-channel gas supply flow rate in stages".

[0184] Step 3: Evaluation of the prediction effect

[0185] Flexible prediction allows for a certain range of error to adapt to on-site uncertainties. The prediction effect is usually evaluated by calculating the difference between the actual value and the predicted value. In this application, the prediction effect is evaluated by comparing the difference between the predicted trapezoidal area and the area enclosed by the actual data and the x-axis, as shown in formulas (15)-(17).

[0186] S a = ∫f(t)dt (15)

[0187] S p = ∫f r (t)dt + ∫f s (t)dt + ∫f d (t)dt (16)

[0188] e = |S a - S p | (17)

[0189] Where S a is the area enclosed by the test set data sample and the x-axis as shown in formula (9), S p is the area of the predicted trapezoid, and e is the error.

[0190] The prediction results of the gas supply flow rates for Channel 1 and Channel 2 are respectively shown in Figure 9 (a)-(c) of Figure 10 and (a)-(d) of

[0191] There is a certain error between the prediction results and the actual values. The specific error values are shown in Table 1 and Table 2.

[0192]

[0193] Table 1 Prediction Error of Channel 1

[0194]

[0195] In summary, this application has modeled and dynamically flexibly predicted the single-furnace discontinuous gas supply flow rate in converter steelmaking. The established segmented prediction model conforms to the converter steelmaking gas production process. This model can perform fixed-period sliding time-domain updates as time progresses during the production process to ensure that the model accurately tracks the equipment and production status. At the same time, based on the prediction model, the gas supply volume can be predicted and corrected in real time, thus realizing the dynamic flexible prediction of the gas supply volume.

[0196] The embodiments described above are some, but not all, of the embodiments of the present application. The detailed description of the embodiments of the present application is not intended to limit the scope of the present application claimed, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application.

Claims

1. A flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking, characterized in that: The following steps are involved: Step S1: Obtaining the CO concentration data of the converter gas, the real-time total flow data, and the start and stop signals of the blower; Step S2: Separate the individual gas flow data of each gas channel according to the start / stop signal of the blower and the CO concentration data; Step S3: Modeling the gas flow data of each gas channel in stages to form three stage models: a rising stage model, a stable stage model and a falling stage model; Step S4: Perform flexible dynamic prediction based on the staged model, and adjust the three staged models in real time to adapt to dynamic changes in the production process; Step S5: Input the test set, obtain the predicted values ​​in segments, compare the predicted values ​​with the actual values ​​to obtain the prediction errors, and evaluate whether the prediction errors reach the set values; analyze the prediction effects of different channel models.

2. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1 is characterized in that: The individual gas flow data of each gas channel is separated into: Obtain segmented data of the total flow of multi-channel gas delivered by a single steel plant, removing data from the time period when all fans are not working; According to the start and stop signals of the blowers in each channel and the CO content data, the flow data samples during single-channel operation are extracted.

3. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1, characterized in that: The staged modeling includes: dividing the air flow data into three stages: rising, stable and falling; using a linear fitting method to establish a prediction model for each stage, forming three staged models: an rising stage model, a stable stage model and a falling stage model.

4. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1, characterized in that: The staged modeling also includes: setting an ascending threshold, a steady threshold and a descending threshold, and determining that the air supply flow enters different stages by monitoring changes in the CO content.

5. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1, characterized in that: The flexible dynamic prediction includes: In the rising stage of air delivery, the rising stage model is used for prediction; In the steady gas delivery phase, the steady phase model is used for prediction; During the gas flow reduction phase, the reduction phase model is used for prediction.

6. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1, characterized in that: The flexible dynamic prediction also includes: dynamically adjusting the parameters of the prediction model according to the real-time monitored CO content changes.

7. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1, characterized in that: It also includes an evaluation of the prediction effect of the prediction model, and the prediction error is evaluated by comparing the difference between the predicted trapezoidal area and the area enclosed by the actual data.

8. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1, characterized in that: The flexible dynamic prediction method for non-continuous gas supply in converter steelmaking is suitable for predicting the gas supply flow rate of multiple converter steelmaking workshops.

9. The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking according to claim 1, characterized in that: The flexible dynamic prediction method for discontinuous gas delivery in converter steelmaking is configured to perform fixed-period sliding time domain updates according to the passage of time in the production process.

10. A modeling and flexible dynamic prediction system for discontinuous air flow in converter steelmaking, characterized in that: include: Data acquisition module, used to obtain the CO concentration data of converter gas, real-time total flow data and start and stop signals of the blower; The data processing module is used to separate the individual flow data of each channel according to the start and stop signals of the blower and the CO concentration data; A modeling module is used to model the air flow data of each channel in stages, including an air flow increase stage, a stable air flow stage, and an air flow decrease stage; The prediction module is used to perform flexible dynamic prediction based on the staged model and adjust the prediction model in real time to adapt to dynamic changes in the production process.

Images