Method for optimizing heating process of annealing furnace

Abstract

The invention relates to an annealing furnace heating process optimization method, and belongs to the field of annealing furnace optimization control. The method comprises the following steps: firstly, collecting and processing related information of a current volume, two previous volumes and two latter volumes to form a plan small group, and preliminarily sorting the two latter volumes according to plan information and transition information of the two previous volumes; establishing a heating process system prediction hybrid model by adopting case reasoning and based on a recurrent neural network; the heating process systems of the current roll and the next roll are obtained according to the heating process system prediction hybrid model; according to the obtained two-coil heating process system, a proper transition strategy is selected, and steel coil production transition management is conducted; and finally, issuing a set value according to the calculated value, monitoring field measurement in real time, and correcting in time. And when the strip steel is discharged from the furnace, digitally collecting the steel coils, and adding or updating the case process library and the historical library in time. The transition strategy is combined to achieve transition management of the strip steel, transition time is greatly shortened, and the length of the strip steel which does not meet the temperature requirement is reduced.

Description

Technical Field

[0001] The invention belongs to the field of annealing furnace optimization control and relates to an annealing furnace heating process optimization method. Background Art

[0002] Annealing furnace is an important equipment in cold rolling coating production line. During the production process, a large amount of fuel is consumed to heat the strip to the target temperature range, and then the coating operation is carried out to finally produce suitable coated products. Annealing furnace has the characteristics of nonlinearity, large hysteresis, high inertia, etc. Annealing furnace control is complex and has multiple influencing factors. Usually the influencing factors can be divided into fixed factors and random factors. The fixed factors include strip specifications (strip width, strip thickness, heating target temperature), strip speed, and fuel input; random factors include temperature measurement fluctuations, speed fluctuations, fuel composition and pressure fluctuations, air temperature and flow fluctuations, furnace performance fluctuations, etc. In general, random factors are unobservable and fixed factors are observable. When optimizing the annealing furnace process system, it is necessary to determine the influence law of fixed influencing factors through steady-state data mining knowledge and existing knowledge. Data is formed under the influence of fixed factors and random factors. Therefore, when analyzing, data processing is first performed, such as using window sliding technology to process data, and then removing the influence of random factors, and finally obtaining fixed influencing factor data, and obtaining the influence law of fixed factors by analyzing related data. Because the heating process is affected by many factors, and the influence of each factor is complex and nonlinear, it is particularly important to optimize the heating process system. By optimizing the heating process system, the transition management time can be reduced, and then the length of the strip that does not meet the target temperature can be reduced, laying the foundation for improving production efficiency and production quality.

[0003] At present, there are two main methods for controlling the heating process of the annealing furnace. One is to operate based on the experience of the operator, but it has the problems of operator experience and operating habits, which leads to instability of the production products, that is, there are individual differences and are greatly affected by random factors. In addition, the use of model control method to establish secondary control is also one of the common methods, that is, to establish system simulation calculation based on mathematical and physical models, but in the actual application process, it is found that this method cannot adapt well to the influence of random factors, and it strongly depends on measurement accuracy and representativeness. If the measurement inaccuracy and the measurement position are not representative enough, it is easy to cause input errors in the mathematical and physical model, and it is easy to cause contradictory results through mathematical and physical iterations, and the credibility of the results is reduced. In addition, the simulation calculation time is relatively long in general, which is difficult to meet the requirements of online real-time optimization process.

[0004] In recent years, with the continuous development of computer hardware and deep learning technology, recurrent neural networks have been applied in many fields. One variant, the long short-term memory network (LSTM) model, can better capture and memorize dependencies in long sequences and process sequence data better. It can predict unknown points by learning from known data, and it has good processing capabilities for nonlinear data. In view of this, the present invention combines recurrent neural networks and case reasoning methods to establish a hybrid model for predicting heating process regimes, thereby more accurately predicting heating process regimes, and at the same time combining effective transition strategies to reduce transition time, reduce the length of strip steel that does not meet temperature requirements, and improve production efficiency and product quality. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide an annealing furnace heating process optimization method. The method establishes a heating process system prediction hybrid model combining a case reasoning model and a recurrent neural network model, and at the same time, combines existing knowledge to fully analyze the influence of fixed factors and random factors, determines the weight coefficient, and on this basis, establishes a prediction model and a reverse calculation model of the heating process system. The model prediction accuracy is higher. On the basis of more accurate prediction of the heating process system of the front and rear rolls, combined with the strip steel information of the front and rear rolls, a reasonable transition strategy is selected, thereby reducing the transition management time and reducing the length of the strip steel that does not meet the target temperature, and finally achieving improved production efficiency and production quality.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] An annealing furnace heating process optimization method comprises the following steps:

[0008] S1: First, read the planning information of the first two rolls, the current roll and the next two rolls to form a small data group of online production planning. According to the target temperature of each roll, product specification transition information, whether over-temperature is allowed, etc., and combined with the transition situation of the first two rolls and the current roll, form a preliminary sorting of the next two rolls.

[0009] S2: A hybrid model for predicting the heating process system is established by combining the case reasoning method and the method based on the recurrent neural network model. The case reasoning method is trained based on the heating process system case library, and the recurrent neural network model is iteratively trained based on the historical database data and the model is saved. The two methods are weighted combined to obtain the hybrid model for predicting the heating process system, and the hybrid model for predicting the heating process system is used to predict the heating process system of the current volume and the next volume after the preliminary sorting of S1.

[0010] S3: According to S2, the corresponding heating process system range under steady state is obtained, and the relevant planning information of S1 is obtained, and the appropriate transition strategy is selected to carry out transition management of steel coil production and realize process optimization of the transition management process.

[0011] S4: The calculated value of step S3 is issued, and the strip temperature is monitored in real time. The PID method is used to monitor the strip temperature, dynamically analyze the impact of changes in influencing factors, and correct the set value in time.

[0012] S5: When the strip steel is out of the furnace, the steel coils are digitally collected, and the case process library and history library are added or updated in time for the training of the hybrid model for heating process system prediction. In the process of collecting relevant parameters, abnormal data and missing data are processed according to the mathematical and physical model knowledge and the previous and next data, and the data is normalized.

[0013] Furthermore, S1 includes the following steps:

[0014] S101: Read the schedule information of the current roll and the two rolls before and after from the first-level system and the third-level system of the annealing furnace. The calculation table information includes strip width, strip thickness, strip specification, upper and lower limits of target temperature, speed requirements, over-temperature requirements, performance requirements, etc.

[0015] S102: Preliminarily calculate the required energy of each roll of strip steel according to the schedule information, sort the last two rolls of strip steel according to the transition strategy, relevant plan information and the transition relationship between the previous and next two rolls and the current roll, and preliminarily determine whether the position relationship between the next two rolls needs to be adjusted.

[0016] Further, the establishment of the heating process system prediction hybrid model described in S2 is characterized by comprising the following steps:

[0017] S201: First, match the case process library according to the relevant plan information of the current roll and the next roll, select the case with the best similarity, and then extrapolate based on the most similar case to obtain the heating process system of the two strips; when matching the cases, it is necessary to calculate the similarity, and calculate the weight of the case reasoning model based on the similarity. When the similarity is high, the case reasoning model is assigned a larger weight. When the similarity is lower than the similarity threshold, the case reasoning model weight is reduced; when establishing the case reasoning model, it is necessary to analyze the weights of various influencing factors. According to production experience and relevant knowledge, the influencing factors are preliminarily determined to include strip speed, strip width, strip thickness, target temperature, and fuel input. The weights of various influencing factors are mainly determined by hierarchical analysis method and single factor analysis method, with hierarchical analysis method as the main method and single factor analysis method as the auxiliary method.

[0018] S202: A data source for training a recurrent neural network is formed in a historical database according to the production sequence. When collecting data, real-time data in seconds is first collected into a historical database at the minute level, and then the historical database at the minute level is collected into a historical database at the strip level (i.e., each strip has only one data). In the process of collection, missing data and abnormal data are handled at the same time. The abnormal data and missing data are corrected and filled according to mathematical knowledge, experience, and previous and subsequent data. At the same time, the numerical mean, median, maximum and minimum values, and variance need to be retained when collecting data. When training, every 10 rolls of strip steel are merged into one data source, and the cross-validation method is used for data division. The recurrent neural network method (such as the STA-LSTM method) is used for training and prediction of the data, and the model is saved. The model is updated every one to three days, that is, new data is added to the model training.

[0019] S203: The heating process conditions obtained in S201 and S202 are combined in a weighted manner to finally obtain the heating process condition of the strip steel.

[0020] Furthermore, the model transition strategy described in S3 is characterized by comprising the following steps:

[0021] S301: If the minimum allowable load of the current coil is greater than the maximum allowable load of the next coil, first determine whether the next coil can be overheated. If the next coil can be overheated, it will run at the current coil speed and minimum allowable load during transition management; if the next coil is not allowed to be overheated, the S2 heating process system prediction hybrid model will be used to reversely calculate the maximum speed and minimum speed range of the next coil under the minimum allowable load of the current coil. When the strip weld comes out of the preheating section of the annealing furnace, start to quickly adjust the speed to the calculated speed upper limit, and make the next coil run at the new speed. When the strip weld comes out of the heating section of the annealing furnace, adjust the load and speed at the same time until the next coil reaches the planned set value.

[0022] S302: If the minimum allowable load of the current coil is less than the maximum allowable load of the next coil, but the maximum allowable load of the current coil is greater than the minimum allowable load of the next coil, the load in the overlapping area is selected for transition management and the load is set without adjusting the speed. The temperature setting value of the strip is calculated by reverse calculation through the S2 heating process system prediction hybrid model.

[0023] S303: If the maximum allowable load of the current coil is less than the minimum allowable load of the next coil, the transition strategy needs to be set in combination with whether the current coil can be overheated. If the current coil can be overheated, the transition state runs at the current coil speed and the minimum load of the next coil; if overheating is not allowed, the S2 heating process system prediction hybrid model is used to reversely calculate the maximum speed and minimum speed range of the next coil corresponding to the maximum load of the current coil. When the strip weld comes out of the preheating section of the annealing furnace, the speed is quickly adjusted to the calculated lower limit speed, and the next coil is operated at the new speed. When the strip weld exits the heating section of the annealing furnace, the load and speed are adjusted at the same time until the next coil reaches the planned set value.

[0024] Further, S5 comprises the following steps:

[0025] S501: When the tail weld of the current coil of strip steel comes out of the heating section of the annealing furnace, the data is aggregated by average, variance, median, mode, maximum, minimum and variance, and the influence of the transition management time before and after is considered. Some data needs to be removed, and the abnormal data and missing data are corrected and filled according to mathematical knowledge, experience and before and after data.

[0026] S502: The real-time collected data is aggregated into a historical database according to two methods for recurrent neural network model training. The data is aggregated by average, variance, median, mode, maximum and minimum values, and variance. Through the interval segmentation method, every 10 groups of strip data are used as a data source.

[0027] This method establishes a hybrid model for predicting the heating process system that combines the case reasoning model and the recurrent neural network model. At the same time, it combines the existing knowledge to fully analyze the influence of fixed factors and random factors. On this basis, a prediction model and a back-calculation model for the heating process system are established. The model has higher prediction accuracy. On the basis of more accurate prediction of the heating process system of the front and rear rolls, combined with the strip information of the front and rear rolls, a reasonable transition strategy is selected, thereby reducing the transition management time and reducing the length of the strip that does not meet the target temperature, ultimately achieving improved production efficiency and quality.

[0028] The beneficial effects of the present invention are: the method establishes a hybrid model for predicting the heating process system that combines the case reasoning model and the recurrent neural network model, and at the same time, combines the existing knowledge to fully analyze the influence of fixed factors and random factors, and on this basis, establishes a prediction model and a reverse calculation model for the heating process system. The model has higher prediction accuracy and strong adaptability, providing strong support for the optimization of the annealing furnace. At the same time, combined with the strip information of the front and rear rolls, a reasonable transition strategy is selected to achieve a reduction in transition management time and reduce the length of the strip that does not meet the target temperature, and ultimately achieve improved production efficiency and production quality. The method of the present invention uses a hybrid model for predicting the heating process system to achieve higher prediction accuracy and faster prediction. At the same time, combined with the transition situation of the annealing furnace, a reasonable transition management strategy library is established to achieve a reduction in transition management time and reduce the length that does not meet the target temperature. The method has strong operability and adaptability and has great prospects for promotion.

[0029] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0031] Figure 1 It is the overall flow chart of the present invention;

[0032] Figure 2 Adjust speed and load flow charts for the same specification. DETAILED DESCRIPTION

[0033] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0034] Among them, the attached drawings are only for illustrative purposes, showing only schematic diagrams rather than physical diagrams, and should not be construed as a limitation on the present invention; in order to better illustrate the embodiments of the present invention, some components in the attached drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the attached drawings may be omitted.

[0035] In the attached drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the attached drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the attached drawings are only for illustrative purposes and should not be construed as a limitation on the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0036] In this example, the annealing furnace heating process is optimized as Figure 1 shown. In this example, a certain cold-rolled strip coating line of a steel enterprise is selected, and the specific implementation process is as follows:

[0037] S1: Read the plan information of the first two coils, the current coil, and the last two coils from the first-level system and the third-level system of the annealing furnace to form a small production plan data group, where the plan information includes the time to establish the production sequence, strip thickness, strip width, strip specifications, strip speed, upper and lower limits of the strip target temperature, whether over-temperature is allowed, etc. Calculate the required heat of each strip according to the relevant information and strip physical property parameters, and record them as q1-q5 respectively. Calculate the following values according to whether the next coil is the next coil in the existing order and whether the next coil is the last coil in the current order. The values include the width change value, the thickness change value, and the required heat change value. Then normalize the above data, and then use the weighted method to calculate the comprehensive change value, which are recorded as b1 and b2 respectively. If b1 < b2, the order is not changed, otherwise the order is changed. In this example, the first-level, second-level, and third-level systems of the annealing line are interconnected to obtain the complete data of the annealing line. The above information is collected in the order of the strip entering the production sequence, and the required heat of the strip is calculated according to the speed, preheating temperature, and median value of the strip target temperature and the strip physical properties. The width change value is the relative value of the width change, the thickness change value is the relative value of the thickness change, and the comprehensive weight is the weighted average. The weight coefficient calculation is first divided according to whether the required heat change value is greater than 0. If power increase is required, the overall power adjustable range needs to be considered, and the weight needs to be fine-tuned in combination with this.

[0038] S2: Establish a hybrid model for predicting the heating process system that combines the case-based reasoning model and the recurrent neural network model. At the same time, combine the existing knowledge to fully analyze the influence of fixed factors and random factors. On this basis, establish a prediction model and a reverse calculation model for the heating process system. First, match the case process library according to the relevant plan information of the current roll and the next roll, select the case with the best similarity, and then extrapolate based on the most similar case to obtain the heating process system of the two strips; when matching the cases, it is necessary to calculate the similarity, and calculate the weight of the case reasoning model based on the similarity. When the similarity is high, the case reasoning model is assigned a larger weight, and when the similarity is lower than the similarity threshold, the case reasoning model weight is reduced; when establishing the case reasoning model, it is necessary to analyze the weights of various influencing factors. According to production experience and relevant knowledge, the influencing factors are preliminarily determined to include strip speed, strip width, strip thickness, target temperature, and fuel input. The weights of various influencing factors are mainly determined by hierarchical analysis method and single factor analysis method, with hierarchical analysis method as the main method and single factor analysis method as the auxiliary method; according to the production sequence, a data source for training the recurrent neural network is formed in the historical library. When collecting data, the real-time data in seconds is first collected into a historical library at the minute level, and then the historical library at the minute level is collected into a historical database at the strip level (that is, each There is only one piece of data for the strip steel). In the process of assimilation, the missing data and abnormal data are processed at the same time. The abnormal data and missing data are corrected and filled according to mathematical knowledge, experience and previous and subsequent data. At the same time, the numerical mean, median, maximum and minimum values ​​and variance need to be retained when performing data assimilation. During training, every 10 rolls of strip steel are merged into one data source, and the cross-validation method is used for data division. The data is trained and predicted using a recurrent neural network method (such as the STA-LSTM method), and the model is saved. The model is updated every one to three days, that is, new data is added to the model training; the two models are mixed and weighted by weight, where the weight is determined based on the case reasoning similarity. If the similarity is high, the case reasoning weight is high, and vice versa; after the model is trained, the model is saved and used to predict the maximum target value and minimum target value of the current roll and the maximum target value and minimum target value of the next roll of strip steel corresponding to the heating process system, and the process system is recorded as k1min, k1max, k2min and k2max respectively.

[0039] S3: Obtain the heating process system corresponding to the maximum target value and minimum target value of the current coil and the maximum target value and minimum target value of the next coil in S2, reasonably select the appropriate transition strategy according to the calculated values ​​of the heating process systems of the previous and next coils and the on-site constraints, and manage the transition strategy according to three modes.

[0040] 1) If the minimum allowable load of the current coil is greater than the maximum allowable load of the next coil (k1min > k2max), first determine whether the next coil can be overheated. If the next coil can be overheated, then during transition management, operate at the speed and minimum allowable load of the current coil; if the next coil does not allow overheating, then use the S2 heating process regime prediction hybrid model to inversely calculate the maximum speed and minimum speed range corresponding to the minimum allowable load of the current coil for the next coil, denoted as speedmax and speedmin. When the strip weld comes out of the preheating section of the annealing furnace, start quickly adjusting the speed to the calculated speed upper limit speedmax and make the next coil operate at the new speed. When the strip weld comes out of the heating section of the annealing furnace, adjust both the load and speed simultaneously until the next coil reaches the planned set value.

[0041] 2) If the minimum allowable load of the current coil is less than the maximum allowable load of the next coil, but the maximum allowable load of the current coil is greater than the minimum allowable load of the next coil, that is, k1min < k2max and k1max > k2min, then select the load in the overlapping area to set the load during transition management, do not adjust the speed, and use the S2 heating process regime prediction hybrid model to inversely calculate the strip temperature set value.

[0042] 3) If the maximum allowable load of the current coil is less than the minimum allowable load of the next coil (k1max < k2min), it is necessary to set the transition strategy in combination with whether the current coil can be overheated. If the current coil can be overheated, then in the transition state, operate at the speed of the current coil and the minimum load of the next coil; if overheating is not allowed, then use the S2 heating process regime prediction hybrid model to inversely calculate the maximum speed and minimum speed range corresponding to the maximum load of the current coil for the next coil, denoted as speedmax and speedmin. When the strip weld comes out of the preheating section of the annealing furnace, start quickly adjusting the speed to the calculated lower limit speed speedmin and make the next coil operate at the new speed. When the strip weld exits the heating section of the annealing furnace, adjust both the load and speed simultaneously until the next coil reaches the planned set value.

[0043] In all strategies, it may be necessary to adjust the temperature and speed of the strip of the same specification to reach the expected value. Its adjustment strategy is as Figure 2As shown, due to the characteristics of the furnace itself, it is necessary to first obtain the time t1 when the heating process parameters are adjusted to the strip temperature and the time t2 when the strip speed is adjusted to the strip temperature, calculate the time difference between the two, match the speed adjustment with the heating process parameter adjustment through the time difference, and then calculate the influence of the heating process parameter change on the strip temperature and the influence of the speed change on the strip temperature according to the heating process system prediction hybrid model of S2, where the influence of speed and heating load changes includes three aspects: the time from the start of reaction to the start of action change, the time from the start of change to the target, and the time from the target to the start of the strip reaching the furnace position. In general, it is considered that it is linear when changing in a small area, and the slope of the speed change is obtained through the model, and then the required power is calculated based on the slope and the model, so as to match the heating process system with the speed change, and finally achieve the expected value of the temperature and speed of the same specification adjusted strip.

[0044] S4: Issue the calculated value according to S3, monitor the strip temperature in real time, and use the PID method to monitor the strip temperature, dynamically analyze the impact of changes in influencing factors, and correct the set value in time.

[0045] S5: When the strip steel is out of the furnace, the steel coils are digitally collected, and the case process library and history library are added or updated in time for the training of the hybrid model for heating process system prediction. In the process of collecting relevant parameters, abnormal data and missing data are processed according to the mathematical and physical model knowledge and the previous and next data, and the data is normalized.

[0046] When the tail weld of the current coil of strip steel comes out of the heating section of the annealing furnace, the data is aggregated by average, variance, median, mode, maximum, minimum and variance, and the influence of the transition management time before and after is considered. It is necessary to remove some data, and correct and fill in the abnormal data and missing data based on mathematical knowledge, experience and previous and subsequent data.

[0047] When the tail weld of the front coiled strip comes out of the heating section of the annealing furnace, the real-time collected data is aggregated into a historical database according to two methods for training the recurrent neural network model. The data is aggregated by average, variance, median, mode, maximum, minimum and variance, and the interval segmentation method is used to take every 10 groups of strip data as a data source and used to train the recurrent neural network model.

[0048] Example: The heating process system of the annealing furnace of a steel company's cold-rolled strip coating line was optimized.

[0049] Step 1: Data reading and sorting

[0050] Read the planning information of the first two rolls, the current roll and the next two rolls from the first-level system and the third-level system of the annealing furnace, including the strip thickness, width, specification, target temperature, speed, whether over-temperature is allowed, etc.

[0051] Calculate the energy required for each coil of strip (q1-q5).

[0052] The width change value, thickness change value, and required heat change value are calculated and normalized to obtain a comprehensive change value (b1, b2).

[0053] Determine whether the order of the last two volumes needs to be adjusted based on b1 and b2.

[0054] Step 2: Build a hybrid model

[0055] Case-based reasoning model:

[0056] Search for similar cases in the case library based on the current volume and next volume plan information.

[0057] Calculate case similarity and determine case reasoning model weights based on the similarity.

[0058] The heating process regime is obtained by extrapolation based on similar cases.

[0059] Recurrent Neural Network Model:

[0060] Extract data from historical database and perform data cleaning and normalization.

[0061] Use the STA-LSTM model for training and prediction, and save the model.

[0062] The model is updated every three days with new data.

[0063] Hybrid Model:

[0064] The prediction results of the case-based reasoning model and the recurrent neural network model are weighted and combined to obtain the final heating process system.

[0065] Step 3: Transition strategy selection

[0066] Mode 1: The minimum allowable load of the current roll > the maximum allowable load of the next roll

[0067] Determine whether the next roll can be overheated.

[0068] If over-temperature is allowed, the current volume will operate at the minimum allowable load.

[0069] If over-temperature is not possible, the next speed range is calculated based on the model and adjusted.

[0070] Mode 2: The minimum allowable load of the current roll < the maximum allowable load of the next roll < the maximum allowable load of the current roll

[0071] Select loads in overlapping areas for transition management.

[0072] The strip temperature setting value is back-calculated based on the model.

[0073] Mode 3: Maximum allowable load of current roll < minimum allowable load of next roll

[0074] Determine whether the current volume can be overheated.

[0075] If over-temperature is allowed, the current volume will operate at the minimum allowable load.

[0076] If over-temperature is not possible, the next speed range is calculated based on the model and adjusted.

[0077] Step 4: Distribution and monitoring

[0078] According to the calculation results of step 3, the heating process system is issued.

[0079] The strip temperature is monitored in real time and PID method is used for monitoring and adjustment.

[0080] Step 5: Data collection and model update

[0081] When the tail weld of the current coil steel comes out of the furnace, data is collected and abnormal data is processed.

[0082] The data is collected into a historical database for recurrent neural network model training.

[0083] Update the case library and historical database.

[0084] Assume that the strip thickness of the current coil and the next coil are 0.68 mm and 1.0 mm respectively, the widths are 904 mm and 924 mm respectively, the target temperatures are 815°C and 705°C respectively, and the speeds are 100 m / min and 100 m / min respectively.

[0085] Through the case-based reasoning model and the recurrent neural network model, the heating process regimes for the current volume and the next volume are obtained as follows:

[0086] Current volume: Full RTF section furnace heating, the heating power ratios of the 11 RTF section pulse burners are 4, 19, 34, 44, 61, 77, 82, 91, 97, 97, 0 respectively, and the target temperature of the RTF section is 815℃.

[0087] Next volume: Full RTF furnace heating, the heating power ratios of the 11 RTF pulse burners are 19, 37, 47, 55, 70, 85, 91, 91, 91, 94, 0 respectively, and the target temperature of the RTF section is 705℃. .

[0088] According to the transition strategy selection, the whole RTF section furnace is heated, the heating power ratios of the 11 RTF section pulse burners are 5, 20, 36, 47, 63, 80, 85, 94, 100, 100, 1, and the target temperature of the RTF section is 830°C. The final selected transition process is.

[0089] By optimizing the heating process system, the transition management time is shortened, the strip temperature control is more accurate, the length of unqualified products is reduced, and the production efficiency and product quality are improved.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for optimizing the heating process of an annealing furnace, characterized in that: The following steps are involved: S1: Read the planning information of the first two rolls, the current roll and the next two rolls to form a small data group of online production planning. According to the target temperature of each roll, the product specification transition information, whether the over-temperature information is allowed, and the transition situation of the first two rolls and the current roll, the preliminary sorting of the next two rolls is formed; S2: Combine the case-based reasoning method and the recurrent neural network model method to establish a hybrid model for predicting the heating process system. The case-based reasoning method is trained based on the heating process system case library, and the recurrent neural network model is iteratively trained based on the historical database data and the model is saved. The two methods are weighted to obtain the hybrid model for predicting the heating process system, and the hybrid model for predicting the heating process system is used to predict the heating process system of the current volume and the next volume after the preliminary sorting of S1. S3: According to S2, the corresponding heating process system range under steady state and the planning information of S1 are obtained, and the appropriate transition strategy is selected to carry out transition management of steel coil production and realize process optimization of transition management process; S4: Send the calculated value according to step S3, monitor the strip temperature in real time, combine the PID method to monitor the strip temperature, dynamically analyze the impact of changes in influencing factors, and correct the set value in time; S5: When the strip steel is out of the furnace, the steel coils are digitally collected, and the case process library and history library are added or updated in time for the training of the hybrid model for heating process system prediction. In the process of collecting relevant parameters, abnormal data and missing data are processed according to the mathematical and physical model knowledge and the previous and next data, and the data is normalized.

2. The annealing furnace heating process optimization method according to claim 1, characterized in that: The S1 comprises the following steps: S101: Read the schedule information of the current roll and the two rolls before and after from the first-level system and the third-level system of the annealing furnace. The calculation table information includes the strip width, strip thickness, strip specification, upper and lower limits of target temperature, speed requirements, over-temperature requirements and performance requirements; S102: Preliminarily calculate the required energy of each roll of strip steel according to the schedule information, sort the last two rolls of strip steel according to the transition strategy, relevant plan information and the transition relationship between the previous and next two rolls and the current roll, and preliminarily determine whether the position relationship between the next two rolls needs to be adjusted.

3. The annealing furnace heating process optimization method according to claim 2, characterized in that: In S2, establishing a heating process system prediction hybrid model comprises the following steps: S201: First, match the case process library according to the relevant plan information of the current roll and the next roll, select the case with the best similarity, and then extrapolate based on the most similar case to obtain the heating process system of the two strips; when matching the cases, it is necessary to calculate the similarity, and calculate the weight of the case reasoning model based on the similarity. When the similarity is high, the case reasoning model is assigned a larger weight, and when the similarity is lower than the similarity threshold, the case reasoning model weight is reduced; when establishing the case reasoning model, it is necessary to analyze the weights of various influencing factors, and preliminarily determine the influencing factors including strip speed, strip width, strip thickness, target temperature, and fuel input based on production experience and relevant knowledge, and use the hierarchical analysis method and the single factor analysis method to determine the weights of various influencing factors, with the hierarchical analysis method as the main method and the single factor analysis method as the auxiliary method; S202: A data source for training a recurrent neural network is formed in a historical database according to the production sequence. When collecting data, real-time data in seconds is first collected into a historical database at the minute level, and then the historical database at the minute level is collected into a historical database at the strip level, that is, each strip has only one piece of data. In the process of collection, missing data and abnormal data are processed at the same time. The abnormal data and missing data are corrected and filled according to mathematical knowledge, experience and previous and next data. At the same time, the numerical mean, median, maximum and minimum values ​​and variance need to be retained when collecting data. When training, every 10 rolls of strip steel are merged into one data source, and the cross-validation method is used for data division. The recurrent neural network method is used for data training and prediction, and the model is saved. The model is updated every one to three days, that is, new data is added to the model training; S203: The heating process conditions obtained in S201 and S202 are combined in a weighted manner to finally obtain the heating process condition of the strip steel.

4. The annealing furnace heating process optimization method according to claim 3, characterized in that: In S3, the transition strategy includes the following steps: S301: If the minimum allowable load of the current roll is greater than the maximum allowable load of the next roll, first determine whether the next roll can be overheated. If the next roll can be overheated, run at the current roll speed and the minimum allowable load during transition management; If the next coil is not allowed to overheat, the maximum speed and minimum speed range of the next coil corresponding to the minimum allowable load of the current coil are calculated through the S2 heating process system prediction hybrid model. When the strip weld comes out of the preheating section of the annealing furnace, the speed is quickly adjusted to the calculated speed upper limit, and the next coil is operated at the new speed. When the strip weld comes out of the heating section of the annealing furnace, the load and speed are adjusted at the same time until the next coil reaches the planned setting value; S302: If the minimum allowable load of the current coil is less than the maximum allowable load of the next coil, but the maximum allowable load of the current coil is greater than the minimum allowable load of the next coil, the load in the overlapping area is selected for transition management and the load is set without adjusting the speed, and the temperature setting value of the strip is calculated by reverse calculation through the S2 heating process system prediction hybrid model; S303: If the maximum allowable load of the current coil is less than the minimum allowable load of the next coil, the transition strategy needs to be set in combination with whether the current coil can be overheated. If the current coil can be overheated, the transition state runs at the current coil speed and the minimum load of the next coil; if overheating is not allowed, the S2 heating process system prediction hybrid model is used to reversely calculate the maximum speed and minimum speed range of the next coil corresponding to the maximum load of the current coil. When the strip weld comes out of the preheating section of the annealing furnace, the speed is quickly adjusted to the calculated lower limit speed, and the next coil is operated at the new speed. When the strip weld exits the heating section of the annealing furnace, the load and speed are adjusted at the same time until the next coil reaches the planned set value.

5. The method for optimizing the heating process of an annealing furnace according to claim 1, characterized in that: The S5 comprises the following steps: S501: When the tail weld of the current coil comes out of the heating section of the annealing furnace, the data average, variance, median, mode, maximum and minimum values ​​and variance are aggregated, and the influence of the transition management time before and after is considered. Some data needs to be removed, and the abnormal data and missing data are corrected and filled according to mathematical knowledge, experience and previous and next data; S502: The real-time collected data are aggregated into a historical database according to two methods for training the recurrent neural network model. Through the aggregation of data mean, variance, median, mode, maximum and minimum values ​​and variance, every 10 groups of strip data are used as a data source through the interval segmentation method and used to train the recurrent neural network model.

Images