Converter multi-source heterogeneous data model generation system and method for predicting temperature

Abstract

The invention provides a converter multi-source heterogeneous data model generation system and method for predicting temperature. The system comprises an input terminal, a processing terminal and a display device. The method is mainly applied to furnace condition simulation in the steelmaking process, and real-time simulation of carbon content and temperature in a converter is completed through the model, so that the purpose of real-time monitoring of the furnace condition is achieved. According to the system, through acquisition and learning of an audio slagging module, a flue gas analysis module, a furnace mouth flame module and a primary module of a converter device, abnormal evaluation rule models of audio data, flue gas data and furnace mouth flame are established respectively, and finally, the various models are fused through a machine learning algorithm in combination with a converter process mechanism model. And finally, a fusion model is formed to achieve the purpose of simulating furnace conditions in real time. According to the system, the real-time performance of steelmaking control is accelerated, and misoperation caused by cooperation of multiple persons is reduced.

Description

Technical Field

[0001] The present application relates to the field of automatic control technology, and in particular to a system and method for generating a multi-source heterogeneous data model of a converter for predicting temperature. Background Art

[0002] In the steel industry, molten iron is produced by reducing iron ore with carbon. The carbon source required to produce the molten iron is about 500 kg per ton of molten iron on average. When cold iron sources such as scrap iron are used as raw materials in the converter to produce molten steel, the carbon source required for the reduction of iron ore is not needed. The main production task of the converter plant is to further smelt the pig iron smelted from the blast furnace into qualified molten steel, and cast it into qualified ingots and send them to various rolling mills for rolling. The purpose of converter smelting is to oxidize the carbon and other impurities (such as silicon, manganese) in the pig iron to form gaseous substances and discharge them out of the furnace to produce steel with better physical, chemical and mechanical properties than iron.

[0003] Converter steelmaking is a process that uses molten iron, scrap steel, and ferroalloys as the main raw materials. It does not rely on external energy sources, but relies on the physical heat of the molten iron itself and the heat generated by the chemical reactions between the components of the molten iron to complete the steelmaking process in the converter. Converters are divided into acidic and alkaline types according to the refractory materials, and top-blown, bottom-blown, and side-blown types according to the location where the gas is blown into the furnace; they are divided into air converters and oxygen converters according to the type of gas. Alkaline oxygen top-blown and top-bottom combined-blown converters are the most commonly used steelmaking equipment because of their fast production speed, large output, high single-furnace output, low cost, and low investment.

[0004] At present, in the process of converter steelmaking, large steel mills cannot install common carbon temperature measuring devices because of the high requirements of converter process, harsh reaction environment, and extremely high risk factors. Therefore, they can only simulate the condition information in the furnace through model calculation and other methods. The traditional converter secondary model only simulates the data of converter smelting under ideal conditions, with large deviations in the early predictions, and only gives the end state results, which cannot cope with the real, changeable and complex smelting conditions. This results in the operator being unable to control the real-time conditions in the furnace, which will reduce the real-time control of steelmaking and even cause losses. Summary of the invention

[0005] The present application provides a system and method for generating a multi-source heterogeneous data model of a converter for predicting temperature, so as to solve the problem that the existing secondary model of the converter has large prediction deviations only in the early stage and only gives the terminal state results, and is unable to cope with the real, changeable and complex smelting conditions, resulting in the operator being unable to control the real-time conditions in the furnace, which reduces the real-time performance of steelmaking control and even causes losses.

[0006] In a first aspect, the present application provides a converter multi-source heterogeneous data model generation system for predicting temperature, comprising: an input terminal, a processing terminal, and a display device; the input terminal and the display device are respectively connected to a communication interface of the processing terminal;

[0007] The input terminal is configured to: obtain furnace data and send it to the processing terminal;

[0008] The processing terminal is configured to:

[0009] receiving the heat data, and establishing an excellent heat set and an abnormal heat set from the heat data based on the abnormal information;

[0010] Establishing multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal heat set;

[0011] Establishing a process mechanism model based on the data in the excellent heat set;

[0012] Calculating a deviation value based on the process mechanism model and the actual measured temperature;

[0013] Building a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and training and learning the model;

[0014] After the model training is completed, the target model is output, and a temperature prediction result is generated based on the target model;

[0015] Sending the temperature prediction result to the display device;

[0016] The display device is configured to generate a target image according to the temperature prediction result, and display the target image on a display of the display device, wherein the target image is used to characterize the distribution of each temperature prediction value.

[0017] In some possible implementations, the processing terminal includes a storage unit, a computing unit, and a control unit, wherein the storage unit and the computing unit are electrically connected to the control unit respectively; the storage unit is used to store the calculated intermediate values ​​of the system; the computing unit is used to perform calculations; and the control unit is used to send control instructions.

[0018] In some possible implementations, calculating the deviation value based on the process mechanism model and the actual measured temperature includes:

[0019] Obtaining a theoretically predicted temperature based on the process mechanism model;

[0020] Get the actual measured temperature;

[0021] A first formula is constructed based on the theoretical predicted temperature and the actual measured temperature. The first formula is theoretical predicted temperature - actual measured temperature = Σ{FN (score of abnormal data of each detection source) * each abnormal weight}, wherein the difference between the theoretical predicted temperature and the actual measured temperature is the deviation value, and FN represents the temperature change caused by each abnormal feature.

[0022] In some possible implementations, constructing the first formula according to the predicted temperature and the actual measured temperature includes:

[0023] Constructing a learning model according to the first formula and multiple abnormal evaluation rule models;

[0024] The learning model performs machine learning training on the weight values ​​corresponding to the abnormal features based on the training set in the abnormal furnace set;

[0025] After the training is completed, the optimal weight value of the abnormal feature is output.

[0026] In some possible implementations, after outputting the optimal weighted value of the abnormal feature, the following is also included:

[0027] Substituting the optimal weight value into the first formula to obtain a target model;

[0028] Substituting the validation set in the abnormal heat set into the target model to obtain a temperature estimate;

[0029] The temperature estimation value is verified with the theoretical predicted temperature to obtain an evaluation value, and when the evaluation value is less than a preset threshold, the target model is output.

[0030] In some possible implementations, the learning model performs machine learning training on the weight values ​​corresponding to the abnormal features based on the training set in the abnormal furnace set, including:

[0031] Collect the actual measured TSC temperature of each furnace and the TSC temperature calculated by the process mechanism model;

[0032] Collect audio slag data to calculate the first temperature change score;

[0033] Collect smoke analysis data to calculate the second temperature change score;

[0034] Collect the furnace mouth flame data to calculate the third temperature change score;

[0035] The optimal weight of the abnormal feature is calculated by combining multiple temperature change scores through a machine learning algorithm.

[0036] In some possible implementations, when the evaluation value is greater than or equal to a preset threshold, a gradient descent algorithm is used to adjust multiple weight values, and the corrected weight values ​​are substituted into a machine learning algorithm to find an optimal solution.

[0037] In some possible implementations, receiving the heat data includes:

[0038] Screening and cleaning the heat data to obtain standard heat data;

[0039] Establishing an abnormal heat set based on the standard heat data characterized as abnormal by abnormal information in the standard heat data;

[0040] An excellent heat set is established based on the standard heat data characterized as normal by abnormal information in the standard heat data.

[0041] In some possible implementations, establishing multiple abnormality evaluation rule models based on abnormal features of data in the abnormal heat set includes:

[0042] Multiple abnormal evaluation rule models are established based on the audio slagging, flue gas analysis, furnace mouth flame and oxygen lance height characteristics of the data in the abnormal furnace set.

[0043] In a second aspect, the present application provides a method for generating a converter multi-source heterogeneous data model for predicting temperature, which is configured in the converter multi-source heterogeneous data model generation system for predicting temperature according to the first aspect, and the method comprises:

[0044] The input terminal obtains the furnace data and sends it to the processing terminal;

[0045] The processing terminal receives the heat data, and creates an excellent heat set and an abnormal heat set based on the abnormal information;

[0046] Establishing multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal heat set;

[0047] Establishing a process mechanism model based on the data in the excellent heat set;

[0048] Calculating a deviation value based on the process mechanism model and the actual measured temperature;

[0049] Building a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and training and learning the model;

[0050] After the model training is completed, the target model is output, and a temperature prediction result is generated based on the target model;

[0051] Sending the temperature prediction result to the display device;

[0052] The display device generates a target image according to the temperature prediction result, and displays the target image on a display of the display device, wherein the target image is used to characterize the distribution of each temperature prediction value.

[0053] From the above content, it can be seen that the present application provides a converter multi-source heterogeneous data model generation system and method for predicting temperature. The system includes an input terminal, a processing terminal and a display device; the input terminal obtains the furnace data and sends it to the processing terminal; the processing terminal receives the furnace data, and, based on the abnormal information, establishes an excellent furnace set and an abnormal furnace set with the furnace data; establishes multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal furnace set; establishes a process mechanism model based on the data in the excellent furnace set; calculates the deviation value based on the process mechanism model and the actual measured temperature; constructs a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and trains and learns the model; after the model training is completed, outputs the target model, and generates a temperature prediction result based on the target model; sends the temperature prediction result to the display device; the display device generates a target image according to the temperature prediction result, and displays the target image on the display of the display device. The system is mainly used for the furnace condition simulation of the steelmaking process, and the simulation of the real-time carbon content and temperature in the converter is completed through the model to achieve the purpose of real-time monitoring of the furnace condition. The system collects and learns the audio slag module, flue gas analysis module, furnace mouth flame module and primary module of the converter device, and establishes the abnormal evaluation rule model of audio data, flue gas data and furnace mouth flame respectively. Finally, it combines the converter process mechanism model with the machine learning algorithm to fuse multiple models and finally forms a fusion model to achieve the purpose of real-time simulation of furnace conditions. The system can reduce the human and material resources in the steelmaking process and reduce the learning cost of steelmaking control. At the same time, it speeds up the real-time control of steelmaking and reduces the misoperation caused by multi-person collaboration. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0055] Figure 1 A schematic diagram of a converter multi-source heterogeneous data model generation system for predicting temperature is provided for an embodiment of the present application;

[0056] Figure 2 A schematic diagram of the audio slag trend provided in an embodiment of the present application;

[0057] Figure 3 A schematic diagram of the trend of flue gas analysis provided in the embodiment of the present application;

[0058] Figure 4 A schematic diagram of the furnace mouth flame trend provided in an embodiment of the present application;

[0059] Figure 5 A schematic diagram of the model fusion principle provided in the embodiment of the present application;

[0060] Figure 6 A schematic diagram of the BP neural network principle provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are only examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

[0062] In the steel industry, molten iron is produced by reducing iron ore with carbon. The carbon source required to produce the molten iron is about 500 kg per ton of molten iron on average. When cold iron sources such as scrap iron are used as raw materials in the converter to produce molten steel, the carbon source required for the reduction of iron ore is not needed. The main production task of the converter plant is to further smelt the pig iron smelted from the blast furnace into qualified molten steel, and cast it into qualified ingots and send them to various rolling mills for rolling. The purpose of converter smelting is to oxidize the carbon and other impurities (such as silicon, manganese) in the pig iron to form gaseous substances and discharge them out of the furnace to produce steel with better physical, chemical and mechanical properties than iron.

[0063] Converter steelmaking is a process that uses molten iron, scrap steel, and ferroalloys as the main raw materials. It does not rely on external energy sources, but relies on the physical heat of the molten iron itself and the heat generated by the chemical reactions between the components of the molten iron to complete the steelmaking process in the converter. Converters are divided into acidic and alkaline types according to the refractory materials, and top-blown, bottom-blown, and side-blown types according to the location where the gas is blown into the furnace; they are divided into air converters and oxygen converters according to the type of gas. Alkaline oxygen top-blown and top-bottom combined-blown converters are the most commonly used steelmaking equipment because of their fast production speed, large output, high single-furnace output, low cost, and low investment.

[0064] The operation process of converter steelmaking includes the following steps: Adding raw materials: Put raw materials such as pig iron, scrap steel and limestone into the converter in a certain proportion. Blowing oxygen: After adding the raw materials, oxygen is blown into the converter by spraying oxygen to oxidize the impurities in the pig iron and form gaseous substances that are discharged out of the furnace. Adding alloys: In the steelmaking process, in order to adjust the composition and properties of the steel, alloys such as chromium, nickel, molybdenum, etc. need to be added. Sampling and analysis: In the steelmaking process, the molten steel needs to be sampled and analyzed to ensure that the composition and properties of the steel meet the requirements. Steelmaking: When the impurities in the molten steel are completely oxidized, the furnace mouth is opened to let the high-quality molten steel flow out.

[0065] However, in the current converter steelmaking process of large steel mills, due to the high requirements of converter technology, harsh reaction environment, and extremely high risk factors, it is impossible to install common carbon temperature measurement devices, so the only way to simulate the furnace condition information is through model calculation. The traditional converter secondary model only simulates the data of converter smelting under ideal conditions, with large deviations in the early predictions, and only gives the end state results, which cannot cope with the real, changeable, and complex smelting conditions. This results in the operator being unable to control the real-time conditions in the furnace, which will reduce the real-time control of steelmaking and even cause losses.

[0066] Based on this, the present application provides a converter multi-source heterogeneous data model generation system and method for predicting temperature. The present application provides a system that collects and learns the audio slag-making module, flue gas analysis module, furnace mouth flame module and primary module of the converter device, respectively establishes abnormal evaluation rule models for audio data, flue gas data and furnace mouth flame, and finally combines the converter process mechanism model through machine learning algorithms to fuse multiple models, and finally forms a fusion model to achieve the purpose of real-time simulation of furnace conditions. The system can reduce human and material resources in the steelmaking process and reduce the learning cost of personnel for steelmaking control. At the same time, it speeds up the real-time control of steelmaking and reduces the misoperation caused by multi-person collaboration.

[0067] In some embodiments, Figure 1 As shown, the present application provides a converter multi-source heterogeneous data model generation system for predicting temperature, comprising: an input terminal, a processing terminal and a display device; the input terminal and the display device are respectively connected to the communication interface of the processing terminal;

[0068] The input terminal is configured to: obtain furnace data and send it to the processing terminal;

[0069] The processing terminal is configured to:

[0070] Receive the heat data, and establish an excellent heat set and an abnormal heat set based on the abnormal information; wherein the abnormal heat set includes a training set and a validation set;

[0071] Establishing multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal heat set;

[0072] Establishing a process mechanism model based on the data in the excellent heat set;

[0073] Calculating a deviation value based on the process mechanism model and the actual measured temperature;

[0074] Building a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and training and learning the model;

[0075] After the model training is completed, the target model is output, and a temperature prediction result is generated based on the target model;

[0076] Sending the temperature prediction result to the display device;

[0077] The display device is configured to generate a target image according to the temperature prediction result, and display the target image on a display of the display device, wherein the target image is used to characterize the distribution of each temperature prediction value.

[0078] In some embodiments, the processing terminal includes a storage unit, a computing unit and a control unit, wherein the storage unit and the computing unit are electrically connected to the control unit respectively; the storage unit is used to store the calculated intermediate values ​​of the system; the computing unit is used to perform calculations; and the control unit is used to send control instructions.

[0079] In some embodiments, calculating the deviation value based on the process mechanism model and the actual measured temperature includes:

[0080] Obtaining a theoretically predicted temperature based on the process mechanism model;

[0081] Get the actual measured temperature;

[0082] A first formula is constructed based on the theoretical predicted temperature and the actual measured temperature. The first formula is theoretical predicted temperature - actual measured temperature = Σ{FN (score of abnormal data of each detection source) * each abnormal weight}, wherein the difference between the theoretical predicted temperature and the actual measured temperature is the deviation value, and FN represents the temperature change caused by each abnormal feature.

[0083] In some embodiments, constructing a first formula according to the predicted temperature and the actual measured temperature includes:

[0084] Constructing a learning model according to the first formula and multiple abnormal evaluation rule models;

[0085] The learning model performs machine learning training on the weight values ​​corresponding to the abnormal features based on the training set in the abnormal furnace set;

[0086] After the training is completed, the optimal weight value of the abnormal feature is output.

[0087] In some embodiments, after outputting the optimal weight value of the abnormal feature, the method further includes:

[0088] Substituting the optimal weight value into the first formula to obtain a target model;

[0089] Substituting the validation set in the abnormal heat set into the target model to obtain a temperature estimate;

[0090] The temperature estimation value is verified with the theoretical predicted temperature to obtain an evaluation value, and when the evaluation value is less than a preset threshold, the target model is output.

[0091] In some embodiments, the present application verifies the temperature estimate with the theoretical predicted temperature to obtain an evaluation value based on R 2 The specific process is as follows:

[0092] numerator=Σ((y_true-y_pred)^2);

[0093] denominator=Σ((y_true-avg(y_true))^2);

[0094] R 2 _score=1-(numerator / denominator);

[0095] Among them, y_true represents the actual result, y_pred represents the predicted result, and avg represents the average value;

[0096] Numerator: Subtract the predicted value from the actual y value, square it, and then sum it;

[0097] Denominator: square the average of each y value minus the actual y value and then sum it up;

[0098] R 2 =1-numerator / denominator.

[0099] In some embodiments, the learning model performs machine learning training on the weight values ​​corresponding to the abnormal features based on the training set in the abnormal heat set, including:

[0100] Collect the actual measured TSC temperature of each furnace and the TSC temperature calculated by the process mechanism model;

[0101] Collect audio slag data to calculate the first temperature change score;

[0102] Collect smoke analysis data to calculate the second temperature change score;

[0103] Collect the furnace mouth flame data to calculate the third temperature change score;

[0104] Combining multiple temperature change scores, TSC actual measured temperature and TSC temperature, the optimal weight value of the abnormal feature is calculated through a machine learning algorithm.

[0105] For large converters controlled by auxiliary lances, the temperature control is divided into two stages, with TSC probe measurement as the dividing point. The stage before is the TSC process stage, and the stage after measurement is the endpoint control stage. The TSC process stage mainly controls carbon and temperature by blowing oxygen to make slag, so as to ensure that the converter endpoint composition and endpoint temperature are hit. The actual TSC measured temperature and the TSC temperature calculated by the process mechanism model are the actual measured temperature of the TSC process stage and the temperature calculated by the process mechanism model, respectively.

[0106] In some embodiments, when the evaluation value is greater than or equal to a preset threshold, a gradient descent algorithm is used to adjust multiple weight values, and the corrected weight values ​​are substituted into a machine learning algorithm to find an optimal solution.

[0107] In some embodiments, receiving the heat data includes:

[0108] Screening and cleaning the heat data to obtain standard heat data;

[0109] Establishing an abnormal heat set based on the standard heat data characterized as abnormal by abnormal information in the standard heat data;

[0110] An excellent heat set is established based on the standard heat data characterized as normal by abnormal information in the standard heat data.

[0111] In some embodiments, establishing multiple abnormality evaluation rule models based on the abnormal characteristics of the data in the abnormal heat set includes:

[0112] Multiple abnormal evaluation rule models are established based on the audio slagging, flue gas analysis, furnace mouth flame and oxygen lance height characteristics of the data in the abnormal furnace set.

[0113] In some embodiments, the present application provides a method for generating a converter multi-source heterogeneous data model for predicting temperature, which is configured in the converter multi-source heterogeneous data model generation system for predicting temperature described in the above embodiment, and the method includes:

[0114] The input terminal obtains the furnace data and sends it to the processing terminal;

[0115] The processing terminal receives the heat data, and creates an excellent heat set and an abnormal heat set based on the abnormal information;

[0116] Establishing multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal heat set;

[0117] Establishing a process mechanism model based on the data in the excellent heat set;

[0118] Calculating a deviation value based on the process mechanism model and the actual measured temperature;

[0119] Building a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and training and learning the model;

[0120] After the model training is completed, the target model is output, and a temperature prediction result is generated based on the target model;

[0121] Sending the temperature prediction result to the display device;

[0122] The display device generates a target image according to the temperature prediction result, and displays the target image on a display of the display device, wherein the target image is used to characterize the distribution of each temperature prediction value.

[0123] This application is mainly aimed at the shortcoming that the secondary static model of the converter cannot accurately simulate the furnace condition. By integrating multi-source real-time information such as audio slagging, flue gas analysis, and furnace mouth flame, the original static model is transformed into a dynamic model, thereby achieving the purpose of real-time and accurate prediction of the furnace condition. This application establishes an excellent furnace data set and an abnormal feature evaluation model for each data source by performing data cleaning and data analysis on multi-source data, giving the data feature evaluation results, and then using the neural network algorithm in machine learning to calculate the weight of each data source information, and finally obtains the fitted fusion model through fitting through a large amount of furnace data.

[0124] Example

[0125] Step 1: Data cleaning. The processing terminal receives the heat data sent by the input terminal, and screens and cleans the heat data with unqualified data. The unqualified heats mainly include heats with missing component test data or untested components, heats with large errors in temperature and other measurement data from the normal range, and heats with major accidents in the smelting process that cause the smelting time to be too long.

[0126] Step 2: Establish an excellent heat data set. After data screening, all collected heat data are divided into two categories.

[0127] Excellent heat set, which refers to the heats with relatively stable and normal smelting process, and the heats where the data of each detection source have not experienced major fluctuations or given alarm information. The data in this set is used to optimize the process model parameters, such as the cooling coefficient of the coolant, the composition ratio coefficient of each compound in the auxiliary materials, etc.

[0128] Abnormal heat collection, which is the heat in which the smelting process is unstable, the data of each detection source fluctuates abnormally or gives alarm information, such as splashing, back-drying, oxygen escape caused by various factors, etc. These abnormal conditions are the main characteristics and measurement objects of the fusion model.

[0129] Step 3: Establish an abnormal evaluation rule model for the abnormal features of multiple detection source data respectively. The abnormal evaluation rule model includes: audio slagging model, flue gas analysis model, furnace mouth flame model and other models.

[0130] (1) Audio slag: The audio trend of audio slag can reflect the current slag condition. When the audio intensity exceeds or falls below the normal range, it indicates that the molten steel slag has abnormal conditions such as splashing or back-drying (see Figure 2 ).

[0131] Abnormal score = Σn{FN(audio abnormal signal strength - audio normal range))};

[0132] |Abnormal audio signal strength - normal audio range| means that when the audio strength is higher than the upper limit of the normal range (splashing abnormality), the value is audio strength - upper limit of the normal range; if the audio strength is lower than the lower limit of the normal range (anti-drying abnormality), the value is lower limit of the normal range - audio strength. n is the number of consecutive abnormalities.

[0133] For example: Figure 2 In the figure, at 500s and 650s, the audio intensity (yellow line) exceeds the red line range. Above the red line, it is considered a splash warning. At 500s, the audio intensity is 8, and the normal range given is 20-40. The score at this time point is |8-20|=8. At a frequency of one sample per second, all time points that exceed the normal range are accumulated, and the cumulative score is about 700.

[0134] (2) Flue gas analysis: When CO or CO2 drops in the flue gas analysis system, it is considered an abnormal situation (see Figure 3 ).

[0135] Abnormal score = Σn{FN(CO break degree)*duration)};

[0136] FN (CO breaking degree) means that the CO content appears to be in a downward state in the middle of smelting and exceeds a certain limit.

[0137] n is the number of consecutive exceptions. If the abnormal state continues without interruption, it is considered as one exception.

[0138] For example (modify according to circumstances): The regulations are as follows,

[0139] When more than 5 < number of declines < 10, FN = 1;

[0140] When more than 10 < number of declines < 20, FN = 2;

[0141] When more than 20 < the number of declines, FN = 3;

[0142] Figure 3 There were two large fluctuations at 280 seconds and 480 seconds of smelting, with FN of 3 and 2 respectively, and duration of 60s and 40s respectively. The score = 3*60+2*40=260.

[0143] (3) Furnace mouth flame: Use infrared camera to collect the flame at the furnace mouth. Because the smoke and dust in the early stage have a greater impact on flame detection, only the flame temperature in the middle and late stages of smelting is collected. At a frequency of one sample per second, if the temperature between the first and last two measurements is greater than 100 degrees (can be modified according to actual conditions), it is considered a fluctuation, and the number of fluctuations is counted as a score.

[0144] Abnormal score = number of flame fluctuations;

[0145] For example (can be modified according to actual situation): Figure 4 In the process of smelting from 550 seconds to 700 seconds, a large number of fluctuations occurred. The number of fluctuations was statistically about 80 times, and the score was 80.

[0146] (4) Other primary information (oxygen lance height):

[0147] It mainly focuses on the adjustment of the oxygen lance position during the smelting process. In order to fully react the solvent and deal with splashing and back-drying during the smelting process, the method of constantly adjusting the lance position is often used. The fluctuation of the lance position will cause some oxygen to escape, resulting in the loss of the thermal effect of the oxygen reaction in the mechanism model.

[0148] Other abnormal score = Σn{primary oxygen lance height fluctuation};

[0149] The oxygen lance height fluctuation score = current oxygen lance height - previous oxygen lance height - 5. If the score is less than 0, the score is considered to be equal to 0.

[0150] Step 4, establish the converter process mechanism model. The mechanism model is used to predict the temperature in the furnace under ideal smelting conditions, and use the excellent furnace data set to complete the adjustment of the equivalent coefficients of the process mechanism model. The mechanism model is mainly based on the principles of material balance and heat balance, by analyzing the chemical reactions in the furnace and adjusting the equivalent coefficients of various calculations. For example, calculate the thermal effect of the reaction of C, P, S, Si, Mn with oxygen, the ratio of generated CO and CO2, oxygen utilization, heat loss carried away by furnace gas, etc., and calculate the amount of coolant needed to reach the target temperature. The mechanism model is a common method for all current converter secondary systems, and is used to calculate the amount of material discharged and carbon temperature prediction.

[0151] Step 5: Establish a multi-source heterogeneous data fusion model based on the converter process mechanism model. The basic principle is that the deviation between the results predicted by the ideal process mechanism model and the actual measurement results is caused by various abnormal events in the reaction process. The formula is:

[0152] Ideally, the temperature predicted by the mechanism model minus the actual measured temperature = Σ{FN (score of abnormal data of each detection source)*weight of each abnormality};

[0153] Where FN represents the temperature change caused by each anomaly.

[0154] This equation is used to build a machine learning model, using each anomaly score as a feature and the prediction deviation of the process model as the target (see Figure 5 ), the neural network algorithm will automatically complete the weight calculation for each anomaly.

[0155] 1. Collect the actual TSC measurement temperature of each furnace.

[0156] 2. Collect the TSC temperature calculated by the process mechanism model for each furnace.

[0157] 3. Collect audio slag data to calculate the temperature change score.

[0158] 4. Collect flue gas analysis data and calculate the temperature change score.

[0159] 5. Collect furnace mouth flame data and calculate temperature change score.

[0160] Combining multiple temperature change scores, TSC actual measured temperature and TSC temperature, the optimal weight value of the abnormal feature is calculated through a machine learning algorithm.

[0161] Step 6: Use the deviation results obtained by machine learning algorithms (BP neural network, XGboost, etc.) to adjust the process mechanism model results to complete the dynamic prediction of real-time furnace condition information.

[0162] For example (such as Figure 6 ): Use BP neural network for training, use two hidden layers and one output layer for training, use parameters as mechanism model prediction temperature (given by the second stage of converter) and TSC measurement temperature (given by the auxiliary gun), use the difference between these two parameters as the learning target y, use audio slagging score, furnace mouth flame score, flue gas analysis score, and other first-level information scores as features x, and learn the weight w of each feature.

[0163] Substitute the optimal weight value into the first formula to obtain the target model; substitute the verification set in the abnormal furnace set into the target model to obtain the temperature estimation value; verify the temperature estimation value with the theoretical prediction temperature to obtain an evaluation value, and when the evaluation value is less than a preset threshold, output the target model. The processing terminal generates a temperature prediction result based on the target model; sends the temperature prediction result to the display device; the display device generates a target image according to the temperature prediction result, and displays the target image on the display of the display device.

[0164] As can be seen from the above embodiments, the present application provides a converter multi-source heterogeneous data model generation system and method for predicting temperature. The system includes an input terminal, a processing terminal and a display device; the input terminal obtains the furnace data and sends it to the processing terminal; the processing terminal receives the furnace data, and, based on the abnormal information, establishes an excellent furnace set and an abnormal furnace set with the furnace data; establishes multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal furnace set; establishes a process mechanism model based on the data in the excellent furnace set; calculates the deviation value based on the process mechanism model and the actual measured temperature; constructs a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and trains and learns the model; after the model training is completed, outputs the target model, and generates a temperature prediction result based on the target model; sends the temperature prediction result to the display device; the display device generates a target image according to the temperature prediction result, and displays the target image on the display of the display device. The system is mainly used for the furnace condition simulation of the steelmaking process, and the simulation of the real-time carbon content and temperature in the converter is completed through the model to achieve the purpose of real-time monitoring of the furnace condition. The system collects and learns the audio slag module, flue gas analysis module, furnace mouth flame module and primary module of the converter device, and establishes the abnormal evaluation rule model of audio data, flue gas data and furnace mouth flame respectively. Finally, it combines the converter process mechanism model with the machine learning algorithm to fuse multiple models and finally forms a fusion model to achieve the purpose of real-time simulation of furnace conditions. The system can reduce the human and material resources in the steelmaking process and reduce the learning cost of steelmaking control. At the same time, it speeds up the real-time control of steelmaking and reduces the misoperation caused by multi-person collaboration.

[0165] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims

1. A converter multi-source heterogeneous data model generation system for predicting temperature, characterized in that: include: An input terminal, a processing terminal and a display device; the input terminal and the display device are respectively connected to the communication interface of the processing terminal; The input terminal is configured to: obtain furnace data and send it to the processing terminal; The processing terminal is configured to: receiving the heat data, and establishing an excellent heat set and an abnormal heat set from the heat data based on the abnormal information; Establishing multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal heat set; Establishing a process mechanism model based on the data in the excellent heat set; Calculating a deviation value based on the process mechanism model and the actual measured temperature; Building a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and training and learning the model; After the model training is completed, the target model is output, and a temperature prediction result is generated based on the target model; Sending the temperature prediction result to the display device; The display device is configured to generate a target image according to the temperature prediction result, and display the target image on a display of the display device, wherein the target image is used to characterize the distribution of each temperature prediction value.

2. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 1, characterized in that: The processing terminal includes a storage unit, a calculation unit and a control unit, wherein the storage unit and the calculation unit are electrically connected to the control unit respectively; the storage unit is used to store the calculated intermediate values ​​of the system; The calculation unit is used to perform calculations; the control unit is used to send control instructions.

3. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 1, characterized in that: The deviation value calculated based on the process mechanism model and the actual measured temperature includes: Obtaining a theoretically predicted temperature based on the process mechanism model; Get the actual measured temperature; A first formula is constructed based on the theoretical predicted temperature and the actual measured temperature. The first formula is theoretical predicted temperature - actual measured temperature = Σ{FN (score of abnormal data of each detection source) * each abnormal weight}, wherein the difference between the theoretical predicted temperature and the actual measured temperature is the deviation value, and FN represents the temperature change caused by each abnormal feature.

4. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 3, characterized in that: The first formula constructed according to the predicted temperature and the actual measured temperature includes: Constructing a learning model according to the first formula and multiple abnormal evaluation rule models; The learning model performs machine learning training on the weight values ​​corresponding to the abnormal features based on the training set in the abnormal furnace set; After the training is completed, the optimal weight value of the abnormal feature is output.

5. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 4, characterized in that: After outputting the optimal weighted value of the abnormal feature, it also includes: Substituting the optimal weight value into the first formula to obtain a target model; Substituting the validation set in the abnormal heat set into the target model to obtain a temperature estimate; The temperature estimation value is verified with the theoretical predicted temperature to obtain an evaluation value, and when the evaluation value is less than a preset threshold, the target model is output.

6. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 4, characterized in that: The learning model performs machine learning training on the weight values ​​corresponding to the abnormal features based on the training set in the abnormal furnace set, including: Collect the actual measured TSC temperature of each furnace and the TSC temperature calculated by the process mechanism model; Collect audio slag data to calculate the first temperature change score; Collect smoke analysis data to calculate the second temperature change score; Collect the furnace mouth flame data to calculate the third temperature change score; The optimal weight of the abnormal feature is calculated by combining multiple temperature change scores through a machine learning algorithm.

7. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 5, characterized in that: When the evaluation value is greater than or equal to a preset threshold, a plurality of weight values ​​are adjusted using a gradient descent algorithm, and the corrected weight values ​​are substituted into a machine learning algorithm to find an optimal solution.

8. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 1, characterized in that: Receiving the heat data includes: Screening and cleaning the heat data to obtain standard heat data; Establishing an abnormal heat set based on the standard heat data characterized as abnormal by abnormal information in the standard heat data; An excellent heat set is established based on the standard heat data characterized as normal by abnormal information in the standard heat data.

9. The converter multi-source heterogeneous data model generation system for predicting temperature according to claim 1, characterized in that: Establishing multiple abnormal evaluation rule models based on the abnormal features of the data in the abnormal heat set includes: Multiple abnormal evaluation rule models are established based on the audio slagging, flue gas analysis, furnace mouth flame and oxygen lance height characteristics of the data in the abnormal furnace set.

10. A method for generating a converter multi-source heterogeneous data model for predicting temperature, characterized in that: In the converter multi-source heterogeneous data model generation system for predicting temperature according to any one of claims 1 to 9, the method comprises: The input terminal obtains the furnace data and sends it to the processing terminal; The processing terminal receives the heat data, and creates an excellent heat set and an abnormal heat set based on the abnormal information; Establishing multiple abnormal evaluation rule models based on the abnormal characteristics of the data in the abnormal heat set; Establishing a process mechanism model based on the data in the excellent heat set; Calculating a deviation value based on the process mechanism model and the actual measured temperature; Building a machine learning model based on the deviation value and multiple abnormal evaluation rule models, and training and learning the model; After the model training is completed, the target model is output, and a temperature prediction result is generated based on the target model; Sending the temperature prediction result to the display device; The display device generates a target image according to the temperature prediction result, and displays the target image on a display of the display device, wherein the target image is used to characterize the distribution of each temperature prediction value.

Images