A converter gas feasible recovery amount prediction method based on multi-source heterogeneous data

Abstract

The application discloses a converter gas feasible recovery amount prediction method based on multi-source heterogeneous data and belongs to the technical field of waste energy recovery in the steel industry. The method comprises the following steps: converter gas generation mechanism analysis, obtaining influencing factors; collecting corresponding primary machine and secondary machine data, and preprocessing the primary machine data into a data set in units of furnaces; fusing the primary and secondary data to form a converter gas feasible recovery amount prediction data set; using a neural network prediction method, taking the influencing factors as an input layer and the gas feasible recovery amount as an output layer, establishing a gas feasible recovery amount prediction model; collecting signals such as converter smelting steel grades, blowing start time and blowing end time in the third machine steelmaking production plan data, determining the converter production process type, according to the corresponding operation factors of the type converter production, using the prediction model, and realizing accurate prediction of the converter gas feasible recovery amount.

Description

Technical Field

[0001] This invention relates to the field of waste energy recovery technology in the steel industry, specifically to a method for predicting the feasible recovery amount of converter gas based on multi-source heterogeneous data. Background Technology

[0002] Recovering energy lost during production and various forms of energy is an indispensable part of energy conservation in the steel industry. Converter gas, one of the three major by-product gases in steel enterprises, exhibits intermittent, complex, and uncertain characteristics in its recovery due to the cyclical and complex nature of converter production. This can easily lead to gas tank overflow and rejection, resulting in gas venting and increasing the difficulty of dynamic balance control of by-product gas throughout the plant. Accurate prediction of converter gas recovery volume is crucial for optimizing by-product gas balance control, effectively improving converter gas recovery rates and reducing gas venting.

[0003] The amount of converter gas recovered is mainly related to the converter gas generation rate, the converter gas recovery conditions, and whether the converter gas holder is full during the recovery period. Specifically, the converter gas generation rate is primarily affected by raw material factors, product factors, and operational factors in steelmaking. The converter gas recovery conditions are mainly influenced by the CO and O2 concentrations during the blowing stage. For steel enterprises, the recovery conditions remain essentially constant; the amount of converter gas that meets these conditions is the feasible amount to be recovered.

[0004] Currently, most predictions of converter gas generation are based solely on historical data. Chinese patent application CN202310626080.0 discloses a method for predicting converter gas generation, which extracts the intermittent duration features from the original converter gas generation data and classifies the data according to these features. Subsequently, based on the generation data features obtained from the intermittent classification, a CPSO-Elman generation prediction model and an Elman intermittent duration prediction model are established. Chinese patent application CN201410066954.2 presents a long-term prediction method for converter gas generation in metallurgical enterprises based on steelmaking rhythm estimation. This method uses a modulus matching method to extract the time-domain and frequency-domain features of the steelmaking rhythm from the on-site converter gas generation data. Then, an improved fuzzy C-means clustering method is used to fuse the data, and finally, a long-term predicted value for converter gas generation is constructed. The above prediction method only uses a data-driven approach and does not consider the impact of converter process parameters (raw materials, products, and operations) on converter gas generation. Furthermore, the CO concentration variation of converter gas differs in the initial, middle, and final stages of blowing, resulting in different recoverable gas zones, which in turn affects the amount of converter gas recovered.

[0005] Some methods for predicting coal gas generation take into account the impact of CO concentration on coal gas recovery during the blowing stage. For example, Chinese patent application CN202210589223.0 discloses a method and system for regulating converter gas generation and supply based on gas holder location. Based on historical field data, it fits characteristic curves of converter flue gas flow rate and CO and O2 concentrations as a function of blowing time, thereby predicting the converter gas recovery rate. Chinese patent application CN201110070892.9 discloses a method for predicting by-product coal gas generation based on a grey multi-factor MGM(l,n) model using principal component analysis. It performs principal component analysis on the converter gas generation and flue gas components (CO flow rate, CO2 flow rate, N2 flow rate, and O2 consumption) for the early, middle, and late stages of converter gas generation. Then, based on the obtained principal components and the converter gas generation, it establishes an MGM(l,3) prediction model for converter gas generation to predict its generation. The above prediction method takes into account the different components of converter gas in different blowing stages, but does not take into account the impact of raw materials, products and operating conditions of converter smelting on the amount and composition of converter gas.

[0006] Chinese patent application CN202310024846.8 discloses a method for predicting converter gas generation based on material carbon balance. Through the carbon balance theory in converter production, a method for predicting converter gas generation based on converter production plan is established. This method considers the raw material and product factors of converter production, but does not consider the influence of operating conditions such as hood height, fan speed, and annular gap opening on converter gas generation. The converter recovery time is simplified to a fixed value. Summary of the Invention

[0007] 1. The problem to be solved

[0008] To achieve accurate prediction of the feasible recovery amount of converter gas, this invention proposes a method for predicting the feasible recovery amount of converter gas based on multi-source heterogeneous data. This method obtains the influencing factors (raw materials, products, and operations) of converter gas recovery amount through converter gas generation mechanism analysis. Based on the primary and secondary data from the field, the data is preprocessed and fused into a dataset for predicting the feasible recovery amount of converter gas. A model for predicting the feasible recovery amount of converter gas is established using a neural network method. Combined with the converter production plan, the feasible recovery amount of converter gas is predicted, providing a basis for optimizing the dynamic balance scheduling of converter gas.

[0009] 2. Technical Solution

[0010] To solve the above problems, the technical solution adopted by the present invention is as follows:

[0011] The method for predicting the feasible recovery of converter gas based on multi-source heterogeneous data proposed in this invention has the following specific implementation steps:

[0012] Step 1: Through the analysis of converter gas generation mechanism, establish a converter gas energy-quality balance model to obtain the raw materials, products, operation and recovery influencing factors for feasible converter gas recovery.

[0013] Step 2: Based on the above factors, collect the corresponding primary and secondary machine data respectively, and preprocess the primary machine data into a dataset based on furnaces;

[0014] Step 3: Based on the blowing time in the primary and secondary data, merge the primary and secondary data to form a dataset for predicting the feasible recovery of converter gas;

[0015] Step 4: Using the neural network prediction method, take the above-mentioned influencing factors as the input layer and the feasible recovery amount of coal gas as the output layer, and use the above-mentioned prediction dataset to establish a prediction model for the feasible recovery amount of coal gas.

[0016] Step 5: Collect signals such as the steel grade being smelted in the converter, the start time of blowing, and the end time of blowing from the three-stage machine data of the steelmaking production plan to determine the converter production process type. Based on the corresponding operating factors of this type of converter production, use the prediction model to accurately predict the feasible recovery amount of converter gas.

[0017] The present invention provides a method for predicting the feasible recovery of converter gas based on multi-source heterogeneous data, comprising the following steps:

[0018] Step 1: Analysis of Influencing Factors

[0019] Based on the analysis of the converter gas generation mechanism, a converter gas mass conservation model and a converter heat balance model are established:

[0020] Converter gas recovery rate V gas,rec It is mainly affected by the amount of converter flue gas generated, the conditions for converter gas recovery, and the position of the gas holder.

[0021] The converter flue gas generation includes the theoretical furnace gas generation V. Fgas and the amount of cold air drawn in from the furnace mouth V air N2 in the middle.

[0022]

[0023] Furnace gas generation V Fgas The main process involves the oxidation reaction between carbon (C) in the molten iron and oxygen (O2) during blowing, producing CO and CO2. Based on the converter heat balance, it can be known that:

[0024]

[0025] Among them, Q iron =m iron ×[c ironS ×(t ironfreeze -tenvir )+Q ironMelt +c ironL ×(t iron -t ironfreeze )]

[0026] Q oxi =29177m Si +6593m Mn +11637m C(co) +34824m C(co2)

[0027] +35874m P +1620m SiO2 +4249m Fe ( FeO) +6459m Fe ( Fe2O3)

[0028] Q steel =m steel ×[c steelS ×(t steelfreeze -t envir )+Q steelMelt +c steelL ×(t steel -t steelfreeze )]

[0029] Q slag =m slag ×[c slagS ×(t slag -t envir )+Q slagMelt ].

[0030] Cold air intake V at the furnace opening air It is mainly affected by the furnace inlet pressure and the dust removal system. Based on the pressure balance of the OG system, it can be known that:

[0031]

[0032] Where, P0 = P2 - (ΔP) Vfp +ΔP twr +ΔP vntr +ΔP rec -P fan )

[0033] ΔP vntr =S vntr (0.78V air +V Fgas ) 2

[0034]

[0035] Comprehensive analysis shows that the converter gas generation V is mainly affected by raw material factors (amount of molten iron charged, amount of scrap steel charged, amount of raw materials fed, temperature of molten iron, C content of molten iron, Si content of molten iron, Mn content of molten iron, and P content of molten iron), product factors (final steel temperature, C content of molten steel, P content of molten steel, Si content of molten steel, and Mn content of molten steel), and operating factors (height of movable fume hood, opening of circumferential gap, and fan speed).

[0036] The feasible recovery amount of converter gas is also affected by the recovery conditions, namely the CO and O2 concentrations of converter gas during the blowing stage. The CO concentration of converter gas during the blowing stage is mainly related to the composition and temperature of molten iron, as well as the oxygen blowing intensity and oxygen lance height.

[0037] The second step is multi-source data acquisition and preprocessing.

[0038] The multi-source data includes time-series data from the primary furnace, data from the secondary furnace at the furnace-by-furnace level, and production plan data from the tertiary furnace. Specifically, the raw material and product factors are collected from the secondary furnace, while the operational and recovery conditions are time-series data that fluctuate during the smelting process; these are collected from the primary furnace. The tertiary furnace data, based on the converter blowing plan, supports subsequent model predictions.

[0039] Due to the different data structures, the primary machine data needs to be preprocessed. The main process involves extracting the blowing period based on the O2 flow rate of the primary machine data, and then extracting the relevant operational factors and gas flow data for the feasible recovery period based on the CO and O2 concentration recovery limits, thus forming the primary machine training dataset.

[0040] The third step is data fusion to form a dataset for predicting feasible gas recovery.

[0041] The prediction dataset is derived from the preprocessed primary machine training set and secondary machine dataset mentioned above. Based on the start and end times of blowing in the secondary machine data, the primary machine dataset for the corresponding time period is extracted. Then, the gas flow rate for that time period is summed, and other operating factors such as hood height, fan speed, and annular gap opening are averaged. The results are then fused with the corresponding secondary machine dataset to form the gas feasible recovery prediction dataset.

[0042] The fourth step is to establish a predictive model for the feasible recovery of coal gas.

[0043] The factors influencing the feasible recovery of coal gas analyzed above (such as the weight of molten iron, the weight of scrap steel, the carbon content of molten iron, the carbon content of molten steel, the total weight of feed, the height of the fume hood, the opening of the circumferential joint, and the fan speed) are used as the input layer of the neural network prediction model, and the feasible recovery of coal gas is used as the output layer. Neural network structures such as LM-BP, RBF, and CNN are adopted. The model is trained using the feasible recovery of coal gas prediction dataset established above to establish a feasible recovery of coal gas prediction model.

[0044] The fifth step is to accurately predict the feasible amount of coal gas to be recovered by combining it with the production plan.

[0045] Information such as the start and end times of blowing and the type of steel being smelted in the production plan of the three-stage converter steelmaking is read. Combined with the operational experience data of smelting this type of steel, a prediction model input dataset is constructed. Using the gas feasible recovery prediction model established above, the gas feasible recovery amount for this furnace is predicted. Then, the gas flow rate is calculated based on the blowing period.

[0046] This invention relates to technologies such as theoretical analysis of factors influencing converter gas generation and recovery, fusion of multi-source heterogeneous data, and neural network prediction of feasible converter gas recovery. It is a method for predicting feasible converter gas recovery based on multi-source heterogeneous data. This invention derives the influencing factors on feasible converter gas recovery through converter gas generation mechanism and heat-mass balance analysis. Then, using primary and secondary field historical data (multi-source heterogeneous data), the multi-source structures are fused as a historical dataset for deep learning. Neural network methods such as LM-BP and RBF are used to predict feasible recovery. Finally, the feasible converter recovery is predicted in conjunction with the three-stage converter production plan. The prediction results can effectively guide converter gas balance scheduling and can also be used to analyze the impact of production plans on feasible converter gas recovery.

[0047] 3. Beneficial effects

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] This invention analyzes the generation mechanism of converter steelmaking gas and the theory of heat and mass balance. Combining multi-source heterogeneous data such as time series data of primary unit, production data of secondary unit, and production plan of tertiary unit, it establishes a method for predicting the feasible recovery of converter gas using neural network prediction. This method takes into account the influence of multiple factors such as raw materials, products, operation, and recovery in the converter production process, and significantly improves the prediction accuracy of converter gas, providing technical support for gas balance scheduling. Attached Figure Description

[0050] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that these drawings are designed for illustrative purposes only and are not intended to limit the scope of the present invention. Furthermore, unless specifically indicated, these drawings are intended only to conceptually illustrate the structural construction described herein and are not necessarily drawn to scale.

[0051] Figure 1 This is a flowchart of the method for predicting the feasible recovery of converter gas based on multi-source heterogeneous data according to the present invention;

[0052] Figure 2The prediction method of this invention is shown at the interface of the primary unit of a converter gas system in a steel plant;

[0053] Figure 3 This is a schematic diagram of the primary machine data processing in the prediction method of the present invention;

[0054] Figure 4 The prediction results (training set) of the BP neural network method used in Embodiment 1 of the present invention are shown.

[0055] Figure 5 The prediction results (test set) of the BP neural network method used in Embodiment 1 of the present invention are shown.

[0056] Figure 6 This is the gas flow prediction result for three furnace cycles in conjunction with the production plan in Embodiment 1 of the present invention;

[0057] Figure 7 This invention provides an example of a comparison between the predicted and actual total gas recovery flow rate of the production plan in Embodiment 1 of the present invention.

[0058] Figure 8 This is the data interface of a three-stage machine in a steel plant according to Embodiment 2 of the present invention;

[0059] Figure 9 This is the result of the feasible gas recovery flow prediction in conjunction with the production plan in Embodiment 2 of the present invention;

[0060] Figure 10 This is a comparison of the feasible recovery amount of converter gas under different recovery start and end concentrations using the prediction method in Example 3 of the present invention.

[0061] Figure 11 The results show the predicted flow rates of converter gas with different start and end recovery concentrations in Example 3 of this invention. Detailed Implementation

[0062] Exemplary embodiments of the present invention are described in detail below. While these exemplary embodiments have been described in sufficient detail to enable those skilled in the art to practice the invention, it should be understood that other embodiments may be implemented and various changes may be made to the invention without departing from its spirit and scope. The more detailed description of embodiments of the invention below is not intended to limit the scope of the claimed invention, but is merely illustrative and does not limit the description of the features and characteristics of the invention, in order to suggest the best mode for carrying out the invention and to enable those skilled in the art to practice it. Therefore, the scope of the invention is defined only by the appended claims.

[0063] The proposed method for predicting the feasible recovery of converter gas based on multi-source heterogeneous data, as described in this invention, includes the following specific implementation steps: Figure 1 As shown:

[0064] Step 1: Through the analysis of converter gas generation mechanism, establish a converter gas energy-quality balance model to obtain the raw materials, products, operation and recovery influencing factors for feasible converter gas recovery.

[0065] Step 2: Based on the above factors, collect the corresponding primary and secondary machine data respectively, and preprocess the primary machine data into a dataset based on furnaces;

[0066] Step 3: Based on the blowing time in the primary and secondary data, merge the primary and secondary data to form a dataset for predicting the feasible recovery of converter gas;

[0067] Step 4: Using the neural network prediction method, take the above-mentioned influencing factors as the input layer and the feasible recovery amount of coal gas as the output layer, and use the above-mentioned prediction dataset to establish a prediction model for the feasible recovery amount of coal gas.

[0068] Step 5: Collect signals such as the steel grade being smelted in the converter, the start time of blowing, and the end time of blowing from the three-stage machine data of the steelmaking production plan to determine the converter production process type. Based on the corresponding operating factors of this type of converter production, use the prediction model to accurately predict the feasible recovery amount of converter gas.

[0069] Example 1

[0070] Taking the converter production of a steel plant A as an example, the method for predicting the feasible recovery of converter gas of the present invention will be described in detail.

[0071] Step 1: By analyzing the converter gas generation mechanism, the converter gas generation mainly consists of two parts: the gas generated by the oxidation of carbon in molten iron during the converter blowing stage and the amount of air drawn in from the furnace mouth and fume hood gaps. From the perspective of converter heat balance, the influencing factors on converter gas generation mainly include: molten iron charge, scrap steel charge, total charge weight, molten iron temperature, carbon content, silicon content, manganese content, phosphorus content, final steel temperature, carbon content, silicon content, manganese content, and phosphorus content in the molten steel.

[0072] The amount of air drawn in from the furnace opening and the gap in the fume hood is mainly affected by the dust removal system and the furnace pressure. Taking the OG dust removal system as an example, this invention analyzes the pressure balance of the OG system and finds that the factors affecting the amount of air mainly include: the height of the movable fume hood, the circumference of the fume hood, the opening of the circumferential gap, the fan speed, and the furnace gas production rate.

[0073] The feasible recovery rate of coal gas is affected not only by the amount of coal gas produced but also by the conditions for coal gas recovery. Analysis of the characteristics of CO and O2 changes in coal gas at different stages of blowing shows that the CO concentration change during the blowing stage is mainly related to the C, Si, Mn, and P content in the molten iron, the temperature of the molten iron, the oxygen blowing intensity, and the height of the oxygen lance.

[0074] Comprehensive analysis shows that the feasible recovery of converter gas is mainly affected by factors such as the amount of molten iron charged, the amount of scrap steel charged, the total weight of the feed, the temperature of the molten iron, the carbon content, silicon content, Mn content, and phosphorus content of the molten iron, the final steel temperature, the carbon content, silicon content, Mn content, and phosphorus content of the molten steel, the oxygen blowing intensity, the height of the oxygen lance, the height of the movable fume hood, the circumference of the fume hood, the opening of the circumferential slit, and the fan speed.

[0075] Step 2: The above-analyzed factors—hot metal charge, scrap steel charge, total feed weight, hot metal temperature, C content, Si content, Mn content, and P content—are raw material factors; the final steel temperature, C content, Si content, Mn content, and P content are product factors. These influencing factors can all be collected from the secondary machine database, as shown in Table 1. This dataset is based on the furnace unit. Oxygen blowing intensity, oxygen lance height, movable fume hood height, fume hood circumference, circumferential gap opening, and blower speed are operational factors (see...). Figure 2 The data set is a time-series dataset, where the perimeter of the smoke hood is a constant for the site, while the rest of the data change over time and need to be collected from the primary database. As shown in Table 2, the dataset is a time-series dataset.

[0076] Table 1 shows the collected data from the converter gas secondary stage machine.

[0077]

[0078] Table 2 shows the collected converter gas primary unit dataset.

[0079]

[0080] Comparing the data in Tables 1 and 2 reveals that the data structures are different, necessitating preprocessing of the primary machine data. For example... Figure 3 As shown in the figure, the variations in O2 flow rate, gas flow rate, CO concentration, and O2 concentration during converter blowing are illustrated. The vertical axis represents the values ​​of these four variables, with units of km. 3 / h, km 3 / h, % and %. From Figure 3 It can be seen that the above-mentioned operational factor data collected from the primary unit can be extracted into a furnace sequence based on O2 flow rate (see...). Figure 3 The vertical dashed line represents the portion where O2 > 0. Then, based on the CO and O2 concentrations, feasible recovery periods are determined (see...). Figure 3 (The section where the horizontal dashed line intersects with the CO concentration line at 30%), the flue gas flow rate and oxygen flow rate are weighted and summed, and the height of the fume hood, the opening of the annular gap, the height of the oxygen lance, and the fan speed are taken as the average value of this interval.

[0081] Step 3: Based on the start time of the blowing process, couple the secondary machine dataset and the processed primary machine dataset to form a dataset for predicting the feasible recovery of converter gas, as shown in Table 3.

[0082] Table 3. Feasible recovery prediction dataset after data fusion

[0083]

[0084] Step 4: Using columns 4 to (n-1) of the dataset (n columns in total) as model input and column n as model output, a feasible model for converter gas recovery is established using the LM-BP neural network method. Taking one month of operation data from a converter in a steel plant as an example, after data preprocessing, a total of 531 heats were collected. This data was randomly shuffled and divided into two groups: 425 groups were used as the training set for model training, and the remaining 106 groups were used as validation test samples for model validation. The prediction results for the training and prediction sets are shown below. Figure 4 and Figure 5 As shown, the model errors MAPE of the training set and prediction set are 1.80% and 1.76%, respectively, and the model accuracies are 98.2% and 98.24%, respectively, indicating high model accuracy.

[0085] Step 5: Collect signals such as the steelmaking blowing plan number, steel grade, blowing start time, blowing end time, and molten iron charging amount from the steelmaking plant's three-stage mill production plan. Based on the steel grade, determine the feed amount, scrap steel charging amount, as well as data such as hood height and circumferential gap opening to form the model input. After inputting these data into the trained prediction model, predict the feasible recovery amount of the converter. Combine this with the blowing period to calculate the converter gas flow rate during the recoverable period.

[0086] Table 4 shows the converter blowing plan of the steel plant during the period from 2:35 to 3:10 on a certain day. It is used into the converter gas recovery prediction model to obtain the feasible gas recovery amount of 3 furnaces. The comparison between the prediction results and the actual values ​​is shown in Table 5.

[0087] Table 4. Converter Steelmaking Production Plan

[0088] First furnace Second furnace Third furnace Time to blow 2:35 2:41 2:57 Steel tapping moment 2:54 3:05 3:12

[0089] Table 5 Comparison of Predicted and Actual Values ​​of Feasible Recoverable Gas Volume from Three Furnaces

[0090] First furnace Second furnace Third furnace <![CDATA[Predicted value (m 3 )]]> 49449.5 40685.8 43445.5 <![CDATA[Actual value (m 3 )]]> 49981.5 40360.8 43514.9 Difference -532 325 -69.4 error% -1.06％ 0.81％ -0.16％

[0091] By averaging the predicted feasible recovery rates for each furnace over the feasible recovery period, the converter gas flow rate for the feasible recovery period can be obtained. Figure 6 It can be seen that the flow rate during the feasible recovery period for a single furnace is 220-270 km³. 3The actual value of the feasible recovery amount for the first furnace fluctuated slightly around the predicted value, but the difference was not significant. The feasible recovery period for the second and third furnaces differed slightly, and the flow rate fluctuation within this period was not significant. The actual flow rate basically coincided with the predicted flow rate line, indicating that the flow rate method for feasible recovery of converter gas proposed in this invention has high accuracy.

[0092] Figure 7 The figure presents a comparison between the predicted and actual values ​​of the gas recovery flow rate for the planned production period. As can be seen from the figure, during this planned production period, there is a period where two furnaces are simultaneously blowing gas, and the gas flow rate during this period is approximately 500 km³. 3 / h, within the predicted period, the actual gas flow rate is very close to the predicted value, basically overlapping, proving that the prediction method is highly accurate, and the predicted converter gas recovery flow rate can effectively guide the control of converter gas and the entire gas process.

[0093] Example 2

[0094] In this embodiment, different neural network prediction methods can be used in the feasible recovery prediction method. In Example 1, the traditional BP neural network prediction method is not used; instead, an RBF radial basis function network is used. Taking the converter production of a steel plant B as an example, the feasible recovery prediction method for converter gas of this invention will be described in detail.

[0095] Step 1: Similar to Example 1, through the analysis of converter gas generation mechanism and pressure balance analysis of converter dust removal system, the influencing factors of converter gas generation are obtained: 1) Raw material factors—hot iron charging amount, scrap steel charging amount, total weight of materials, hot iron temperature, hot iron C content, hot iron Si content, hot iron Mn content, hot iron P content; 2) Product factors—final steel temperature, hot steel C content, hot steel Si content, hot steel Mn content, hot steel P content; 3) Operational factors—movable fume hood height, fume hood circumference, circumferential gap opening, fan speed.

[0096] The feasible recovery rate of converter gas is also constrained by the gas recovery conditions: namely, the starting and ending concentration limits of CO and O2 recovery during the blowing stage. The characteristics of CO concentration changes during the blowing process are mainly affected by operational factors—oxygen blowing intensity and oxygen lance height.

[0097] In summary, the feasible recovery amount of converter gas is mainly affected by the aforementioned raw material factors, product factors, operational factors, and recovery conditions.

[0098] Step 2: Similar to Example 1, collect the above-mentioned operational factors and time-series factors such as CO concentration, O2 concentration and gas flow rate from the primary unit on site, and convert them into a primary unit dataset by furnace number according to the method of Example 1.

[0099] Step 3: Based on the start time of the blowing process, couple the pre-treated secondary machine dataset and the converted primary machine dataset to form a converter gas feasible recovery prediction dataset for model training and model validation.

[0100] Step 4: Taking the converter's data from August to September 2024 as an example, the dataset organized using the above method is a 1062×19 matrix. After randomizing the data, it is divided into a training set and a test / validation set, with 850 groups and 212 groups respectively. The model uses a radial basis function (RBF) neural network. The first 18 columns of the dataset contain the aforementioned influencing factors, serving as the input layer neurons of the RBF neural network. There is one hidden layer with a Gaussian kernel function to perform spatial mapping transformation on the input information. The 19th column represents the feasible recovery amount of converter gas, serving as the output layer of the model. The mean absolute percentage error (MAPE) is used as the error function. Calculations show that the MAPE of the feasible recovery amount for the converter in the training and prediction sets are 1.53% and 1.90% respectively, with model accuracies of 98.47% and 98.1%, indicating very high model accuracy.

[0101] Examples 1 and 2 demonstrate that the converter gas feasible recovery prediction method based on multi-source heterogeneous data proposed in this invention can achieve high model accuracy by using different neural network prediction models.

[0102] Step 5: Read the production plan of the converter stage 3, including the plan number, steel grade, converter start time, converter end time, and planned steel output. Based on historical data, establish a model mean library for each steel grade, including data on feed rate, scrap steel loading, hood height, and circumferential gap opening. Using the planned steel grade, read the parameters from the corresponding model mean library, along with the molten iron quantity, temperature, and composition processed by KR, as model inputs. Substitute these into the prediction model to predict the feasible recovery amount for the corresponding plan number. Combine this with the blowing period to calculate the converter gas flow rate during the recoverable period.

[0103] Figure 8 This is the steelmaking schedule for a set of three-stage mills in the steel plant. The blowing start time for this heat is 08:49, and the converter steelmaking end time is 09:18. Based on the steel grade produced by the three-stage mills (see...), Figure 8 (The section within the box) Using the mean library and data on molten iron quantity, temperature, and C, Si, Mn, and P content read from the KR primary machine, the predictive model forecasts a feasible converter gas recovery rate of 44324.7 m³. 3 If this is averaged over the recoverable gas flow period during the blowing process (08:50-09:01), the recovered gas flow rate for that period is 221.6 km³. 3 / h, see comparison with actual results. Figure 9 As shown. From Figure 9It can be seen that the gas flow prediction accuracy during the feasible recovery period is very high, which can effectively guide on-site gas regulation and dispatch.

[0104] Example 3

[0105] The converter gas feasible recovery prediction method of this embodiment can be used to predict the feasible recovery amount of gas under different recovery conditions with the same raw materials, products and operating conditions. Taking the converter production of a steel plant C as an example, the converter gas feasible recovery prediction method of this invention will be described in detail.

[0106] Step 1: Similar to Examples 1 and 2, through the analysis of converter gas generation mechanism and the pressure balance analysis of converter dust removal system, a total of 16 factors are obtained regarding the raw material factors, product factors, and operational factors of converter gas generation.

[0107] In addition, the feasible amount of converter gas to be recovered is also constrained by the gas recovery conditions: namely, the starting and ending concentration limits of CO and O2 recovery during the blowing stage. Enterprises can adjust the starting and ending CO recovery concentrations during the blowing stage according to their own needs to adjust the amount of converter gas recovered and the calorific value of the recovered gas.

[0108] Step 2: Similar to Example 2, the above 13 raw material and product factors are selected from the secondary machine database, and the above 3 operating factors, oxygen flow rate, oxygen lance height, CO concentration, and O2 concentration are selected from the primary machine database. Using the same method, the primary machine database is transformed, converting the time-series data into data per furnace. The feasible recovery amount of converter gas can be obtained by extracting data based on different start and end recovery concentration limits.

[0109] Step 3: Taking converter C of a steel plant as an example, data from June 2024 was selected. Following the method described above, the primary unit data was processed and merged with the secondary unit data according to the start time of blowing, forming a dataset for predicting the feasible recovery of converter gas, totaling 500 data points. The feasible recovery of converter gas was extracted using different CO concentration limits: 25%, 30%, 35%, and 40%. Figure 10 Comparative results of feasible converter gas recovery rates under 20 different CO concentration limits are presented. It can be seen that, under the same recovery conditions, the feasible recovery rates vary for each furnace, with a range of 40,000 m³ / s. 3 up to 55000m 3 Fluctuations. In the same furnace, when the CO concentration increases from 25% to 40%, the decrease in the feasible recovery amount of converter gas also varies. Therefore, if it is necessary to analyze the impact of different recovery conditions on the feasible recovery amount of converter gas, it is necessary to predict them separately.

[0110] Step 4: Randomly shuffle the dataset prepared in Step 3 and divide it into a training set (400 groups) and a test / validation set (100 groups). The model uses a radial basis function (RBF) neural network, and the mean absolute percentage error (MAPE) is selected as the error function. Calculations show that when the starting and ending recovery concentrations are 25%, 30%, 35%, and 40%, the MAPE of the model on the test set are 2.49%, 2.59%, 3.5%, and 4.17%, respectively, all demonstrating high accuracy and proving the model's effectiveness.

[0111] Step 5: Read the blowing plan of the converter in the three-stage machine. This time, the selected heat is the one with a blowing start time of 14:44 and a blowing end time of 15:00. Similar to Example 2, after calling the steel grade average database and the relevant parameters of the molten iron from the KR primary machine, the feasible recovery flow rate of converter gas under different starting and ending recovery concentrations is predicted using the prediction model established above with CO start and end concentrations of 25% and 35%. The prediction results are as follows: Figure 11 As shown. From Figure 11 It can be seen that the feasible recovery period varies depending on the CO start and end recovery concentrations for the same furnace, resulting in different feasible recovery amounts. Using the prediction method of this invention, the predicted feasible recovery flow rates for different CO start and end recovery concentrations are highly accurate, effectively guiding on-site adjustments to the recovered gas volume and calorific value by modifying recovery conditions.

[0112] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A converter gas feasible recovery amount prediction method based on multi-source heterogeneous data, characterized in that, The method comprises the following steps: Through converter gas mechanism analysis, a converter gas energy and quality balance model is established, and raw material, product, operation and recovery influencing factors of the feasible converter gas recovery amount are obtained; Combined with the influencing factors, corresponding first-stage machine data and second-stage machine data are collected, and the first-stage machine data is preprocessed into a data set in units of furnaces; According to the blowing time in the first-stage machine data and the second-stage machine data, the first-stage machine data and the second-stage machine data are fused to form a converter gas feasible recovery amount prediction data set; Using a neural network prediction method, the influencing factors are taken as an input layer, the converter gas feasible recovery amount is taken as an output layer, the prediction data set is used, and a converter gas feasible recovery amount prediction model is established; The converter steelmaking production plan three-stage machine data is collected, converter smelting steel grades, blowing start time and blowing end time signals are determined, a converter production process type is determined, corresponding operation factors of the converter production are determined according to the type, the prediction model is used, and accurate prediction of the converter gas feasible recovery amount is realized.

2. The converter gas feasible recovery amount prediction method based on multi-source heterogeneous data according to claim 1, characterized in that, The specific steps are as follows: First step, influencing factor analysis Through converter gas mechanism analysis, a converter gas quality conservation model and a converter heat balance model are established, and the main influencing factors of the converter gas generation amount V are analyzed and obtained; Second step, multi-source data collection and preprocessing According to the first step of the main influencing factor analysis, it is known that the raw material factors and product factors affecting the converter gas feasible recovery amount are in units of furnaces, and the field data need to be collected from the second-stage machine; and the operation factors and recovery condition factors are time series data collected in the first-stage machine; According to the O2 flow rate of the first-stage machine data, the blowing period is intercepted, then the CO and O2 concentration recovery limit values are used to intercept the operation factors and gas flow data in the feasible recovery period, and a first-stage machine training data set is formed; Third step, data fusion to form a converter gas feasible recovery amount prediction data set According to the blowing start time and the blowing end time of the second-stage machine data, the first-stage machine data set in the corresponding period is intercepted, then the gas flow in the period is summed, the remaining operation factors such as the hood height, the fan speed and the ring gap opening are processed by averaging, and the processing results are fused with the corresponding second-stage machine data set to form a converter gas feasible recovery amount prediction data set; Fourth step, establishment of a converter gas feasible recovery amount prediction model The influencing factors of the third step are taken as the input layer of the neural network prediction model, the converter gas feasible recovery amount is taken as the output layer, a neural network structure is adopted, the converter gas feasible recovery amount prediction data set established above is used for model training, and a converter gas feasible recovery amount prediction model is established; Fifth step, accurate prediction of the converter gas feasible recovery amount combined with the production plan The blowing start time, the blowing end time and the smelting steel grade information in the converter steelmaking production plan are read, the prediction model input data set is formed by combining the operation experience data of smelting the steel grade, the converter gas feasible recovery amount prediction model established in the fourth step is used, the converter gas feasible recovery amount of the furnace is predicted, and then the gas flow is calculated according to the blowing period. 3.The converter gas feasible recovery amount prediction method based on multi-source heterogeneous data according to claim 2, characterized in that, Converter gas recovery amount V gas,rec The converter gas recovery amount V is mainly affected by the converter flue gas generation amount, the converter gas recovery conditions, and the gas tank position. The converter flue gas generation amount includes the theoretical converter gas generation amount V Fgas and the N2 in the cold air suction amount V air at the converter mouth. Amount of furnace gas generated V Fgas The main reaction is the oxidation of carbon in the hot metal with the blown O2 to produce CO and CO2. Based on the heat balance of the converter, it is known that: wherein, ， Cold air intake volume at the furnace mouth V air Mainly affected by the furnace mouth pressure and dust removal system, based on the OG system pressure balance: wherein, 。 4. The converter gas feasible recovery amount prediction method based on multi-source heterogeneous data according to claim 2, characterized in that, The converter gas feasible recovery amount influencing factors include molten iron weight, scrap steel weight, molten iron C content, molten steel C content, total charge weight, hood height, ring gap opening and fan speed. 5.The converter gas feasible recovery amount prediction method based on multi-source heterogeneous data according to claim 2, characterized in that, The neural network structure includes LM-BP, RBF and CNN.

Images