Method and device for automatic roasting and process intelligent control of belt roaster

Abstract

The application discloses a kind of automatic roasting and process intelligent control method and device of belt roaster, it is related to the technical field of pellet roasting of belt roaster in metallurgical iron and steel industry.The method includes four steps of historical data slice acquisition and modeling, target process parameter determination, feedback check and online correction, rolling prediction control, establishes model history library through business slice and time slice, is modeled based on three-layer feedforward neuron network, is combined with the flexible weighted deviation calculation of dynamic weight and the rolling optimization of softening coefficient, realizes equipment parameter accurate regulation and control.It is mainly used for the automatic control of belt roaster pellet production process, improves product quality stability, reduces energy consumption, reduces manual labor intensity, provides support for production unmanned, intelligent.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pellet induration of a belt induration machine in the metallurgical iron and steel industry. More specifically, the present application relates to a method and device for automatic induration and intelligent control of the process of a belt induration machine. BACKGROUND

[0002] In the metallurgical iron and steel industry, reducing the energy consumption of the ironmaking process and reducing carbon emissions is an inevitable trend, and increasing the blast furnace pellet ratio (i.e. the proportion of pellet ore in the blast furnace ironmaking charge) is a key means. Pellet ore has good reducibility and fewer harmful impurities, and increasing its ratio can significantly reduce the consumption of coke in blast furnace ironmaking, thereby reducing energy consumption and carbon emissions. Based on this, the current main direction of ironmaking industry structure adjustment at home and abroad is two paths: one is to adopt high-proportion pellet blast furnace ironmaking technology, and the other is to develop global pellet DRI preparation technology to produce direct reduced iron using pellet ore as raw material. The carbon emissions of this process are much lower than those of traditional blast furnace ironmaking.

[0003] The belt induration machine is the core equipment supporting the above two technical paths, and its core function is to complete the full-process conversion from green balls to pellet ore. Through drying, preheating, induration, cooling and other sections, the green balls made of iron ore powder are processed into pellet ore with high strength and excellent metallurgical properties.

[0004] To achieve stable and efficient operation of the belt induration machine, it is necessary to rely on automatic induration and intelligent control technology, and the core support of this technology is the belt induration machine balance model. It is necessary to base on accurate actual production data, comprehensively calculate the mass balance, heat balance and pressure balance of each process section of the equipment, and achieve precise control of equipment parameters accordingly. Among them, heat balance is the core basis for establishing the model, and the analysis of heat balance needs to be based on material balance. It needs to consider key factors such as heat change of green balls, gas combustion heating, water evaporation heat absorption, oxidation reaction heat release, carbonate decomposition heat absorption, equipment surface heat loss and flue gas circulation, and clearly understand the operation rules of hot air in the system, which is the key to accurate analysis of heat balance. Through thermal process detection and energy balance analysis of the belt induration machine system, the energy distribution and utilization level of the system can be scientifically evaluated, and equipment improvement suggestions and operation parameter adjustment schemes can be proposed, so as to realize economic operation of the system and reduce energy consumption.

[0005] However, the belt induration machine is a high-temperature closed "black box", and the induration process state cannot be described by a single parameter. It is necessary to consider the state of each process stage respectively. If only a single state at different positions is considered at the same time, different control results will occur. It is impossible to analyze the reason by relying on the state of only one condition. If control is only based on the state of one factor, it is easy to make mistakes. Therefore, through big data analysis and process intelligent control, the states of all positions are comprehensively analyzed to find the root cause of the problem and effective measures, so as to realize comprehensive consideration and overall coordination of control parameters.

[0006] Further, the pellet roasting production itself is a complex dynamic system with many influencing factors and large lag, and the roasting process is a typical complex controlled object with multivariable, nonlinear, and strong coupling characteristics. Due to the dynamic complexity and time-varying characteristics of the roasting process, the occurrence of an abnormal phenomenon is inevitably accompanied by fluctuations or abnormalities in multiple monitoring data; in order to narrow the state solving space and feature space of the abnormal diagnosis and improve the solving efficiency, the frequently occurring or major abnormality in the pellet roasting production is defined as a typical abnormality, and only the parameters related to these typical abnormalities are extracted during the diagnosis process. The selection of such characteristic parameters is mainly based on the experience of skilled operators on site, combined with the control ideas and methods of stable and balanced pellet roasting.

[0007] However, in the face of such a complex production process and diagnosis requirement, the traditional control method has shown its inadequacy: with the expansion of industrial production scale and the increase of complexity, the requirements for key product quality, process optimal solution and real-time control are constantly improving. The simple PID control system and offline historical data expert library system cannot meet the production needs of the belt roaster. At present, it is more necessary to realize the multivariable model predictive control of the optimal parameter solution in the production process control by taking system identification and model calculation as the core and taking big data analysis and factory control as the carrier, and then to achieve the intelligent and stable production automatic control with the optimal product and the lowest energy consumption through a method and device for automatic roasting and intelligent control of the belt roaster.

[0008] Therefore, the specific implementation direction of the automatic roasting and intelligent control of the belt roaster includes fan control, material thickness control, machine speed control, valve control, burner heat source control, negative pressure control, etc., and the ultimate goal is to automatically achieve the optimal production plan, raw material matching, process temperature, process pressure, physical / chemical performance, process energy consumption, and key process indicators such as air heat balance and material balance in the roasting process. SUMMARY

[0009] The present application provides a method for automatic roasting and intelligent control of a belt roaster, which is based on historical data modeling and dynamic adjustment, can accurately control process parameters, stabilize product quality (such as compressive strength and drum index), reduce energy consumption, cope with process disturbances, reduce manual monitoring intensity, and provide support for intelligent production.

[0010] The present application provides a device for automatic roasting and intelligent control of a belt roaster, which can accurately adjust equipment parameters, adapt to multiple working conditions, ensure production stability, optimize quality and energy consumption, and promote the development of belt roaster production towards unmanned and intelligent production by executing the above-mentioned method through each functional module.

[0011] In order to achieve the objects and other advantages according to the present application, a method for automatic roasting and process intelligent control of a belt roaster is provided, comprising:

[0012] A historical data slice collection and modeling step collects monitoring data samples in the operation process of the belt roaster, the monitoring data samples including raw material codes, slice codes, process monitoring item values, device parameter measured signal values, process energy consumption data, and product qualification rate index data, a business slice is triggered by a production event change, a time slice is triggered by a sampling interval time, a model history library is established according to the business slice and the time slice, and based on the model history library, an automatic roasting big data model is established by using a three-layer feedforward neuron network;

[0013] A target process parameter determination step queries a record with the same business slice label in the model history library according to the current raw material code and the current business slice condition, filters a product code domain divided according to the time slice if the corresponding record exists, selects a quality qualified record satisfying a pre-defined product qualification rate index from the product code domain, and extracts process monitoring item values and device parameter measured signal values of a record with the lowest process energy consumption from the quality qualified record as predicted device adjustment parameters;

[0014] A feedback checking and online correction step performs a flexible weighted deviation calculation based on dynamic weights on the current process monitoring item values and the device parameter measured signal values and the predicted device adjustment parameters, and adjusts and issues the device parameters online according to the flexible weighted deviation calculation result;

[0015] A rolling prediction control step periodically executes the target process parameter determination step and the feedback checking and online correction step of the automatic roasting big data model, estimates a future deviation value by using the prediction model, introduces a softening coefficient for rolling optimization, determines a current control strategy, and the control strategy includes adjustment of the device parameters to minimize the deviation between the controlled variables and the expected values.

[0016] Preferably, the process monitoring item values are the drum drying section cover temperature, the drying section wind box temperature, the preheating section cover temperature, the preheating section wind box temperature, the roasting section cover temperature, the roasting section wind box temperature, the soaking section wind box temperature, the second cooling section cover temperature, the drum drying section cover pressure, the drum drying section wind box pressure, the drying section cover pressure, the preheating section cover pressure, the roasting section cover pressure, the first cooling section cover backheating wind main pipe pressure, and the second cooling section cover pressure.

[0017] The measured signal values ​​of the equipment parameters are: roasting machine speed, roasting material thickness, opening of the hot air valve on the outlet pipe of the drying blower, opening of the butterfly valve in the exhaust section, opening of the butterfly valve in the preheating section, opening of the butterfly valve in the secondary cooling section, opening of the valve connecting the outlet pipe of the drying blower to the furnace hood, opening of the butterfly valve in the exhaust section, heat source flow rate of the 16 burners on each side of the roasting section, speed of the No. 1 main exhaust fan, speed of the No. 2 main exhaust fan, speed of the regenerating blower, speed of the furnace hood blower, speed of the cooling blower, and speed of the drying blower.

[0018] The energy consumption data for each process includes electricity consumption per unit of gas consumption per unit of gas and finished product quantity.

[0019] The finished product qualification rate indicators are compressive strength, drum index, and particle size.

[0020] Preferably, the business slice is triggered by at least one of the following events: change in product type, change in production specifications, change in production line status, change in proportion, change in process production status, material addition / reduction operation, and timeout. The sampling interval of the time slice is calculated by dividing the trolley length by the real-time machine speed.

[0021] Preferably, the compressive strength, drum index, and particle size indicators include: compressive strength in the range of 2400-2600N, drum index in the range of 88-92%, and particle size of 10-16mm in the range of 73-77%. The minimum process energy consumption is the sum of electricity consumption and gas consumption divided by the finished product quantity.

[0022] Preferably, when there is no record of the corresponding raw material code in the model history database, the raw material number and time slice record are added to the model history database.

[0023] Preferably, the specific process for calculating the flexible weighted deviation in the feedback verification and online correction steps is as follows:

[0024] For the i Each process monitoring item, its equipment parameter adjustment amount Δ U i Determined by the following formula:

[0025]

[0026] in, E i For the first i The deviation between the current real-time value of each process monitoring item and the corresponding target value in the predicted equipment adjustment parameters;

[0027] Δ E i For deviation E i Rate of change within a sampling period;

[0028] Ki For the first i The basic adjustment coefficients corresponding to each process monitoring item are obtained by training historical data using an automatic roasting big data model.

[0029] f ( E i ,Δ E i ) is the dynamic weight function, and its expression is:

[0030]

[0031] α and β These are preset weighting factors, with values ​​ranging from 0.1 to 0.5 and from 0.05 to 0.3, respectively. E i，max and Δ E i，max The first i The statistical maximum values ​​of the absolute value of deviation and the absolute value of the rate of change of deviation for each process monitoring item under qualified operating conditions in the model history database.

[0032] Preferably, the softening coefficient in the rolling predictive control step is... λ For dynamic adaptive variables:

[0033]

[0034] in, λ 0 is the basic flexibility coefficient, ranging from 0.8 to 1.0, and Δ is the average of the absolute values ​​of the relative deviations of several key process monitoring items. Summation range j From 1 to m , m The number of key process monitoring items, m The value range is 3-6. The key process monitoring items include the temperature inside the roasting section hood, the pressure inside the roasting section hood, and the thickness of the roasted material as mandatory items, and the temperature inside the preheating section hood, the temperature of the wind box in the homogenizing section, and the pressure inside the drying section hood as optional supplementary items. The combination of mandatory items and optional supplementary items constitutes the key process monitoring items.

[0035] k ( σ ) is the dynamic adjustment coefficient related to the volatility dispersion. ,in C The preset weighting coefficients range from 0.03 to 0.08. σ Let be the composite standard deviation of the relative deviations of m key process monitoring items over the most recent 10-30 sampling periods. The composite standard deviation is the arithmetic mean of the standard deviations of the relative deviations of each key process monitoring item.σ ref is the reference standard deviation, which is obtained by querying the model history library for the 90th percentile of the comprehensive standard deviation of the relative deviation of the key process monitoring items under all qualified working conditions.

[0036] Preferably, the control strategy further comprises balancing the pressures of the drum dry section and the exhaust dry section:

[0037] By adjusting the valve opening degree of the drum dry fan outlet pipeline and the communication valve with the furnace cover and the rotation speed of the second main exhaust fan, the ratio of the actual measured value of the drum dry section cover pressure P 鼓干段 to the actual measured value of the exhaust dry section cover pressure P 抽干段 is stabilized within the preset balance interval D ; D min , D max ;

[0038] Wherein, according to the current raw material code and the current business slice condition, the model history library with the same business slice label is queried to obtain the quality qualified record data segment that meets the pre-defined finished product qualified rate index, the target data segment with the lowest process energy consumption is extracted, the value distribution of the target data segment is counted, and the value range with a frequency of ≥90% is taken as the initial D ; D D min , D max .

[0039] The automatic roasting and process intelligent control device of the belt roaster comprises:

[0040] A historical data slice acquisition and modeling module is used to acquire monitoring data samples, which include raw material codes, slice codes, process monitoring item values, equipment parameter measured signal values, process energy consumption data, and finished product qualified rate index data. A model history library is established according to business slices and time slices, the business slices are triggered by production changes, the time slices are triggered by sampling interval time, and an automatic roasting big data model is established by using a three-layer feedforward neuron network based on the model history library;

[0041] A target process parameter determination module is used to query the records with the same business slice label in the model history library according to the current raw material code and the current business slice condition. If there is a corresponding record, the finished product code domain divided according to the time slice is selected, the quality qualified record that meets the pre-defined finished product qualified rate index is selected from the finished product code domain, and the record with the lowest process energy consumption is extracted from the quality qualified record. The process monitoring item values and the equipment parameter measured signal values of the record are used as the predicted equipment adjustment parameters; ​

[0042] a feedback check and online correction module, configured to collect current process monitoring item values and device parameter measured signal values in real time, perform flexible weighted deviation calculation on the predicted device adjustment parameters, and adjust and issue the device parameters online according to the results of the flexible weighted deviation calculation;

[0043] a rolling prediction control module, configured to periodically perform target process parameter determination, feedback check and online correction through automatic baking big data models, estimate future deviation values by using a prediction model, and perform rolling optimization by using a softening coefficient to determine a current control strategy, the control strategy including adjustment of the device parameters to minimize the deviation of the controlled variables from the expected values;

[0044] a control execution module, configured to execute the control strategy, and the control execution module is configured to adjust the fan speed, the valve opening, the machine speed and the material thickness;

[0045] a data management module, configured to add the raw material number and the time slice record to the model history library when there is no record corresponding to the raw material code in the model history library.

[0046] Preferably, the process monitoring item values are the drum drying section cover temperature, the drying section wind box temperature, the preheating section cover temperature, the preheating section wind box temperature, the baking section cover temperature, the baking section wind box temperature, the soaking section wind box temperature, the second cooling section cover temperature, the drum drying section cover pressure, the drum drying section wind box pressure, the drying section cover pressure, the preheating section cover pressure, the baking section cover pressure, the first cooling section cover back heat recovery air main pressure and the second cooling section cover pressure.

[0047] The device parameter measured signal values are the belt baking machine speed, the belt baking material thickness, the drum drying fan outlet pipeline hot air valve opening, the drying section wind box butterfly valve opening, the preheating section wind box butterfly valve opening, the second cooling section wind box butterfly valve opening, the drum drying fan outlet pipeline and the furnace cover communication valve opening, the wind box butterfly valve opening, the baking section each side 16 burner heat source flow, the 1# main exhaust fan speed, the 2# main exhaust fan speed, the back heat recovery fan speed, the furnace cover fan speed, the cooling fan speed and the drum drying fan speed.

[0048] The process energy consumption data are the electricity consumption, the gas consumption and the finished product quantity.

[0049] The finished product qualification rate indicators are the compressive strength, the drum index and the particle size.

[0050] The business slice is triggered by at least one of the following events: product variety change, production specification change, production line state change, ratio change, process production state change, material adding or subtracting operation and timing time to.

[0051] The sampling interval time of the time slice is calculated by dividing the trolley length by the real-time machine speed.

[0052] The present application at least comprises the following advantages:

[0053] The present application is based on information mining of historical and real-time data, combined with production process knowledge, application of big data analysis regression calculation, optimal control variable, measured feedback variable, flexible weighted deviation optimization, accurate system identification and multivariate algorithm modeling for multiple sections, multiple devices and multiple parameters, to issue more accurate and reasonable variable input for the control system, reduce the standard deviation of the temperature curve of the indurating machine through process intelligent control, control the correlation and dynamic rolling optimization control of the thermal parameters such as pressure and temperature of each process section, thereby improving the pellet quality and reducing the fuel consumption, realizing the balance of product quality, temperature and pressure balance.

[0054] Other advantages, objects and features of the present application will be apparent from the following description, and will be understood by those skilled in the art. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 The flowchart of one technical solution of the present application. DETAILED DESCRIPTION

[0056] The present application will be further described in detail below with reference to the accompanying drawings, so that those skilled in the art can implement it according to the description.

[0057] It should be understood that the terms such as "have", "contain" and "include" used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0058] It should be noted that the experimental methods described in the following embodiments are all conventional methods unless otherwise specified, and the reagents and materials can be obtained from commercial channels unless otherwise specified, so they cannot be understood as limiting the present application.

[0059] In the pellet production process, the indurating system is the core technology, which is composed of a material distribution system, a combustion system and a hot air circulation system. The entire indurating system is a thermal process, and the thermal process is realized by means of the combustion system and the air flow system, which is a quite large and complex heat exchange process. In this process, the system has the characteristics of interaction, strong coupling, mass / heat transfer and large lag. Therefore, the process parameters, equipment performance and product quality are the most important system control parameters.

[0060] The purpose of roasting production is to make the index parameters and state parameters optimal by adjusting the raw material parameters, operation parameters and equipment parameters. The state parameters reflect the state of the roasting process. The index parameters refer to the yield and quality index of the roasting pellets. The influencing factors of the yield index mainly include the environmental protection and energy efficiency of the roasting process, the finished product rate and the pallet speed, and these parameters are related to the control of the state of the roasting process. The roasting energy consumption is mainly related to the consumption of fuel and then related to the thermal state. The quality index includes three aspects of chemical composition, physical properties and metallurgical properties of the roasting pellets. The physical properties and metallurgical properties are mainly controlled by adjusting the state and working condition of the roasting process and reducing the fluctuation of the intermediate operation index.

[0061] The process of the belt roaster includes the following steps: the air drying section (the material layer is dried by air from bottom to top), the suction drying section (the material layer is dried by air from top to bottom), the preheating section (FeO oxidation reaction), the roasting section (the rest of FeO oxidation reaction), the soaking section (recrystallization / stress and strength), the first cooling section (temperature reduction and oxidation reaction) and the second cooling section (temperature reduction and cooling). The automatic roasting and process intelligent control includes the related control of the pallet speed, the material amount, the material thickness and the longitudinal flatness of the material surface. The temperature, the negative pressure control and the constraint coupling control of the furnace cover, the air box of the drying section, the preheating section, the roasting section, the soaking section, the first cooling section and the second cooling section are also included.

[0062] Therefore, the purpose of the present application is to provide a large data modeling analysis and online optimization control method for the key temperature, pressure, fan frequency and valve opening of the seven process sections of the drum drying section, the suction drying section, the preheating section, the roasting section, the soaking section, the first cooling section and the second cooling section in the pellet production process of the belt roaster.

[0063] As shown in Figure 1 The present application provides a method for automatic roasting and process intelligent control of a belt roaster, which comprises the following steps:

[0064] 1) historical data slice collection and modeling step, collecting monitoring data samples in the running process of the belt roaster, the monitoring data samples including raw material code (used to distinguish different raw material batches), slice code (uniquely identifying each data slice), process monitoring item value (reflecting the state of each process section), equipment parameter measured signal value (recording the running state of the equipment), process energy consumption data (statistical production energy consumption) and finished product qualification rate index data (evaluating product quality), business slice is triggered by production event change, when any one of the events occurs, the system automatically triggers the generation of a new business slice to distinguish production data under different working conditions, time slice is a data segment divided according to fixed time interval, ensuring the timeliness and continuity of data collection, triggered by sampling interval time, that is, the time interval between two data collections, a model historical library is established according to the business slice and the time slice, and each record in the library is associated with the corresponding raw material code, business slice label, time slice information and various monitoring data, a three-layer feedforward neural network (Three-layer Feedforward Neural Network) is used to establish an automatic roasting big data model, including an input layer, a hidden layer and an output layer, the input layer receives key data such as raw material code and process monitoring item value, the hidden layer processes data features through an activation function, and the output layer outputs the predicted device adjustment parameter related results. During the model training process, the weights are constantly updated using historical data to improve the model's ability to capture data patterns.

[0065] Based on the mastery of process data and long-term exploration and analysis, the system collects monitoring data samples, including raw material code, slice code, 15 key process parameters, 15 device adjustment parameters, 3 process energy consumption and 3 qualified index data history modeling, parameter prediction, online correction, rolling optimization, develops an automatic control model of belt roaster process parameters, and realizes big data analysis and parameter automatic control of belt roaster production process. Specifically, the process monitoring item value mainly includes temperature and pressure parameters, which directly reflect the thermal state of the roasting process. The process monitoring item values are the drum drying section cover temperature, the drying section wind box temperature, the preheating section cover temperature, the preheating section wind box temperature, the roasting section cover temperature, the roasting section wind box temperature, the soaking section wind box temperature, the two cooling section cover temperature, the drum drying section cover pressure, the drum drying section wind box pressure, the drying section cover pressure, the preheating section cover pressure, the roasting section cover pressure, the one cooling section cover back heat air main pipe pressure, and the two cooling section cover pressure; the device parameter measured signal value is the belt roaster speed, the belt roasting material thickness, the drum drying fan outlet pipeline hot air valve opening degree, the drying section wind box butterfly valve opening degree, the preheating section wind box butterfly valve opening degree, the two cooling section wind box butterfly valve opening degree, the drum drying fan outlet pipeline and the furnace cover communication valve opening degree, the wind box butterfly valve opening degree, the roasting section each side 16 burner heat source flow, the 1# main exhaust fan speed, the 2# main exhaust fan speed, the back heat fan speed, the furnace cover fan speed, the cooling fan speed, and the drum drying fan speed; the process energy consumption data is the electric power consumption, the gas consumption and the product quantity; the product qualified rate index is the compressive strength, the drum index and the particle size.

[0066] The system uses a unique slice technology to slice the process data and automatic control data continuously collected by the pellet belt roaster to generate slice feature data, providing data portraits for big data self-learning modeling. Specifically, the system collects monitoring data samples, establishes a model history library according to business and time slicing. Business slicing (7 types of production event triggers: product variety change; production specification change; production line state change; ratio change; process production state change; material addition or subtraction operation (one-time operation); timing time (for example, timing 10 minutes). The sampling interval time of time slicing, i.e. the trend tracking sampling interval (the time position conversion of each monitoring point of time series data), is calculated by the length of the trolley divided by the real-time machine speed.

[0067] The belt roaster optimal parameter prediction algorithm and process intelligent control model are based on the stable assumption, and the problems of large data analysis modeling, online feedback correction, multivariate rolling optimization, and precise control in the continuous production process with dynamic changes in operating conditions are coordinated by using computer technology and automatic control technology. Before using the algorithm and model, the production state condition needs to be judged according to the quality of the detection data within a certain period. When the data quality does not exceed the reasonable range of error, it is considered that the production state is stable, the model calculation is started, and the system uses the process production data with the optimal calculation result of the model historical library and the current real-time production monitoring information to control and guide together.

[0068] The automatic roasting big data model is established by using artificial neural network method. Here, a three-layer feedforward neural network is used to establish the automatic roasting big data model. The neural network is composed of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function, called activation function. The connection between each two nodes represents a weighted value for the signal passing through the connection, called weight. The adaptive learning of the model is to continuously update the optimal solution and deviation trend control of the target and control according to the real-time collected production data during use, so that the model can timely track, analyze and control the dynamic changes of the roasting process. The complex data modeling process can be avoided, the nonlinear mapping of the optimal process parameters and process intelligent control of the system can be realized, and the adaptive and self-learning ability of the neural network can be used to track the dynamic changes of the system. According to the state of the roasting process, the adjustment amount of the process and equipment parameters is calculated by using the big data self-learning modeling and optimal control algorithm, and the control parameters are flexibly weighted and optimized online according to the actual state of the seven process sections. The parameters of the control are the parameters of the control.

[0069] 2) Target process parameter determination step, according to the current raw material code and the current business slice condition, query the record with the same business slice tag in the model historical library, if the corresponding record exists, select the relevant data of each batch of qualified products corresponding to each product code in the product code domain according to the time slice division), select the quality qualified record that meets the pre-defined product qualified rate index from the product code domain, and extract the record with the lowest energy consumption from the quality qualified record. The process monitoring item value and the equipment parameter measured signal value of the record are used as the predicted equipment adjustment parameter.

[0070] Through accurate modeling of the production process and artificial neural network analysis and optimization technology, the golden batch of products (including: optimal solution of quality, lowest solution of energy consumption, highest solution of total iron) is obtained, and the optimal process parameters and the best control parameters of the equipment under the actual production conditions are obtained according to the constraint conditions and optimization calculation.

[0071] First, the model history library is queried according to the raw material code. Then, the quality qualified product code domain meeting the pre-defined finished product qualification rate index is further queried based on the first result data, and the query is directly ended if there is no solution; finally, the finished product code record with the lowest process energy consumption is further queried based on the second result data, and the 15 process monitoring item values and the corresponding 15 device adjustment parameters of the record are extracted as the predicted device adjustment parameters in the model history library under the raw material and working condition.

[0072] The system first queries the model history library according to the raw material code of the current production. After the production of each batch of finished product is completed, the system associates the corresponding quality detection data and energy consumption data to the corresponding time slice record, and gradually accumulates the historical production data of the raw material. When the raw material is put into production again, the system can query the related records already in the model history library, providing data support for the determination of target process parameters. In the query process, the system quickly searches the library for whether there is a historical record corresponding to the same raw material code through the unique identification of the raw material code. If the search result shows that there is no related record of the raw material code in the model history library, it means that the raw material is put into production for the first time, and there is no historical production data for reference. At this time, the system automatically starts the new record process, enters the current raw material code into the model history library, and generates the corresponding initial time slice record. The initial time slice record contains the current business slice conditions (such as the current product variety, production specification, production line state, etc.), the initial sampling interval time (calculated according to the current trolley length and real-time machine speed), and the subsequent real-time collection of process monitoring item values, device parameter measured signal values, process energy consumption data and finished product qualification rate index data. The slice code of the new record follows a unified coding rule, including the raw material code, business slice identifier and time information, ensuring the uniqueness and traceability of the record.

[0073] 3) Feedback verification and online correction step, real-time collection of current process monitoring item values, device parameter measured signal values through sensors and other detection devices, comparison with the predicted device adjustment parameters obtained in the target process parameter determination step, calculation of the deviation of each parameter, statistics of the change rate of each deviation within a sampling period, determination of dynamic weight combined with the preset basic adjustment coefficient, maximum deviation and maximum deviation change rate, flexible weighted deviation calculation based on dynamic weight, the basic adjustment coefficient is obtained by training the historical data through the automatic roasting big data model, the dynamic weight is adjusted according to the deviation size ratio and the deviation change rate ratio, the larger the deviation and the faster the change rate, the higher the dynamic weight, online adjustment and issuance of device parameters according to the flexible weighted deviation calculation result.

[0074] By using big data history library, time series tracking and self-learning, the data model of the equipment is automatically optimized. By using optimal process parameter solving and parameter prediction, the influence trend of various feedbacks and production processes is calculated in real time, so as to adjust and control the strategy in real time, stabilize the optimal operation of the equipment, quantify and patrol the relationship between variables, and provide data and model support for optimal and accurate control.

[0075] The system collects the current values of 15 process monitoring items in real time, matches the 15 process monitoring item corresponding 15 device parameter measured signal values, and completes the flexible weighted deviation optimization with the predicted device adjustment parameters derived in the model history library in step 2), considers the deviation size and deviation change rate, and finally issues the optimal adjustment parameters of the equipment.

[0076] Feedback verification, online correction module identifies the model, and recognizes the influence of various prediction trends on the controlled parameters. When the controlled parameters occur, the feedback verification and online real-time correction control strategy are adjusted in time according to the deviation of the controlled parameters caused by the future time prediction trend, so that the controlled parameters can be close to the model prediction value.

[0077] 4) Rolling prediction control step, automatic roasting big data model periodically executes target process parameter determination step, feedback verification and online correction step, in each execution process, the prediction model is used to estimate the parameter deviation trend in the future several sampling periods, and a flexible coefficient is introduced for rolling optimization. The flexible coefficient is dynamically adjusted according to the relative deviation average value and the fluctuation dispersion of the current key process monitoring items (the required items are the roasting section cover temperature, the roasting section cover pressure and the belt roasting thickness, and the optional supplementary items are the preheating section cover temperature, the soaking section air tank temperature and the drying section cover pressure, a total of 3-6 items). The smaller the relative deviation average value and the fluctuation dispersion, the closer the flexible coefficient to the basic value, and the more stable the optimization effort. Otherwise, the flexible coefficient decreases, and the optimization effort is appropriately reduced to avoid control shock. The current control strategy is determined to realize the continuous adjustment of the equipment parameters. The control strategy includes the adjustment of the equipment parameters, so that the controlled variables and the expected value deviation are minimized.

[0078] The optimal adjustment parameters of the equipment calculated by the model once are only the actual execution at the current time, and the optimal adjustment parameter values of the equipment are solved again by this method at the next model calculation period and sampling time. Not only the current and past deviation values are used, but also the prediction model is used to predict the future deviation values of the process. The flexible coefficient is put in to determine the current optimal control strategy by rolling optimization. The overall optimization control time sequence is solved, the optimal control is issued according to the time sequence and position tracking, and the controlled variables and the expected value deviation are minimized in the future period of time.

[0079] The rolling optimization and optimal control module finds the optimal operating point of the device through an optimal solution, and the feedback correction and online correction module stabilizes the device at the optimal point for operation. The rolling optimization and optimal control module is implemented on the basis of improving control quality, and only by reducing the fluctuation of variables can the device be operated on the boundary of the constraint condition to stabilize the optimal control.

[0080] In the above technical solution, real-time working condition slices are first collected and recorded in a history library, and then the optimal process and device setting parameters under each type of production working condition are found by screening historical data with high quality and low consumption; then, rolling optimization is performed in combination with the model history library and real-time correction, and is realized through computer programming, data slice collection, multi-constraint multi-solution optimization, and flexible rolling optimization by comparing predicted values with actual values. The device parameter full-automatic control of the belt-type indurating machine under the working condition of stable balance of combustion temperature and pressure, optimal product quality, and lowest production energy consumption is realized, the optimal setting values of each link in the process are calculated, the response efficiency and control accuracy of production control are improved, the operation efficiency is improved, the product quality is stabilized, and the product cost is reduced, so that real-time stability and optimal control of the pellet production are realized; at the same time, the labor intensity of the operators for all-weather continuous monitoring and adjustment is greatly reduced, and method guidance and technical support are provided for the unmanned and intelligent production of the belt-type indurating machine.

[0081] In one example, the following steps are specifically included:

[0082] 1. Historical data slice collection, data time sequence tracking, and big data self-learning modeling;

[0083] First, the trend tracking sampling interval is calculated, and data is collected as a time sequence unit, and the trend tracking sampling interval (i.e. the time position conversion of each monitoring point of the time sequence data) = the length of the pellet belt-type indurating machine trolley / real-time machine speed;

[0084] For a batch of iron ore powder raw materials (corresponding to a specific raw material code) pretreated by a high-pressure roller mill, monitoring data samples are collected, including raw material code, slice code, 15 process monitoring item values, 15 device parameter measured signal values, 3 process energy consumption data, and 3 finished product qualification rate index data, and a model history library is established according to business and time slices;

[0085] 2. Optimal process parameter solving and parameter prediction;

[0086] 2.1. According to the raw material code, the model history library is queried, if there is, the corresponding material code is queried to obtain the finished product code domain E1-n1 (N1 records) divided according to the time slice T1-x, if there is no solution, the raw material number and time slice record are added to the model history library;

[0087] 2.2、According to the first result data, further query the qualified product code domain E1-n2 (N2 records) with the conditions of compression resistance >=2500N, drum index >=90%, and particle size 10-16 ratio >=75%. If there is no solution, end the model directly.

[0088] 2.3、According to the second result data, further query the product code record E3 (1 record) with the lowest process energy consumption min[(electricity consumption + gas consumption) / product quantity sum], and extract the 15 process monitoring item values and the corresponding 15 device adjustment parameters of the record as the predicted device adjustment parameters in the model history library under the raw material and working condition. If there is no solution, end the model directly.

[0089] 3、Feedback verification and online correction;

[0090] The system collects the current 15 process monitoring item values in real time, matches the 15 process monitoring item corresponding 15 device parameter measured signal values, and completes the flexible weighted deviation optimization with the predicted device adjustment parameters derived in step 2 in the model history library. Finally, the optimal device adjustment parameters are issued in combination with the process characteristics of drying, preheating and other sections in the roasting process.

[0091] 4、Rolling optimization and optimal control;

[0092] Steps 2 and 3 are periodically executed according to the sampling interval. When the production is stable, the process parameters fluctuate little, the softening coefficient is close to the basic value, and only the parameters such as air valve opening degree are slightly adjusted to maintain stability. When the moisture content of the raw material suddenly increases, causing the roasting temperature to drop and the pressure fluctuation to intensify, the softening coefficient automatically decreases, slowly increasing the fan speed and the heat supply of the burner to avoid parameter shock caused by sudden changes. After the moisture fluctuation is alleviated and the parameters tend to be stable, the softening coefficient rises, and the adjustment is adjusted to a small amplitude of fine tuning, so that each index is finally stabilized in the qualified range.

[0093] The drum drying and suction drying section pressure balance control, main suction fan frequency and roasting section front air box temperature gradient control, main suction fan frequency and suction drying section upper cover pressure, preheating section upper cover pressure, roasting section upper cover pressure deviation control rules and parameter optimization are shown in Table 1.

[0094] Table 1

[0095]

[0096] The traditional control has fixed adjustment strength for process monitoring items, and cannot dynamically adapt according to parameter deviation (difference between actual value and target value) and deviation change trend (change speed of deviation in unit time). When the process parameter deviates from the target value by a large margin or the deviation rapidly expands, the fixed adjustment strength is difficult to quickly suppress the deviation, resulting in that the process state continuously deviates from the optimal interval; when the parameter is close to the target value and the deviation is small, the fixed adjustment strength is easy to cause overshoot (the parameter deviates in the opposite direction after passing the target value), causing system oscillation, affecting the temperature and pressure stability of the roasting process, and further leading to product quality fluctuation or energy consumption increase. In addition, the qualified working condition deviation ranges of each process monitoring item (such as the temperature inside the drum drying section cover and the pressure inside the roasting section cover) of the belt roaster are different, and the traditional fixed adjustment method cannot adapt to the characteristic differences of different monitoring items, further reducing the control accuracy. In another technical solution, the specific process of flexible weighted deviation calculation in the feedback verification and online correction steps is as follows:

[0097] For each process monitoring item of the belt roaster, the real-time value of the current process monitoring item is collected in real time, and the target value corresponding to the predicted device adjustment parameter in the target process parameter determination step is called to calculate the difference between the two, which is the deviation of the process monitoring item E i ; at the same time, the deviation change amount of the process monitoring item in a sampling period (the sampling period is determined by the time slicing rule) is recorded, and the deviation change rate Δ E i is obtained by dividing the sampling period length.

[0098] For each process monitoring item, the qualified working condition data in the model history library is trained in advance by an automatic roasting big data model to obtain the basic adjustment coefficient K i of the process monitoring item K i The value of the basic adjustment coefficient is related to the characteristics of the process monitoring item, for example, the temperature inside the roasting section cover has a significant impact on the product quality and a large response lag, K i which will be adapted to a value more suitable for temperature adjustment sensitivity through training to ensure that the basic adjustment direction and strength meet the process requirements of the monitoring item.

[0099] A dynamic weight function f ( E i , Δ E i ) is introduced, which is a function of the deviation E i and the deviation change rate Δ E iFor input, combine preset weighting factor and historical statistical maximum value to calculate dynamic weight under current working condition. First, extract the absolute value of the deviation of the process monitoring item under all qualified working conditions from the model history library, and statistically obtain the maximum value E i，max ; similarly, extract the absolute value of the deviation change rate under qualified working conditions, and statistically obtain the maximum value Δ E i，max ; then, substitute the preset weighting factor α and β , and calculate the result of the dynamic weight function.

[0100] Multiply the basic adjustment coefficient K i , the result of the dynamic weight function f ( E i , Δ E i ) and the deviation E i to obtain the equipment parameter adjustment amount Δ U i of the process monitoring item corresponding to the equipment parameter adjustment amount Δ U i ; according to the adjustment amount Δ

[0101] , adjust the equipment parameter associated with the process monitoring item (such as the inner temperature of the drum dry section cover associated with the hot air valve opening degree of the drum dry fan outlet pipeline), and issue the adjusted equipment parameter to the actuator (such as the valve, fan controller), complete a feedback check and online correction. i U Specifically, for the first process monitoring item, its equipment parameter adjustment amount Δ

[0102] i is determined by the following formula:

[0103] E i is the deviation of the current real-time value of the first i process monitoring item from the corresponding target value in the predicted equipment adjustment parameter, E i is positive when the parameter is higher than the target value, and is negative when it is lower than the target value, Δ E i is the deviation E i of the change rate in a sampling period;

[0104] K i is the first iThe basic adjustment coefficient corresponding to each process monitoring item is obtained by training historical data through an automatic roasting big data model. For example, the adjustment of the roasting section wind box temperature is strong in hysteresis, and the model learns from historical data that a slightly large basic coefficient is needed to speed up the response, while the adjustment of the inner cover pressure of the second cooling section is fast, K i will be adapted to a smaller value to avoid over-adjustment;

[0105] f E i , Δ E i is a dynamic weight function, and its expression is:

[0106]

[0107] α and β are preset weighting factors, and the value ranges are 0.1-0.5 and 0.05-0.3, respectively, α the greater E i , the greater the contribution of the dynamic weight, β the smaller, the contribution of the deviation change rate to the weight is more gentle, and for monitoring items that are large in response hysteresis and critical to quality (such as the inner cover temperature of the roasting section), a larger K i and α is set to ensure that the adjustment is strong enough when the deviation is large; for monitoring items that are easily disturbed and need to be stabilized first (such as the inner cover pressure of the draining section), a smaller β is set to avoid adjustment shock caused by short-term fluctuations; E i，max and Δ E i，max are the statistical maximum values of the absolute value of the deviation and the absolute value of the deviation change rate of the i th process monitoring item in the model history library under qualified working conditions, reflects the severity of the current deviation relative to the maximum deviation of the historical qualified working conditions, reflects the urgency of the current deviation change speed relative to the maximum change speed of the historical qualified working conditions, f E i , Δ E i The value range is greater than 1, and the adjustment force needs to be increased, and close to 1 needs to maintain / reduce the adjustment force.

[0108] In the above technical solution, according to the actual deviation E i of each process monitoring item, as well as the change trend Δ E ​​i The system adaptively adjusts the adjustment intensity. When the parameter deviates significantly from the target value, or when the deviation changes rapidly, the dynamic weight automatically increases, thereby increasing the adjustment intensity of the equipment parameters and quickly suppressing the spread of deviation. When the parameter is close to the target value, the deviation is small, and the change is gradual, the dynamic weight decreases accordingly, and the adjustment intensity also decreases, effectively avoiding system overshoot and achieving stable and precise control, making the adjustment behavior more in line with the severity of deviation in real-time operating conditions.

[0109] In the rolling predictive control stage of a belt calciner, the belt calciner, as a complex controlled object with multiple variables, strong coupling, and large time lag, often faces two typical types of fluctuations during production: one is small, stable fluctuations (such as slight changes in raw material moisture content or instantaneous air pressure disturbances), and the other is large, drastic fluctuations (such as sudden changes in raw material composition, recovery after equipment failure, or adjustments to production line load). Traditional model predictive control (MPC) often uses fixed values ​​for its softening coefficients, which cannot adapt to the dynamic fluctuation characteristics of the belt calciner production process. In another technical solution, this invention achieves adaptive optimization of rolling predictive control through a process of selecting key process monitoring items → quantifying fluctuations and deviations → calculating dynamic coefficients → solving for softening coefficients → adapting control strategies.

[0110] First, select 3-6 key process monitoring items from the 15 process monitoring items of the belt roaster: the mandatory items are the temperature inside the roasting section hood (which directly determines the roasting intensity of the pellets), the pressure inside the roasting section hood (which affects the hot air circulation efficiency), and the thickness of the roasted material (which determines the roasting uniformity). Optional supplementary items are selected according to the current working conditions (such as the temperature inside the preheating section hood, the temperature of the air box in the homogenizing section, and the pressure inside the drying section hood) to ensure that the selected parameters can fully reflect the core state of the roasting process (avoiding parameter redundancy or omission).

[0111] For each selected key process monitoring item, its current real-time value is collected and compared with the expected value obtained from the target process parameter determination step. The absolute value of its relative deviation is calculated to eliminate the difference in magnitude between different parameters. The absolute values ​​of the relative deviations of all key process monitoring items are summed and the arithmetic mean is taken to obtain the average relative deviation Δ, which reflects the overall degree of deviation between the current production process and the target operating condition. The larger the Δ, the more serious the deviation of the current parameter from the target.

[0112] The process stability of a belt roaster requires attention to both static deviation Δ and dynamic fluctuations. σ Traditional control only focuses on Δ, but in reality, Δ is small but... σ The parameters are large, close to the target but fluctuate frequently, requiring careful control; Δ is large, but... σSmall, parameter deviation from target but stable change, can adjust moderately aggressive. Select the key process monitoring item data in the last 10-30 sampling periods, calculate the standard deviation of the relative deviation of each key item, understand the fluctuation amplitude of a single item, then sum the standard deviations of all key items and take the arithmetic mean to get the comprehensive standard deviation σ , reflecting the overall fluctuation intensity of the current production process, σ The larger the value, the more frequent the parameter fluctuation.

[0113] Query all qualified conditions that meet the finished product qualification rate index in the model history library, extract the relative deviation comprehensive standard deviation of the key process monitoring items under these qualified conditions σ ref , take the 90% quantile as the reference standard deviation, which is used to judge whether the current fluctuation is beyond the historical qualified range.

[0114] Pre-set weight coefficient C Take larger for the condition sensitive to fluctuation C , such as temperature control in the baking section, take smaller for the condition with high tolerance to fluctuation C , combined with fluctuation dispersion σ and reference standard deviation σ ref , calculate the dynamic adjustment coefficient k ( σ ), reflecting the deviation of the current fluctuation relative to the historical qualified fluctuation, σ The larger the value, σ ref The smaller the value, k ( σ ) is larger, and vice versa.

[0115] Specifically, the softening coefficient λ in the rolling prediction control step is a dynamic self-adaptive variable, λ The larger the value, the more range of parameter adjustment is allowed to pursue long-term optimization, λ The smaller the value, the range of parameter adjustment is limited to ensure current stability:

[0116]

[0117] Among them, λ 0 is the basic softening coefficient, the value range is 0.8-1.0, which is set according to the conventional working condition of the belt baking machine, to ensure that the system can continuously pursue the optimal target when there is no fluctuation, is the adaptive adjustment factor of the softening coefficient, the value range is 0-1, which determines λ the attenuation degree relative to λ 0, Δ is the average value of the absolute value of the relative deviation of the current several key process monitoring items, , the summation range jFrom 1 to m , m The number of key process monitoring items, m The value ranges from 3 to 6. The larger Δ is, the better. k ( σ The larger the exponent, the closer it is to 0. λ The smaller; conversely λ The closer λ 0. The key process monitoring items include the temperature inside the roasting section hood, the pressure inside the roasting section hood, and the thickness of the roasted material as mandatory items, and the temperature inside the preheating section hood, the temperature of the wind box in the homogenization section, and the pressure inside the drying section hood as optional supplementary items. The combination of mandatory items and optional supplementary items constitutes the key process monitoring items.

[0118] k ( σ This is a dynamic adjustment coefficient related to volatility dispersion, reflecting the severity of the current volatility relative to historical acceptable volatility. ,in C The preset weighting coefficient ranges from 0.03 to 0.08. If temperature fluctuations in the roasting section have a significant impact on the quality of the finished product, a weighting factor can be set. C =0.08, which slightly increases the fluctuation. k ( σ Rapid rise, λ Rapid decay, minimal impact from pressure fluctuations in the pumping section, can be set C =0.03, to avoid small fluctuations causing λ Over-adjustment σ for m The composite standard deviation of the relative deviations of the key process monitoring items over the most recent 10-30 sampling periods. The composite standard deviation is the arithmetic mean of the standard deviations of the relative deviations of the key process monitoring items. σ ref The reference standard deviation is determined by querying the historical database of the model and taking the 90th percentile of the comprehensive standard deviation of the relative deviations of the key process monitoring items under all qualified operating conditions.

[0119] In the above technical solution, this invention obtains the basic adjustment coefficient through model training and automatically adapts to deviation changes by combining dynamic weights, achieving optimal adjustment without manual intervention. Based on the real-time stability of the production process, the predicted time-domain attributes are dynamically adjusted; when process fluctuations are severe (the dispersion of key process monitoring items...), the adjustment is applied accordingly. σ When the relative deviation average value Δ is large, the softening coefficient is... λ This automatically reduces the effective prediction time domain, allowing the controller to focus more on suppressing current drastic fluctuations and behave more cautiously to avoid system instability due to over-optimization; when the production process is stable ( σ When the coefficient of flexibility is small (Δ is small), the softening coefficient is also small. λThis will increase the predictive time domain of the controller, making its control behavior more aggressive. It will be able to optimize and adjust process parameters in advance, fully pursuing the optimal performance of product quality and energy consumption, significantly reducing the labor intensity of operators in monitoring and adjusting the feedback verification process, and providing support for the intelligent control of belt roasters.

[0120] In the production stages of the belt calciner's drying and extraction sections, the drying section removes surface moisture from the green pellets by blowing air upwards, while the extraction section removes internal moisture from the pellets by drawing air downwards. The relative pressure balance between the two sections directly determines the drying efficiency, hot air utilization rate, and finished product moisture content. In another technical solution, the control strategy also includes balancing the pressure between the drying and extraction sections:

[0121] The valve connecting the outlet pipe of the drying blower to the furnace hood only directly affects the pressure inside the drying section hood (changes in valve opening - changes in hot air leakage in the drying section - pressure changes), and does not interfere with the pressure in the extraction section or other process sections (such as the preheating section and calcination section). Changes in the speed of the No. 2 main exhaust fan only directly affect the pressure in the extraction section (increased speed - enhanced extraction capacity - decreased pressure in the extraction section). By adjusting the opening of the valve connecting the outlet pipe of the drying blower to the furnace hood and the speed of the second main exhaust fan, the measured pressure inside the drying section hood can be adjusted. P 鼓干段 Measured pressure inside the desiccant hood P 抽干段 ratio D Stable within the preset equilibrium range [ D min , D max ]Inside;

[0122] D Excessive pressure means that the pressure in the blower section is relatively dominant, and hot air tends to accumulate there, failing to effectively enter the extraction section. D Too low a pressure means the desiccant section has a relatively dominant pressure, which can easily lead to excessive heat removal from the blower section, resulting in decreased drying efficiency. Pressure changes in one section can indirectly affect the pressure in another section through hot air flow (for example, an increase in the pressure in the blower section can cause some hot air to enter the desiccant section, passively increasing its pressure). D Stable at [ D min , D max Inside, regardless of how the absolute values ​​of the two pressures fluctuate, their relative balance always adapts to the drying requirements of the current working conditions. The hot air in the blowing drying section can effectively penetrate the material layer to remove surface moisture, and the desiccation section can accurately remove internal moisture, avoiding situations where the surface is too dry and the inside is too wet or the surface is too wet and the inside is too dry, thus fundamentally solving the coupling interference problem.

[0123] Among them, according to the current raw material code and the current business slicing conditions, query the quality qualified record data segments in the model history library that meet the predefined finished product qualification rate index under the same business slicing label, extract the target data segment with the lowest process energy consumption, and count the D value distribution within the target data segment, excluding the extreme D values (such as abnormal values caused by instantaneous pressure fluctuations) in the target data segment, and take the D value range with a frequency of occurrence ≥ 90%, the mainstream stable area under the historical optimal working conditions, as the initial D min , D max .

[0124] When D > D max , the pressure in the drying section is relatively too high:

[0125] Adjust the opening degree of the connecting valve between the outlet pipe of the drying fan and the furnace hood: increase the opening degree of the valve to directly connect a part of the hot air at the outlet of the drying fan to the furnace hood, reducing the pressure inside the drying section hood ( P 鼓干段 decreases);

[0126] Auxiliary adjustment of the speed of the 2# main exhaust fan: If the D still does not fall back after the valve adjustment, the speed of the 2# main exhaust fan can be slightly increased to enhance the exhaust capacity of the drying section and increase P 抽干段 , so that D drops to within the range.

[0127] When D <Ymin (the pressure in the drying section is relatively too low):

[0128] Adjust the opening degree of the connecting valve between the outlet pipe of the drying fan and the furnace hood: reduce the opening degree of the valve to reduce the leakage of hot air from the outlet of the drying fan to the furnace hood and increase the pressure inside the drying section hood ( P 鼓干段 increases);

[0129] Auxiliary adjustment of the speed of the 2# main exhaust fan: If the D still does not rise after the valve adjustment, the speed of the 2# main exhaust fan can be slightly reduced to weaken the exhaust capacity of the drying section and reduce P 抽干段 , so that D rises to within the range.

[0130] When D is in D min , D maxInside: maintain the current valve opening and fan speed, only through real-time monitoring to ensure D The value is stable, avoid the waste of energy consumption caused by meaningless adjustment.

[0131] In the above technical scheme, by pressure ratio D Stable in the historical calibration D min , D max ] interval, ensure that two hot air penetration and exhaust matching, product pellet moisture fluctuation is greatly reduced, for subsequent baking process to lay a stable foundation, indirectly improve the compressive strength, drum index qualified rate, while reducing drying energy consumption, avoid excessive blowing or exhaust, reduce the fan invalid work, significantly reduce the drum-dryer section of the electric power consumption and gas consumption, but also can adapt to multi-working condition, through the raw material coding and business slice automatic call corresponding balance interval, without manual re-set, flexible response to different raw materials and varieties production, greatly reduce the labor intensity of manual monitoring.

[0132] Automatic baking and process intelligent control device of belt baking machine, comprising:

[0133] The historical data slice acquisition and modeling module is used to acquire monitoring data samples, the monitoring data samples including raw material codes, slice codes, process monitoring item values, device parameter measured signal values, process energy consumption data and finished product qualification rate index data, a model historical library is established according to business slices and time slices, the business slices are triggered by production changes, the time slices are triggered by sampling interval times, and based on the model historical library, an automatic roasting big data model is established by using a three-layer feedforward neuron network; the process monitoring item values are drum drying section cover-in temperature, drying section wind box temperature, preheating section cover-in temperature, preheating section wind box temperature, roasting section cover-in temperature, roasting section wind box temperature, soaking section wind box temperature, two cooling section cover-in temperature, drum drying section cover-in pressure, drum drying section wind box pressure, drying section cover-in pressure, preheating section cover-in pressure, roasting section cover-in pressure, one cooling section cover-in backheating wind main pipe pressure and two cooling section cover-in pressure; the device parameter measured signal values are belt roasting machine speed, belt roasting material thickness, drum drying fan outlet pipeline hot air valve opening degree, drying section wind box butterfly valve opening degree, preheating section wind box butterfly valve opening degree, two cooling section wind box butterfly valve opening degree, drum drying fan outlet pipeline and furnace cover communication valve opening degree, wind box butterfly valve opening degree, roasting section each side rear 16 burner heat source flow, 1# main exhaust fan speed, 2# main exhaust fan speed, backheating fan speed, furnace cover fan speed, cooling fan speed and drum drying fan speed; the process energy consumption data is electric energy consumption, gas energy consumption and finished product quantity; the finished product qualification rate index is compressive strength, drum index and particle size; the business slices are triggered by at least one of the following events: product variety change, production specification change, production line state change, ratio change, process production state change, material adding or subtracting operation and timing time; the sampling interval time of the time slice is calculated by the length of the trolley divided by the real-time machine speed;

[0134] The target process parameter determination module is used to query the records with the same business slice label in the model historical library according to the current raw material code and the current business slice condition, if there is a corresponding record, the finished product code domain divided according to the time slice is selected, the quality qualified record meeting the pre-defined finished product qualification rate index is selected from the finished product code domain, and the record with the lowest process energy consumption is extracted from the quality qualified record, the process monitoring item values and the device parameter measured signal values of the record are used as the predicted device adjustment parameters;

[0135] The feedback checking and online correction module is used to acquire the current process monitoring item values and the device parameter measured signal values in real time, perform flexible weighted deviation calculation on the predicted device adjustment parameters, and adjust and issue the device parameters online according to the flexible weighted deviation calculation result;

[0136] The rolling prediction control module is used for periodically performing target process parameter determination, feedback checking and online correction through automatic roasting big data model, predicting future deviation values by using a prediction model, and placing a softening coefficient to perform rolling optimization, so as to determine a current control strategy, wherein the control strategy comprises adjustment of device parameters, so that a controlled variable and an expected value are minimized.

[0137] The control execution module is used for executing the control strategy, and the control execution module adjusts fan rotating speed, valve opening, machine speed and material thickness.

[0138] The data management module is used for adding the raw material number and time slice record to the model history library when there is no record corresponding to the raw material code in the model history library.

[0139] In the above technical solution, real-time working condition slices are first collected and recorded in the history library, then the optimal process and device setting parameters under each type of production working condition are found by screening historical data with high quality and low consumption, then rolling optimization is performed in combination with the model history library and real-time correction, and finally, the flexible rolling optimization is realized through computer programming, data slice collection, multi-constraint multi-solution optimization, and comparison of predicted values and actual values. The device parameters of the belt roaster are automatically controlled under the working condition of stable balance of combustion temperature and pressure, optimal product quality and lowest production energy consumption, the optimal setting values of each link in the process are calculated, the response efficiency and control accuracy of production control are improved, the operation rate is improved, the product quality is stabilized, and the product cost is reduced, so that real-time stability and optimal control of the pellet production are realized. Meanwhile, the labor intensity of continuous monitoring and adjustment of operators is greatly reduced, and method guidance and technical support are provided for unmanned and intelligent pellet production of the belt roaster.

[0140] The number of devices and the processing scale described herein are used to simplify the description of the present application. Applications, modifications and changes to the present application are obvious to those skilled in the art.

[0141] Although the embodiments of the present application have been disclosed as above, they are not limited to the applications and embodiments listed in the specification, and can be fully applied to various fields suitable for the present application, and additional modifications can be easily realized by those skilled in the art, therefore, the present application is not limited to specific details and the figures shown and described herein, without departing from the general concept defined by the claims and the equivalent scope.

Claims

1. A method for automatic roasting and intelligent process control of a belt roaster, characterized in that, include: The historical data slicing acquisition and modeling steps involve collecting monitoring data samples during the operation of the belt roaster. The monitoring data samples include raw material codes, slice codes, process monitoring item values, measured signal values ​​of equipment parameters, process energy consumption data, and finished product qualification rate index data. Business slices are triggered by changes in production events, and time slices are triggered by sampling intervals. A model history library is established according to the business slices and time slices. Based on the model history library, an automatic roasting big data model is established using a three-layer feedforward neural network. The steps for determining target process parameters are as follows: Based on the current raw material code and current business slice conditions, query the records with the same business slice label in the model history database. If a corresponding record exists, filter out the finished product code domain divided by time slice. Select the quality qualified records that meet the predefined finished product qualification rate index from the finished product code domain. Then, extract the record with the lowest process energy consumption from the quality qualified records. Use the process monitoring item value and the measured signal value of the equipment parameter of this record as the predicted equipment adjustment parameters. The feedback verification and online correction steps involve real-time acquisition of current process monitoring item values ​​and measured equipment parameter signal values, and performing flexible weighted deviation calculation based on dynamic weights with the predicted equipment adjustment parameters. Based on the flexible weighted deviation calculation results, the equipment parameters are adjusted online and issued. The specific process of calculating the flexible weighted bias is as follows: For the i Each process monitoring item, its equipment parameter adjustment amount Δ U i Determined by the following formula: in, E i For the first i The deviation between the current real-time value of each process monitoring item and the corresponding target value in the predicted equipment adjustment parameters; Δ E i For deviation E i Rate of change within a sampling period; K i For the first i The basic adjustment coefficients corresponding to each process monitoring item are obtained by training historical data using an automatic roasting big data model. f ( E i ,Δ E i ) is the dynamic weight function, and its expression is: α and β These are preset weighting factors, with values ​​ranging from 0.1 to 0.5 and from 0.05 to 0.3, respectively. E i，max and Δ E i，max The first i The statistical maximum values ​​of the absolute value of deviation and the absolute value of the rate of change of deviation for each process monitoring item under qualified operating conditions in the model history database. The rolling predictive control step involves the automatic roasting big data model periodically executing the target process parameter determination step, feedback verification and online correction step. The predictive model is used to estimate the future deviation value of the process, and a softening coefficient is introduced for rolling optimization to determine the current control strategy. The control strategy includes adjusting the equipment parameters to minimize the deviation between the controlled variable and the expected value. Softening coefficient λ For dynamic adaptive variables: in, λ 0 is the basic flexibility coefficient, ranging from 0.8 to 1.0, and Δ is the average of the absolute values ​​of the relative deviations of several key process monitoring items. Summation range j From 1 to m , m The number of key process monitoring items, m The value range is 3-6. The key process monitoring items include the temperature inside the roasting section hood, the pressure inside the roasting section hood, and the thickness of the roasted material as mandatory items, and the temperature inside the preheating section hood, the temperature of the wind box in the homogenizing section, and the pressure inside the drying section hood as optional supplementary items. The combination of mandatory items and optional supplementary items constitutes the key process monitoring items. k ( σ ) is the dynamic adjustment coefficient related to the fluctuation dispersion. ,in C The preset weighting coefficients range from 0.03 to 0.

08. σ for m The composite standard deviation of the relative deviations of the key process monitoring items over the most recent 10-30 sampling periods. The composite standard deviation is the arithmetic mean of the standard deviations of the relative deviations of the key process monitoring items. σ ref The reference standard deviation is determined by querying the historical database of the model and taking the 90th percentile of the comprehensive standard deviation of the relative deviations of the key process monitoring items under all qualified operating conditions.

2. The method for automatic roasting and intelligent process control of a belt roaster according to claim 1, characterized in that, The process monitoring values ​​are: temperature inside the drying section hood, temperature inside the extraction section air box, temperature inside the preheating section hood, temperature inside the preheating section air box, temperature inside the calcination section hood, temperature inside the calcination section air box, temperature inside the homogenizing section air box, temperature inside the second cooling section hood, pressure inside the drying section hood, pressure inside the drying section air box, pressure inside the extraction section hood, pressure inside the preheating section hood, pressure inside the calcination section hood, pressure in the main regenerative air pipe inside the first cooling section hood, and pressure inside the second cooling section hood. The measured signal values ​​of the equipment parameters are: roasting machine speed, roasting material thickness, opening of the hot air valve on the outlet pipe of the drying blower, opening of the butterfly valve in the exhaust section, opening of the butterfly valve in the preheating section, opening of the butterfly valve in the secondary cooling section, opening of the valve connecting the outlet pipe of the drying blower to the furnace hood, opening of the butterfly valve in the exhaust section, heat source flow rate of the 16 burners on each side of the roasting section, speed of the No. 1 main exhaust fan, speed of the No. 2 main exhaust fan, speed of the regenerating blower, speed of the furnace hood blower, speed of the cooling blower, and speed of the drying blower. The energy consumption data for each process includes electricity consumption per unit of gas consumption per unit of gas and finished product quantity. The finished product qualification rate indicators are compressive strength, drum index, and particle size.

3. The method for automatic roasting and intelligent process control of a belt roaster according to claim 1, characterized in that, The business slice is triggered by at least one of the following events: change in product type, change in production specifications, change in production line status, change in ratio, change in process production status, material addition / reduction operation, and timeout. The sampling interval of the time slice is calculated by dividing the trolley length by the real-time machine speed.

4. The method for automatic roasting and intelligent process control of a belt roaster according to claim 1, characterized in that, The requirements for compressive strength, drum index, and particle size include: compressive strength in the range of 2400-2600N, drum index in the range of 88-92%, and particle size of 10-16mm in the range of 73-77%. The minimum energy consumption of the process is the sum of electricity consumption and gas consumption divided by the quantity of finished product.

5. The method for automatic roasting and intelligent process control of a belt roaster according to claim 1, characterized in that, If the corresponding raw material code is not found in the model history database, add the raw material number and time slice record to the model history database.

6. The method for automatic roasting and intelligent process control of a belt roaster according to any one of claims 2-5, characterized in that, The control strategy also includes pressure balancing between the drying section and the extraction section: By adjusting the opening of the valve connecting the outlet pipe of the drying blower to the furnace hood and the speed of the second main exhaust fan, the measured pressure inside the drying section hood is adjusted. P 鼓干段 Measured pressure inside the desiccant hood P 抽干段 ratio D Stable within the preset equilibrium range [ D min , D max ]Inside; Specifically, based on the current raw material code and current business segment conditions, the system queries the historical database for quality qualified record data segments that meet the predefined finished product qualification rate index under the same business segment label. It then extracts the target data segment with the lowest process energy consumption and statistically analyzes the data within that target data segment. D The value distribution is selected based on the frequency of occurrence being ≥90%. D value range as initial D min , D max .

7. An automatic roasting and intelligent process control device for a belt roaster applied to the method described in any one of claims 1-6, characterized in that, include: The historical data slice acquisition and modeling module is used to collect monitoring data samples, which include raw material codes, slice codes, process monitoring item values, measured signal values ​​of equipment parameters, process energy consumption data, and finished product qualification rate index data. A model history library is established according to business slices and time slices. Business slices are triggered by production changes, and time slices are triggered by sampling intervals. Based on the model history library, an automatic roasting big data model is established using a three-layer feedforward neural network. The target process parameter determination module is used to query records with the same business slice label in the model history database based on the current raw material code and the current business slice conditions. If a corresponding record exists, the finished product code field divided by time slice is filtered out. From the finished product code field, the quality qualified records that meet the predefined finished product qualification rate index are selected. Then, the record with the lowest process energy consumption is extracted from the quality qualified records. The process monitoring item values ​​and measured signal values ​​of equipment parameters in this record are used as predicted equipment adjustment parameters. The feedback verification and online correction module is used to collect the current process monitoring item values ​​and measured signal values ​​of equipment parameters in real time, perform flexible weighted deviation calculation with the predicted equipment adjustment parameters, and adjust and issue equipment parameters online based on the flexible weighted deviation calculation results. The rolling predictive control module is used to periodically determine, verify, and correct target process parameters through an automatic roasting big data model. It uses the predictive model to estimate future deviations in the process and incorporates a softening coefficient for rolling optimization to determine the current control strategy. The control strategy includes adjusting equipment parameters to minimize the deviation between the controlled variable and the expected value. A control execution module is used to execute the control strategy, and the control execution module adjusts the fan speed, valve opening, machine speed and material thickness; The data management module is used to add a record of the corresponding raw material code and time slice to the model history database when the record does not exist in the model history database.

8. The automatic roasting and intelligent process control device for the belt roaster according to claim 7, characterized in that, The process monitoring values ​​are: temperature inside the drying section hood, temperature inside the extraction section air box, temperature inside the preheating section hood, temperature inside the preheating section air box, temperature inside the calcination section hood, temperature inside the calcination section air box, temperature inside the homogenizing section air box, temperature inside the second cooling section hood, pressure inside the drying section hood, pressure inside the drying section air box, pressure inside the extraction section hood, pressure inside the preheating section hood, pressure inside the calcination section hood, pressure in the main regenerative air pipe inside the first cooling section hood, and pressure inside the second cooling section hood. The measured signal values ​​of the equipment parameters are: roasting machine speed, roasting material thickness, opening of the hot air valve on the outlet pipe of the drying blower, opening of the butterfly valve in the exhaust section, opening of the butterfly valve in the preheating section, opening of the butterfly valve in the secondary cooling section, opening of the valve connecting the outlet pipe of the drying blower to the furnace hood, opening of the butterfly valve in the exhaust section, heat source flow rate of the 16 burners on each side of the roasting section, speed of the No. 1 main exhaust fan, speed of the No. 2 main exhaust fan, speed of the regenerating blower, speed of the furnace hood blower, speed of the cooling blower, and speed of the drying blower. The energy consumption data for each process includes electricity consumption per unit of gas consumption per unit of gas and finished product quantity. The finished product qualification rate indicators are compressive strength, drum index, and particle size; Business segments are triggered by at least one of the following events: changes in product type, changes in production specifications, changes in production line status, changes in proportion, changes in process production status, material addition / reduction operations, and the expiration of a scheduled time. The sampling interval of the time slice is calculated by dividing the trolley length by the real-time machine speed.

Images