A method for subzone calcination of high purity alumina

Abstract

The application discloses a high-purity alumina subarea calcination method and belongs to the field of high-purity alumina continuous calcination and industrial furnace subarea control. In order to solve the problems that the uneven thermal field in the furnace, local overburning or underburning and the difficulty in synchronous control of quality indexes are caused by the fluctuation of feed particle size, layer thickness and precursor water content, the application realizes the technical effects of reducing the uneven thermal field and stabilizing the whiteness, ignition loss, specific surface area, particle size and alpha phase proportion by collecting subareas, coupling subarea maps with hot materials, predicting the map time sequence, calculating risks and rolling solving adjustment actuators.

Description

Technical Field

[0001] This invention relates to the field of continuous calcination of high-purity alumina and zoned control of industrial furnaces, and particularly to a zoned calcination method for high-purity alumina. Background Technology

[0002] High-purity calcined alumina typically requires heat treatment processes such as dehydration, phase transformation, α-phase transformation, and grain stabilization to be completed in a continuous calcination furnace. The temperature distribution in each temperature zone of the furnace, the material residence time, and the heat transfer state of the material layer directly affect the whiteness, loss on ignition, specific surface area, particle size, and α-phase ratio of the output. Current continuous calcination control largely relies on fixed process curves, feedback from single or a small number of temperature measurement points, and manual experience corrections, which can meet basic calcination requirements when the feed conditions are relatively stable.

[0003] However, in actual continuous production, the precursor particle size distribution, moisture content, bed thickness, and feed rate fluctuate with batch and operating conditions, causing changes in the heat load, heat transfer resistance, and material residence time in different zones. Single-point temperature feedback is difficult to reflect the internal thermal history of the bed and the thermal coupling state of adjacent zones in a timely manner, which can easily lead to local overburning causing abnormal grain growth, local underburning causing insufficient α-phase transformation, and making it difficult to synchronously stabilize indicators such as whiteness, loss on ignition, specific surface area, and particle size.

[0004] Therefore, a high-purity alumina partition calcination method is needed to overcome the shortcomings of the existing technology. Summary of the Invention

[0005] One objective of this invention is to propose a high-purity alumina zonal calcination method. Addressing the problems of uneven furnace thermal field, localized over- or under-calcination, and difficulties in synchronously controlling quality indicators caused by fluctuations in feed particle size, material layer thickness, and precursor moisture content during continuous calcination in existing technologies, this invention proposes a technical solution based on zonal data acquisition, thermal-material coupled zonal diagrams, zonal disturbance weight matrices, graph time-series prediction, generation of under- and over-calcination risks, and rolling solution of control quantities. This invention achieves the technical effects of early identification of zonal thermal load disturbances, reducing the impact of single-point temperature measurement lag, and stabilizing multiple quality indicators of the output material.

[0006] This invention provides a method for partitioned calcination of high-purity alumina, comprising: S1, dividing a continuous calcining furnace into a preheating and dehydration zone, a phase transformation induction zone, an α-phase transformation zone, a grain stabilization zone, and a cooling and slow-release zone along the material travel direction; collecting material state data and thermal state data for each zone; collecting discharge quality data corresponding to the sampling period or discharge batch; and obtaining historical control quantities corresponding to the sampling period or control period to obtain partitioned data; S2, constructing a heat-material coupling partition diagram with partitions as nodes and heat transfer relationships between adjacent partitions as edges based on the partitioned data; and determining the particle size distribution deviation relative to the target particle size distribution, the moisture content difference relative to the target moisture content, and the material thickness relative to the target material layer thickness. S3. Correct the node weights and edge weights for layer thickness deviation to obtain the partition disturbance weight matrix; S4. Input the hot material coupled partition diagram, partition disturbance weight matrix and historical control quantities into the time series prediction model of the diagram, and output the temperature field, material thermal history and material layer heat flux of each partition in the future control cycle; S5. Generate the underburning risk index, overburning risk index and quality offset based on the temperature field, material thermal history, material layer heat flux and preset quality target range of each partition; S6. Based on the underburning risk index, overburning risk index and quality offset, perform rolling solution on the partition temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure adjustment, feed rate and material layer thickness correction, and adjust the corresponding actuators according to the solution results.

[0007] Optionally, S1 includes: Obtain the current product specification number, current furnace type number, current control cycle, and zone number based on the current production task; In each zone, the precursor particle size distribution, precursor moisture content, material layer thickness, feed rate, zone temperature, fuel flow rate, hot air flow rate, furnace pressure, and exhaust oxygen content are collected. The discharge whiteness, loss on ignition, specific surface area, particle size, and α phase ratio obtained from the discharge end or discharge batch are used as the discharge quality data. Read the historical control quantity corresponding to the sampling period or control period from the control history record or the execution record of the previous control period, and associate the historical control quantity with the corresponding partition number and sampling time or control period; According to the corresponding relationship of sampling period or the corresponding relationship of discharge batch, the material state data, the thermal state data and the discharge quality data are time-aligned or batch-aligned, and the discharge quality data corresponding to the discharge batch is associated with the sampling time window of the corresponding material passing through each partition to obtain the partition collection data with partition number and sampling time.

[0008] Optionally, S2 includes: The temperature, material layer thickness, material residence time, and heat load of each zone are used as node features, and the hot air flow direction, temperature difference, and material transfer time between adjacent zones are used as edge features. The material residence time, heat load, hot air flow direction, and material transfer time are determined based on the data collected from the zones and the furnace configuration parameters associated with the furnace type number and zone number. Read the target particle size distribution, target moisture content, and target bed thickness from the process formulation associated with the current product specification number, furnace type number, and zone number; The difference between the current particle size distribution and the target particle size distribution is determined as the particle size distribution deviation; the difference between the current moisture content and the target moisture content is determined as the moisture content difference; and the difference between the current material layer thickness and the target material layer thickness is determined as the material layer thickness deviation. The moisture content difference is mapped to the dehydration heat load increment, the particle size distribution deviation is mapped to the phase transformation hysteresis increment, and the material layer thickness deviation is mapped to the heat transfer resistance increment and residence time offset, and the node weight and edge weight are corrected accordingly.

[0009] Optionally, S3 includes: The node features, edge features, partition disturbance weight matrix, partition temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure adjustment, feed rate, and material layer thickness correction value of the thermal coupling partition diagram are used as model inputs. The time series prediction model is used to jointly encode the heat transfer relationship between adjacent zones and the time series relationship in the material travel direction to obtain the temperature field of each zone, the thermal history of the material, and the heat flux of the material layer at multiple future sampling times. Furthermore, the graph time series prediction model is trained using historical calcination samples, which include historical partition data, historical partition perturbation weight matrix, historical control variables, historical discharge quality data, and future partition temperature field labels, material thermal history labels, and material layer heat flux labels corresponding to each historical control cycle. During training, the loss function is composed of the prediction error of the future zone temperature field, the prediction error of the material thermal history, the prediction error of the material layer heat flux, and the prediction error of the discharge quality relative to the historical discharge quality data. The predicted discharge quality is generated based on the temperature field of each zone, the material thermal history, and the material layer heat flux. The trained model parameter version is associated with and stored with the corresponding sampling period, furnace type number and product specification number; When step S3 is executed, the model parameter version that matches the current sampling period, current furnace type number, and current product specification number is called.

[0010] Optionally, S4 includes: The material residence time is determined based on the feeding rate and material layer thickness in the data collected from the partitions, as well as the partition length associated with the furnace type number and partition number. The material thermal history is input into the α-phase conversion effective thermal history integration channel, and the α-phase conversion sufficiency index is calculated according to the temperature duration of the phase change induction zone and the α-phase conversion zone, the material residence time and the heat flux of the material bed. The material thermal history is input into the grain growth risk thermal history integration channel. The grain growth risk index is calculated based on the duration of temperature exceeding the upper boundary of the grain growth initiation temperature range within the grain stability region and the heat flux of the material layer. The grain growth initiation temperature range is a preset temperature range associated with the current product specification number and furnace type number. Based on the temperature field of each zone, the thermal history of the material, and the heat flux of the material layer, a predicted value for the discharge quality is generated. The predicted value for the discharge quality includes predicted whiteness, predicted loss on ignition, predicted specific surface area, predicted particle size, and predicted α phase ratio. The preset quality target range includes whiteness target range, ignition loss target range, specific surface area target range, particle size target range, and α phase target range; The underburn risk index is generated based on the α-phase conversion sufficiency index, the deviation of the predicted loss on ignition from the target range, and the deviation of the predicted α-phase ratio from the target range. The overburn risk index is generated based on the grain growth risk index, the deviation of the predicted specific surface area from the target range, the deviation of the predicted particle size from the target range, and the deviation of the predicted whiteness from the target range. The mass offset is generated based on the boundary difference between the predicted output quality and the preset quality target range. Furthermore, the quality offset is also corrected according to the quality confidence interval, specifically including: from the historical batches that match the current product specification number and furnace type number, the historical qualified batches and historical deviation batches are divided according to whether the whiteness, loss on ignition, specific surface area, particle size and α phase ratio are all within the corresponding quality target interval. Using the absolute values ​​of whiteness prediction error, loss on ignition prediction error, specific surface area prediction error, particle size prediction error, and α phase proportion prediction error in the historical qualified batches and the historical deviation batches respectively, construct the corresponding quality index qualified batch non-consistency fractional subsets and deviation batch non-consistency fractional subsets. The whiteness prediction error, loss on ignition prediction error, specific surface area prediction error, particle size prediction error, and α phase proportion prediction error are respectively the differences between the predicted values ​​of the corresponding quality index of the historical batches and the corresponding measured values ​​in the historical output quality data. According to the preset target coverage rate associated with the current control cycle, the quantile values ​​of qualified batches and the quantile values ​​of deviating batches are determined from the subset of qualified batches with inconsistent scores and the subset of deviating batches with inconsistent scores, respectively. The last quantile value after sorting the qualified batch quantile values ​​and the deviating batch quantile values ​​in ascending order of value is used as the quantile value of the corresponding quality indicator. Using the predicted value of the quality index corresponding to the current control cycle as the interval benchmark, the confidence interval of the quality index is generated according to the quantile value of the corresponding quality index. Thus, confidence intervals for whiteness, ignition loss, specific surface area, particle size and α phase proportion are generated respectively. When the lower bound of any quality confidence interval is less than the lower bound of the corresponding target interval, the lower limit offset component of the corresponding quality index is generated based on the difference between the lower bound of the corresponding target interval and the lower bound of the quality confidence interval. When the upper bound of any quality confidence interval is greater than the upper bound of the corresponding target interval, the upper limit offset component of the corresponding quality index is generated based on the difference between the upper bound of the quality confidence interval and the upper bound of the corresponding target interval. The lower limit offset component and the upper limit offset component are incorporated into the mass offset.

[0011] Optionally, S5 includes: The control constraints are: the predicted temperature difference between adjacent zones within the future control cycle does not exceed a preset zone temperature difference threshold; the predicted α phase ratio is within the α phase target range; the predicted loss on ignition is within the loss on ignition target range; the predicted specific surface area is within the specific surface area target range; the predicted particle size is within the particle size target range; and the predicted whiteness is within the whiteness target range. The preset zone temperature difference threshold is a preset threshold associated with the current furnace type number, adjacent zone combination, product specification number, and control cycle. The predicted α phase ratio, the predicted loss on ignition, the predicted specific surface area, the predicted particle size, and the predicted whiteness are determined by the predicted discharge quality value, which is generated based on the temperature field of each zone, the thermal history of the material, and the heat flux of the material layer. The zone temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure regulation, feed rate, and material layer thickness correction are constrained within the allowable adjustment range and allowable rate of change associated with the current furnace type number, actuator number, or product specification number; The penalty weights of the rolling optimization objective function are determined based on the underburn risk index, the overburn risk index, and the quality offset. The rolling optimization objective function is generated based on the quality offset objective item, the underburn risk objective item, the overburn risk objective item, and the control quantity adjustment range objective item, and the penalty weights are used to weight the corresponding objective items. Within each control cycle, the rolling optimization objective function is solved to obtain the zone temperature setpoint, the fuel flow rate, the hot air flow rate, the furnace pressure adjustment, the feed rate, and the material layer thickness correction. Furthermore, the rolling solution also includes partition response lag compensation, specifically including: determining the response lag time of each partition based on the time difference between the time of change of historical control quantity and the time of partition temperature response; The current material residence time is determined based on the feeding rate and material layer thickness in the data collected from the partitions, as well as the partition length associated with the furnace type number and partition number. The control lead time is determined based on the response lag time and the current material residence time. The solution time windows for the zone temperature setpoint, the fuel flow rate, and the hot air flow rate, as well as the effective time of the control quantity of the corresponding actuator, are adjusted according to the control advance amount, so that the corresponding actuator completes the control quantity adjustment before the material enters the zone to be adjusted corresponding to the control advance amount. Furthermore, when the underburn risk index and the overburn risk index both exceed their respective preset risk thresholds, conflict handling control quantities are generated in the order of first, second and third order, where the preset risk thresholds are thresholds associated with the current product specification number and control cycle. The first order is to constrain the predicted α phase ratio and the predicted scorch reduction within the corresponding target interval; The second sequence is to constrain the set value of the grain stabilization zone temperature to not exceed the upper boundary of the grain growth initiation temperature range. The grain growth initiation temperature range is a preset temperature range associated with the current product specification number and furnace type number, and is used as a variable boundary correction amount to participate in the generation of feasible control quantities. When the first order and the second order cannot be satisfied simultaneously, the equipment safety boundary associated with the current furnace type number is taken as an unrelaxable constraint, and preset relaxation variables and penalty weights are set according to the order of priority of the first order over the second order, so as to perform feasible relaxation on the corresponding constraints in the first order and the second order and generate feasible control quantities. The feasible control quantities are constrained by the equipment safety boundary and preset allowable deviations. The preset allowable deviations include the predicted α phase ratio allowable deviation, the predicted ignition allowable deviation, the predicted specific surface area allowable deviation, the predicted particle size allowable deviation, the predicted whiteness allowable deviation, the grain stability zone temperature allowable deviation, and the control quantity adjustment allowable deviation associated with the current product specification number, furnace type number, and control cycle. After satisfying the first order and the second order, or after generating the feasible control quantity, the remaining mass offset is formed based on the boundary differences that have not been eliminated in the mass offset. The third step is to generate a feed rate adjustment amount and a material layer thickness correction amount based on the remaining mass offset, and to constrain the feed rate adjustment amount and the material layer thickness correction amount within the feed rate adjustment range and material layer thickness correction range associated with the current furnace type number. The conflict handling control variable is used as the priority constraint and variable boundary correction variable of the rolling optimization objective function.

[0012] The beneficial effects of this invention are: 1. By dividing the continuous calcining furnace into multiple functional zones and constructing a thermal-material coupling zone diagram based on material state data, thermal state data, and discharge quality data, the fluctuations in particle size distribution, moisture content, and material layer thickness can be transformed into node weights and edge weight corrections, so that the furnace thermal field control no longer relies solely on single-point temperature feedback.

[0013] 2. The graphical time-series prediction model outputs the zoned temperature field, material thermal history, and material layer heat flux for the future control cycle. The underburning risk index and overburning risk index are generated by the α-phase transformation effective thermal history integral channel and the grain growth risk thermal history integral channel, which can respectively constrain the risks of insufficient α-phase transformation and abnormal grain growth.

[0014] 3. By incorporating predicted quality indicators, quality confidence intervals, zonal response lag compensation, and conflict handling control quantities into the rolling solution process, it is possible to adjust the temperature, fuel, hot air, furnace pressure, feed rate, and material layer thickness correction quantities in conjunction with the simultaneous existence of feed disturbance and thermal field lag, thereby improving batch quality stability. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a method for calcining high-purity alumina in a partitioned manner.

[0016] Figure 2 This is a flowchart of step S3, time series prediction, of the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2A method for partitioned calcination of high-purity alumina includes: S1, dividing the continuous calcination furnace into a preheating and dehydration zone, a phase transformation induction zone, an α-phase transformation zone, a grain stabilization zone, and a cooling and slow-release zone along the material travel direction; collecting material state data and thermal state data for each zone; collecting discharge quality data corresponding to the sampling period or discharge batch; and obtaining historical control quantities corresponding to the sampling period or control period to obtain partitioned data; S2, constructing a heat-material coupling partition diagram with partitions as nodes and heat transfer relationships between adjacent partitions as edges based on the partitioned data; and determining the particle size distribution deviation relative to the target particle size distribution, the moisture content difference relative to the target moisture content, and the material layer thickness relative to the target material layer thickness. S3. Correct the node weights and edge weights to obtain the partition disturbance weight matrix; S4. Input the hot material coupled partition map, partition disturbance weight matrix and historical control quantities into the time series prediction model of the map, and output the temperature field, material thermal history and material bed heat flux of each partition in the future control cycle; S5. Generate the underburning risk index, overburning risk index and quality offset based on the temperature field, material thermal history, material bed heat flux and preset quality target range of each partition; S6. Based on the underburning risk index, overburning risk index and quality offset, perform rolling solution on the partition temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure adjustment, feed rate and material bed thickness correction, and adjust the corresponding actuators according to the solution results.

[0019] In this specific embodiment, S1 includes: The control system reads the current production task from the production execution system and generates the current product specification number, the current furnace type number, the current control cycle number, and the partition number that corresponds one-to-one with the furnace body configuration parameter table. The furnace body configuration parameter table includes at least the partition length, measuring point number, actuator number, and data acquisition channel number for each partition to ensure that subsequent data acquisition and historical control quantity reading have a definite index relationship. The continuous calcining furnace is divided into a preheating and dehydration zone, a phase change induction zone, and a phase change induction zone along the material travel direction. The system is divided into a phase transformation zone, a grain stabilization zone, and a cooling slow-release zone. Material state acquisition points and thermal state acquisition points are set in each zone. Material state data, such as precursor particle size distribution, precursor moisture content, material layer thickness, and feed rate, are collected according to the sampling period. The precursor particle size distribution is obtained by sampling a quantitative amount of material at regular intervals from the side-line sampling mechanism of the zone and sending it to the online particle size analyzer to obtain a volume fraction sequence divided into preset particle size ranges. The precursor moisture content is output by the online moisture meter of the zone and converted by temperature compensation through a calibration curve bound to the current product specification number. The material layer thickness is output by the material layer thickness measuring device at the top of the furnace body of the zone at a fixed measuring point and its validity is verified within the same sampling period. The feed rate is jointly verified by the speed feedback signal of the feeding mechanism and the feed mass flow rate per unit time to eliminate single signal distortion caused by slippage or blockage. Thermal status data are collected according to the sampling period, including zone temperature, fuel flow rate, hot air flow rate, furnace pressure, and exhaust oxygen content. Zone temperature is collected by a multi-point thermocouple array in that zone and aggregated according to the measurement point number to obtain the measured value of zone temperature. Fuel flow rate and hot air flow rate are output by the corresponding mass flow meter and bound to the corresponding zone according to the actuator number. Furnace pressure is collected by the zone pressure transmitter to characterize air leakage and draft changes. Exhaust oxygen content is collected by an online oxygen analyzer in the exhaust pipeline to reflect the combustion and air leakage status. The discharge quality data is collected by discharge batch and bound to the discharge batch number. The discharge batch is divided with the current control cycle as the boundary and is automatically sampled at the discharge end to obtain the discharge whiteness, loss on ignition, specific surface area, particle size, and other parameters. The corresponding measured results are then written into the quality database; Historical control quantities are read from the control history records according to the sampling time or control cycle, including the zone temperature setpoint, fuel flow setpoint, hot air flow setpoint, furnace pressure adjustment, feed rate setpoint, and material layer thickness correction of the previous control cycle. Each historical control quantity is recorded and associated with the zone number and control cycle number to form a traceable control sequence. To align material status data, thermal status data, and discharge quality data, the system timestamps all collected data using a unified clock and aggregates multi-source data within the sampling period into a single partition collection record according to partition number. Simultaneously, it inversely correlates the discharge quality data corresponding to each discharge batch with the sampling time window of the material passing through each partition. The sampling time window is determined by the following formula: ; in, The batch number of the output material is The materials are in the partition number The start time of the time window corresponding to the partition entry point. The batch number of the output is The materials are in the partition number The end time of the time window corresponding to the partition exit. The batch number of the output is The time of material discharge sampling. This represents the total number of partitions. For partition summation index and satisfying the material movement direction from 1 to... Increasing, The partition number is The partition length is provided by the furnace configuration parameter table. The batch number of the output material is The corresponding material travel speed is obtained from the speed conversion relationship solidified by the control system based on the collected feed speed value. The partition number to be associated and satisfying This refers to the batch number of the material being discharged; The system will fall into each partition The partition collection records and the batch number of the output are as follows The output quality data is bound to the data, thereby forming a partition collection data that simultaneously includes partition number, sampling time, material status data, thermal status data, output quality data, and historical control quantities, and is written into the process database for subsequent steps.

[0020] In this specific embodiment, S2 includes: The control system collects data based on zones, and the current control cycle is numbered as follows: A thermal coupling partitioning diagram is constructed for the continuous calcining furnace at all times. The thermal coupling partitioning diagram uses partitions as a set of nodes and is numbered according to partition. From 1 to the total number of partitions Sequential arrangement, nodes Corresponding steps Measured temperature values ​​of this partition after time alignment Material layer thickness Material residence time With heat load Four types of node characteristics, among which material residence time The partition length is determined by the furnace body configuration parameter table. With steps Material travel speed obtained by converting the collected and verified feed rate The calculated heat load The fuel heat input is obtained by multiplying the fuel mass flow rate and the lower heating value of the fuel in this zone, and then adding the sensible heat input of the hot air by the hot air mass flow rate, the specific heat at constant pressure, the hot air temperature, and the difference between the hot air temperature and the reference temperature. The lower heating value of the fuel, the specific heat at constant pressure, and the reference temperature are given by a physical property parameter table bound to the current furnace type number and are respectively taken as... and and to and Normalize according to a fixed dimension range to form a node feature vector with consistent numerical scale for use in subsequent models; The heat-coupled partitioning diagram uses the heat transfer relationship between adjacent partitions as the edge set. For each pair of adjacent partitions... Simultaneously establish edges With edge To express bidirectional thermal coupling, the side features include three categories: hot air flow direction, temperature difference, and material transfer time. The hot air flow direction is determined by the relationship between the compaction measurements of adjacent furnace zones and is discretely selected. It means that among them Indicates the mainstream direction of hot air from point to and Indicates the mainstream direction of hot air from point to The temperature difference is determined by the difference between the measured temperatures of adjacent zones and is used to characterize the potential difference driving heat transfer. The material transfer time is determined by the material conveying distance and the material travel speed between adjacent zones. Determine and use the timescale for characterizing the propagation of disturbances along the material direction; While constructing the thermal coupling zoning diagram, the control system reads the target particle size distribution from the process formula table bound to the current product specification number, current furnace type number, and zoning number. Target moisture content With the target material layer thickness Among them, the target particle size distribution Adoption and Steps Consistent particle size range division and storage as a volume fraction sequence, followed by steps The obtained current particle size distribution volume fraction sequence and The particle size distribution deviation is obtained by summing the absolute differences of each term, and then proceeds in steps... The current water content collected is consistent with The difference yields the moisture content difference, which is obtained through the steps. The current material layer thickness collected and The difference is used to obtain the material layer thickness deviation, and the three types of deviations are normalized into a dimensionless disturbance quantity to eliminate dimensional differences. The process formula table simultaneously stores a set of coefficients that map disturbances to the increments of dehydration endothermic load, phase transformation hysteresis, heat transfer resistance, and residence time offset. This set of coefficients is determined according to the functional type of the zones and takes fixed values ​​in this embodiment. The coefficients for the preheating dehydration zone are... The coefficient of the phase transition induced region is taken as The coefficient of the phase transformation region is taken as The coefficient of the grain stability region is taken as The coefficient of the cooling slow-release zone is taken as Based on this, partition perturbation weight factors are generated. The partitioned disturbance weighting factor is designed to uniformly handle changes in dehydration endothermic load caused by fluctuations in moisture content, changes in phase transformation hysteresis caused by fluctuations in particle size, and changes in heat transfer resistance and residence time caused by fluctuations in bed thickness. Determine using the following formula: ; in, The partition number is In the control cycle number The partition perturbation weighting factor at that time, The partition number is The corresponding moisture content disturbance mapping coefficient is given by the process formula table. The partition number is The corresponding particle size perturbation mapping coefficients are given by the process formulation table. The partition number is The corresponding material layer thickness disturbance mapping coefficient is given by the process formula table. The partition number is In the control cycle number The moisture content at that time is a dimensionless disturbance and is calculated by dividing the moisture content difference by the target moisture content. get, The partition number is In the control cycle number The dimensionless perturbation of particle size at that time is obtained by normalizing the particle size distribution deviation according to the total deviation by volume fraction. The partition number is In the control cycle number is The material layer thickness is a dimensionless disturbance and is calculated by dividing the material layer thickness deviation by the target material layer thickness. get, For partition numbering and satisfying Number the control cycle; The control system is as described The node weights and edge weights are corrected to form a partition perturbation weight matrix. ,in The diagonal elements are taken from each partition. To correct node weights, The off-diagonal elements corresponding to adjacent partitions are taken as the arithmetic mean of the perturbation weight factors of the two adjacent partitions to correct the heat transfer edge weights of the adjacent partitions. The remaining elements are set to zero to maintain the graph structure where only adjacent partitions are coupled. The node feature matrix, edge feature set, edge connection relationship and partition perturbation weight matrix are then combined. As the same control cycle The thermal coupling partition diagram data structure is written into the process database for each step. Direct call.

[0021] In this specific embodiment, S3 includes: The control system is numbered in the current control cycle. The thermal coupling zoning map data structure is read from the process database and organized into a model input sequence. The thermal coupling zoning map data structure includes a node feature matrix. Edge connection relationship Edge feature set and partition perturbation weight matrix ,in By partition number Sequentially assemble each partition during the control cycle Measured temperature values ​​of the zones Material layer thickness Material residence time With heat load Form a numerical vector. It is stored as a set of bidirectional edges between adjacent partitions, and its node index is consistent with the partition number. Store three types of edge features—hot air flow direction, temperature difference, and material transfer time—using an edge index, and then... One-to-one correspondence, Storage partition perturbation weighting factors are based on diagonal and adjacent off-diagonal rules; The control system simultaneously reads the previous control cycle number as Historical control values ​​are aligned by partition number to form a control vector sequence. The The parameters are obtained by splicing together the zone temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure adjustment, feeding speed and material layer thickness correction in a fixed order, and the order is fixed in the control software to ensure consistency between training and online inference. To ensure the model receives sufficient temporal context, the control system uses a sampling period... Length of the solidified structure history time window And form the input sequence ,in and The timing index is shifted by one beat to express the impact of the control action on the state of the subsequent process. The control system retrieves key values ​​from the model parameter version library ( The system retrieves and loads the graph time-series prediction model parameter version that matches the current operating condition, based on the current furnace type number and current product specification number. The graph time-series prediction model employs a deterministic spatiotemporal co-encoder-decoder architecture and consists of a parameter set. Define, where the spatial encoder parameters The corresponding two-layer graph attention network has a hidden dimension of 64 in each layer and uses 4 attention heads. The temporal encoder parameters are as follows. Corresponding to a single-layer gated loop unit with a hidden dimension of 128, decoder parameters Corresponding to two fully connected layers to output multi-step prediction sequences, quality prediction header parameters The corresponding two-layer fully connected network with 32 hidden units in the middle and ReLU activation is used to map the predicted temperature field of each zone, the thermal history of the material and the heat flux of the material layer into the predicted value of the output quality. The model first bases on each historical moment. and Message passing is performed between adjacent partitions, and node inputs are weighted during message weight calculation. Scaling diagonal elements and side messages The adjacent off-diagonal elements are scaled to allow the effects of moisture content, particle size distribution, and material layer thickness disturbances on heat transfer coupling strength and zoned thermal response to be spatially represented. The spatial representation sequence and the control vector sequence are then input into a gated loop unit to encode the time-series relationship along the material's movement direction and form the joint hidden state for the current control cycle. Subsequently, the decoder calculates the prediction step size. The rolling prediction window outputs the predicted temperature fields of each zone at multiple future sampling times all at once. Material thermal history prediction Prediction of heat flux of material bed ,in A matrix arranged by partition number and prediction step index, and its elements Indicates the partition number is In the future Predicted temperature values ​​for each zone at each sampling time. A matrix arranged by partition number and prediction step index, and its elements This indicates the representative material unit corresponding to this partition in the future. The thermal history state variables at each sampling time are defined as a standardized representation of the temperature-time cumulative quantity according to the training labels. A matrix arranged by partition number and prediction step index, and its elements This indicates that the material layer in this zone will be in the future. Predicted net heat flux per unit area at each sampling time; The graph time-series prediction model is trained offline using historical calcination samples. These historical calcination samples consist of historical partition data, historical partition perturbation weight matrices, historical control variables, historical output quality data, and future partition temperature field labels corresponding to each historical control cycle. Material thermal history label With material layer heat flux label The system is structured as follows: the future zone temperature field label is obtained by aggregating the measured temperatures of multi-point thermocouple arrays in each zone; the material thermal history label is calculated in the label construction script by combining the measured temperatures and residence times of each zone according to a fixed integration rule; and the material layer heat flux label is derived from the zone energy balance calculation script based on the fuel heat input, hot air sensible heat input, and zone temperature changes. During training, the Adam optimizer is used, and the learning rate is fixed. The batch size is 64, the number of training rounds is 200, and the mean and standard deviation of all input features are used to evaluate the performance of the data. - Score standardization ensures that the data distribution is consistent between training and inference. The training loss function is determined by the following formula: ; in, To train the scalar values ​​of the loss function, As the weighting of the temperature field prediction error for the region, Weights for the prediction error of material thermal history. As the weighting factor for the prediction error of the heat flux of the material bed, As the weight for the prediction error of the output quality, The model outputs a future zone temperature field prediction matrix and is compared with the label matrix. The partition number and the prediction step index are identical. The model outputs a prediction matrix of the future material thermal history, along with the label matrix. Same shape, The model outputs a prediction matrix of future bed heat flux and is compared with the label matrix. Same shape, For quality prediction head The generated output quality prediction vector for the input includes, in sequence, predicted whiteness, predicted loss on ignition, predicted specific surface area, predicted particle size, and predicted... The proportion consists of five components. The vector of historical output quality data corresponding to the same control cycle as the training samples, and the order of the components is the same as... Consistent The squared L2 norm of a matrix or vector is obtained by summing the elements after squaring each element. After offline training is completed, the control system will set the parameters. Along with its applicable The furnace type number and product specification number are written into the model parameter version library and a traceable version number is generated. The online execution steps are then performed. Call the matching version number Perform one forward inference on the input sequence and output the temperature field of each zone, material thermal history, and bed heat flux in the future control cycle for use in the next step. Risk calculation and quality prediction.

[0022] In this specific embodiment, S4 includes: The control system is numbered in the current control cycle. Read future prediction sequences in time and And read the material travel speed calculated from the feed rate. The partition lengths of each zone in the furnace body configuration parameter table Each partition is numbered as Material residence time according to The timescale was determined and used as the integral timescale for the thermal history calculation. Subsequently, the predicted sequences were grouped according to the functional set of the partitions, including the phase transition induced region and... The set of partition numbers corresponding to the phase transformation region is denoted as The set of partition numbers corresponding to the grain stable region is denoted as The threshold parameters used in this embodiment are read and fixed from the process formula table bound to the current product specification number and the current furnace type number. These threshold parameters include... Lower limit of effective temperature for phase transformation K and the upper boundary of the grain growth initiation temperature range Based on this, the control system implements two risk thermal history integral channels, the first of which is... Phase transformation effective thermal history integral channel, which is indexed at each prediction step Inner Traversal The partition number is used to read the predicted temperature of that partition. With predicted heat flux of the bed ,when When With sampling period Multiply to obtain the effective heat contribution of this prediction step and calculate based on the material's residence time within the partition. After normalization, the sum is obtained Phase transformation sufficiency index ,when The effective heat contribution of the first prediction step is set to zero to reflect that the phase change is not effectively driven. The second path is the integral channel for the thermal history of grain growth risk, which is indexed in each prediction step. Inner Traversal Partition number and read and ,when When and Multiplying these values ​​yields the overtemperature heat contribution of this prediction step, and summing them up yields the grain growth risk index. ,when The predicted overheating contribution is set to zero to reflect that the grain growth risk has not been triggered. Control system based on and Generate output quality prediction vector ,in To predict whiteness, To predict burn-off, To predict specific surface area, To predict granularity, For prediction In comparison, the From steps The medium-quality prediction head outputs the prediction target at the end of the prediction window. The corresponding batch quality; The control system reads the preset quality target range from the process formula table and solidifies it as the whiteness target range. Target range for ablation Specific surface area target range Particle size target range and Phase target range The whiteness offset component, ignition offset component, specific surface area offset component, and particle size offset component are calculated according to the rule that "if the offset is within the target range, it is zero; if it is below the lower bound, the offset is equal to the difference between the lower and upper bounds; and if it is above the upper bound, the offset is equal to the difference between the upper and lower bounds." The mass offset vector is obtained by concatenating the proportional offset components. ; In the calculation of the underburning risk index, the control system will Phase transformation sufficiency index The sufficiency threshold of curing according to the process formula table Normalization is performed to obtain the underburned driving component, and then... Relative ablation target range Upper bound offset and relatively Phase target range The under-bound offsets, after being normalized, are weighted and compared with the underburned driving components. Weighted summation yields the underburn risk index In the calculation of the overheating risk index, the control system incorporates the grain growth risk index. Risk threshold for curing according to process formula Normalization is performed to obtain the overburning driving component, and then... Target range of relative surface area Under-bounded offset Relative particle size target range Upper bound offset and Relative whiteness target range The under-bound offsets, after being normalized, are weighted and compared with the overburning driving components. The overheating risk index is obtained by weighted summation. ; During the quality offset confidence interval correction, the control system reads the most recent 300 historical batches from the historical batch library that matches the current product specification number and the current furnace type number, and sorts them by "whiteness, loss on ignition, specific surface area, particle size and..." The proportions of each batch fall within their respective quality target ranges. The system divides historical qualified batches into sets and historical deviation batches. For each quality indicator, it constructs a subset of inconsistent scores for qualified batches and a subset of inconsistent scores for deviation batches. The inconsistent score is the absolute value of the difference between the predicted value of the quality indicator and the corresponding measured value of the historical output quality data. A traceable index is established using historical batch numbers, and the control system uses the target coverage rate... Take the first from each of the two subsets. The final quantile value of the quality indicator is determined by taking the larger of two ordinal statistics as candidate quantile values. ,in The sample size for the corresponding subset is automatically determined by the system based on the number of historical batches read. Then, using the quality prediction value of the current control cycle as the interval benchmark, a quality confidence interval for the quality indicator is generated, and the quality offset is corrected accordingly. The quality confidence interval is determined by the following formula: ; in, For quality indicators In the control cycle number is The corresponding quality confidence interval, For quality indicators In the control cycle number is The corresponding predicted value and from Read by index. For quality indicators The corresponding final quantile value is determined jointly by the inconsistency quantile values ​​of historical qualified batches and historical deviation batches. The quality index is indexed and the following values ​​are taken in sequence: whiteness, loss on ignition, specific surface area, particle size, and... In proportion, Number the control cycle; The control system will The lower bound is compared with the corresponding target interval lower bound, and the difference is included in the lower limit offset component of the index when the lower bound is less than the target lower bound. The upper bound is compared with the corresponding target interval upper bound, and if the upper bound is greater than the target upper bound, the difference is included in the upper limit offset component of the index. Finally, the lower limit offset component and the upper limit offset component are merged into the quality offset vector. This outputs the steps. Underburn risk index obtained by rolling solution Overheating risk index mass offset after confidence interval correction .

[0023] In this specific embodiment, S5 includes: The control system is numbered in the current control cycle. Read the underburn risk index at time Overheating risk index and the mass offset corrected for confidence intervals and read Prediction step size used With sampling period To construct a rolling solver window, the rolling solver uses a sequence of control quantities. As decision variables, the single-step control vector By partition number Align and sequentially include the zone temperature settings Fuel flow rate Hot air flow rate Furnace pressure regulation Feeding speed With material layer thickness correction amount ,and Acting on the zone heating controller, Acting on fuel valve actuator, Acting on the hot air valve actuator, Acting on the induced draft or damper actuator, Acting on the feeding and conveying actuator, Acting on the fabric or scraping mechanism; During the rolling optimization process, The graph time series prediction model is used as the process predictor and the injective method is employed. Injecting the control input channel of the predictor, thereby in each candidate The predicted sequences of future temperature fields, material thermal histories, and bed heat fluxes for each zone are obtained and combined with the predicted sequences of the same candidate Next step of calling The calculation logic yields the candidate underburn risk index. Candidate overheating risk index offset from candidate quality At the same time, a slack variable vector is introduced. The rolling optimization objective function, used to make relaxable constraints feasible during conflict resolution, is determined by the following formula: ; in, The control cycle number is The corresponding rolling optimization objective function value, For length is The control sequence decision variables, Let be a vector of relaxation variables, and let each component correspond one-to-one with the relaxed constraint. In the control sequence Under the action of steps The calculated mass offset vector, In the control sequence Under the action of steps The calculated underburning risk index In the control sequence Under the action of steps The calculated overheating risk index, The change in the control sequence and by The difference between the control quantity sequence executed in the previous control cycle and the corresponding component constitutes the result. This is the square of the L2 norm obtained by summing the squares of each element. The weight of the quality offset target term is taken as follows: In the software implementation, the actual measurement is based on the current cycle. renew, Weight the underburning risk target item and take Weight the target item for overheating risk and take To control the adjustment range of the target item weight and take Penalize the weights of slack variables and take Number the control cycle. To predict the step size; In this embodiment, the control constraints for the rolling solution are fixed and include zone temperature difference constraints, quality target constraints, allowable adjustment range and allowable rate of change constraints of the actuator, and equipment safety boundary constraints. Among them, the zone temperature difference constraints are combined according to adjacent zones. Apply and require that the predicted temperature difference between adjacent zones within a future prediction step does not exceed a preset zone temperature difference threshold. Quality target constraints require prediction Phase ratio located at Predicted burn reduction located at Predicted specific surface area is located at Prediction granularity is located at Predicted whiteness is located at The allowable adjustment range constraint of the actuator is fixed in this embodiment. The allowed rate of change constraint is fixed in this embodiment as Equipment safety boundary constraints are non-relaxable constraints and are fixed, including the maximum allowable temperature of the furnace. and the upper limit of the absolute value of furnace pressure And during model inference and constraint checking, any factors that cause the predicted temperature to exceed [the limit] will be considered. This could lead to the predicted furnace pressure exceeding [a certain value]. Candidates Determined as infeasible; To achieve the partition response lag compensation, the control system numbers each partition as follows: Calculate response lag time The The timing of the temperature setpoint change in this zone and the magnitude of the change when the measured temperature in this zone reaches the setpoint are compared from the most recent 50 historical control change events. The time difference at each moment is determined by taking the median and stored in seconds. The control system simultaneously follows the steps. Material travel speed The arrival time of materials from the feed end to the zone inlet is calculated using the furnace configuration parameter table. And based on this, the control lead time is determined. The control lead time is expressed in terms of the number of control cycles and is calculated according to... The software calculates the temperature settings for each zone, and then... Fuel flow rate With hot air flow Press the button at the effective time within the scrolling solver window. Perform forward scheduling and write the scheduling results into the control instruction queue so that the corresponding actuator can complete the control quantity adjustment before the material enters the zone; To achieve the generation of the conflict handling control quantity, when the control system detects... and Simultaneously, when the condition is met, conflict handling logic is triggered and the threshold is fixed. and The first order of conflict resolution will predict The phase ratio and the target interval constraint of the predicted burn-off are set as non-relaxable constraints and directly enter the current cycle's rolling solution constraint set. The second order of conflict handling is to set the partition number set corresponding to the grain stability region. The upper limit of the temperature setting is corrected to not exceed the upper boundary of the grain growth initiation temperature range. It also participates in the rolling solution as a variable boundary correction. When the first and second orders result in an empty feasible region due to the inability to relax the equipment safety boundary, the control system enables the relaxation variable vector. And set slack variables and penalty weights in the order of first priority over second priority, and so on. The slack variable penalty weights corresponding to the proportionality and reduction constraints are fixed as follows: And solidify the relaxation variable penalty weights corresponding to the upper bound constraint of the grain stability region temperature as At the same time, the preset allowable deviation amount will be fixed and the prediction will be made. The allowable deviation of the phase ratio is set to 0.003, the allowable deviation of the predicted loss on ignition is set to 0.05, the allowable deviation of the predicted specific surface area is set to 0.3, the allowable deviation of the predicted particle size is set to 0.05, the allowable deviation of the predicted whiteness is set to 0.5, the allowable deviation of the grain stability zone temperature is set to 20K, and the allowable deviation of the control quantity adjustment is solidified at 1.5 times the above allowable change rate, thereby generating feasible control quantities under the premise of meeting the equipment safety boundary; After satisfying the first and second orders or generating feasible control quantities, the control system is based on the mass offset. Subtracting the boundary differences eliminated by the first sequence constraints to form the remaining mass offset and performing the third sequence, only for the feed rate. With material layer thickness correction amount Open up additional degrees of freedom to reduce the remaining mass offset while keeping the temperature, fuel, hot air and furnace pressure control values ​​within the range of feasible solutions determined by the first two orders; In this embodiment, the rolling numerical solver employs a sequential quadratic programming algorithm and sets the maximum number of iterations to 10 and the objective function convergence threshold to [value missing]. Each iteration's quadratic programming subproblem is solved using the interior-point method, with a fixed maximum step size and backtracking count of 8. After completing the solution for the current cycle, the solution is simply deployed and executed. Control commands after delay compensation scheduling and will Write the control history record, and then proceed to the next control cycle numbered as follows. The system uses newly acquired data from different partitions and model predictions to repeatedly perform rolling solutions to achieve closed-loop control.

[0024] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0025] This invention maps the fluctuations in precursor moisture content, particle size distribution, and bed thickness to changes in furnace heat load, heat transfer resistance, and residence time by using a thermal-material coupling partition diagram and a partition perturbation weight matrix. This enables the graphical time-series prediction model to output partition temperature fields, material thermal histories, and bed heat flux for future control cycles, providing closed-loop inputs for subsequent risk calculations and rolling solutions.

[0026] This invention forms a continuous control chain in stages such as partition map construction, thermal history risk calculation, quality confidence interval correction, and response lag compensation, so that under-burning risk, over-burning risk, and quality offset can all be included in the rolling optimization objective function, thereby better achieving partitioned stable control of the continuous calcination process of high-purity alumina.

Claims

1. A method for partitioned calcination of high-purity alumina, characterized in that, include: S1. Divide the continuous calcining furnace into a preheating and dehydration zone, a phase transformation induction zone, an α-phase transformation zone, a grain stabilization zone, and a cooling and slow release zone along the material travel direction. Collect material state data and thermal state data for each zone, collect discharge quality data corresponding to the sampling period or discharge batch, and obtain historical control quantities corresponding to the sampling period or control period to obtain the zone collection data. S2. Construct a thermal material coupling partition diagram with partitions as nodes and heat transfer relationships between adjacent partitions as edges based on the partition data. Then, adjust the node weights and edge weights according to the particle size distribution deviation relative to the target particle size distribution, the moisture content difference relative to the target moisture content, and the material layer thickness deviation relative to the target material layer thickness to obtain the partition perturbation weight matrix. S3. Input the thermal coupling partition diagram, partition disturbance weight matrix and historical control quantity into the time series prediction model, and output the temperature field, material thermal history and material layer heat flux of each partition in the future control cycle. S4. Based on the temperature field of each zone, the thermal history of the material, the heat flux of the material layer, and the preset quality target range, generate the underburning risk index, the overburning risk index, and the quality deviation. S5. Based on the underburning risk index, overburning risk index, and quality deviation, perform rolling solutions for the zone temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure adjustment, feeding speed, and material layer thickness correction, and adjust the corresponding actuators according to the solution results.

2. The method for partitioned calcination of high-purity alumina according to claim 1, characterized in that, S1 includes: Obtain the current product specification number, current furnace type number, current control cycle, and zone number based on the current production task; In each zone, the precursor particle size distribution, precursor moisture content, material layer thickness, feed rate, zone temperature, fuel flow rate, hot air flow rate, furnace pressure, and exhaust oxygen content are collected. The discharge whiteness, loss on ignition, specific surface area, particle size, and α phase ratio obtained from the discharge end or discharge batch are used as the discharge quality data. Read the historical control quantity corresponding to the sampling period or control period from the control history record or the execution record of the previous control period, and associate the historical control quantity with the corresponding partition number and sampling time or control period; According to the corresponding relationship of sampling period or the corresponding relationship of discharge batch, the material state data, the thermal state data and the discharge quality data are time-aligned or batch-aligned, and the discharge quality data corresponding to the discharge batch is associated with the sampling time window of the corresponding material passing through each partition to obtain the partition collection data with partition number and sampling time.

3. The high-purity alumina partition calcination method according to claim 2, characterized in that, S2 include: The temperature, material layer thickness, material residence time, and heat load of each zone are used as node features, and the hot air flow direction, temperature difference, and material transfer time between adjacent zones are used as edge features. The material residence time, heat load, hot air flow direction, and material transfer time are determined based on the data collected from the zones and the furnace configuration parameters associated with the furnace type number and zone number. Read the target particle size distribution, target moisture content, and target bed thickness from the process formulation associated with the current product specification number, furnace type number, and zone number; The difference between the current particle size distribution and the target particle size distribution is determined as the particle size distribution deviation; the difference between the current moisture content and the target moisture content is determined as the moisture content difference; and the difference between the current material layer thickness and the target material layer thickness is determined as the material layer thickness deviation. The moisture content difference is mapped to the dehydration heat load increment, the particle size distribution deviation is mapped to the phase transformation hysteresis increment, and the material layer thickness deviation is mapped to the heat transfer resistance increment and residence time offset, and the node weight and edge weight are corrected accordingly.

4. The method for partitioned calcination of high-purity alumina according to claim 3, characterized in that, S3 includes: The node features, edge features, partition disturbance weight matrix, partition temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure adjustment, feed rate, and material layer thickness correction value of the thermal coupling partition diagram are used as model inputs. The time series prediction model is used to jointly encode the heat transfer relationship between adjacent zones and the time series relationship in the direction of material movement to obtain the temperature field of each zone, the thermal history of the material, and the heat flux of the material layer at multiple future sampling times.

5. The method for partitioned calcination of high-purity alumina according to claim 4, characterized in that, S4 includes: The material residence time is determined based on the feeding rate and material layer thickness in the data collected from the partitions, as well as the partition length associated with the furnace type number and partition number. The material thermal history is input into the α-phase conversion effective thermal history integration channel, and the α-phase conversion sufficiency index is calculated according to the temperature duration of the phase change induction zone and the α-phase conversion zone, the material residence time and the heat flux of the material bed. The material thermal history is input into the grain growth risk thermal history integration channel. The grain growth risk index is calculated based on the duration of temperature exceeding the upper boundary of the grain growth initiation temperature range within the grain stability region and the heat flux of the material layer. The grain growth initiation temperature range is a preset temperature range associated with the current product specification number and furnace type number. Based on the temperature field of each zone, the thermal history of the material, and the heat flux of the material layer, a predicted value for the discharge quality is generated. The predicted value for the discharge quality includes predicted whiteness, predicted loss on ignition, predicted specific surface area, predicted particle size, and predicted α phase ratio. The preset quality target range includes whiteness target range, ignition loss target range, specific surface area target range, particle size target range, and α phase target range; The underburn risk index is generated based on the α-phase transformation sufficiency index, the deviation of the predicted loss on ignition from the target range, and the deviation of the predicted α-phase ratio from the target range. The overburn risk index is generated based on the grain growth risk index, the deviation of the predicted specific surface area from the target range, the deviation of the predicted particle size from the target range, and the deviation of the predicted whiteness from the target range. The mass offset is generated based on the boundary difference between the predicted output quality and the preset quality target range.

6. The high-purity alumina partition calcination method according to claim 5, characterized in that, S5 include: The control constraints are: the predicted temperature difference between adjacent zones within the future control cycle does not exceed a preset zone temperature difference threshold; the predicted α phase ratio is within the α phase target range; the predicted loss on ignition is within the loss on ignition target range; the predicted specific surface area is within the specific surface area target range; the predicted particle size is within the particle size target range; and the predicted whiteness is within the whiteness target range. The preset zone temperature difference threshold is a preset threshold associated with the current furnace type number, adjacent zone combination, product specification number, and control cycle. The predicted α phase ratio, the predicted loss on ignition, the predicted specific surface area, the predicted particle size, and the predicted whiteness are determined by the predicted discharge quality value, which is generated based on the temperature field of each zone, the thermal history of the material, and the heat flux of the material layer. The zone temperature setpoint, fuel flow rate, hot air flow rate, furnace pressure regulation, feed rate, and material layer thickness correction are constrained within the allowable adjustment range and allowable rate of change associated with the current furnace type number, actuator number, or product specification number; The penalty weights of the rolling optimization objective function are determined based on the underburn risk index, the overburn risk index, and the quality offset. The rolling optimization objective function is generated based on the quality offset objective item, the underburn risk objective item, the overburn risk objective item, and the control quantity adjustment range objective item, and the penalty weights are used to weight the corresponding objective items. Within each control cycle, the rolling optimization objective function is solved to obtain the zone temperature setpoint, the fuel flow rate, the hot air flow rate, the furnace pressure adjustment, the feed rate, and the material layer thickness correction.

7. The method for partitioned calcination of high-purity alumina according to claim 5, characterized in that, The mass offset is further corrected based on the mass confidence interval, specifically including: From historical batches that match the current product specification number and furnace type number, historical qualified batches and historical deviation batches are classified according to whether whiteness, loss on ignition, specific surface area, particle size and α phase ratio are all within the corresponding quality target range. Using the absolute values ​​of whiteness prediction error, loss on ignition prediction error, specific surface area prediction error, particle size prediction error, and α phase proportion prediction error in the historical qualified batches and the historical deviation batches respectively, construct the qualified batch non-consistency fractional subset and the deviation batch non-consistency fractional subset for the corresponding quality indicators. The whiteness prediction error, loss on ignition prediction error, specific surface area prediction error, particle size prediction error, and α phase proportion prediction error are the differences between the predicted values ​​of the corresponding quality indicators of the historical batches and the corresponding measured values ​​in the historical output quality data. According to the preset target coverage rate associated with the current control cycle, the quantile values ​​of qualified batches and the quantile values ​​of deviating batches are determined from the subset of qualified batches with inconsistent scores and the subset of deviating batches with inconsistent scores, respectively. The last quantile value after sorting the qualified batch quantile values ​​and the deviating batch quantile values ​​in ascending order of value is used as the quantile value of the corresponding quality indicator. Using the predicted value of the quality index corresponding to the current control cycle as the interval benchmark, the confidence interval of the quality index is generated according to the quantile value of the corresponding quality index. Thus, confidence intervals for whiteness, ignition loss, specific surface area, particle size and α phase proportion are generated respectively. When the lower bound of any quality confidence interval is less than the lower bound of the corresponding target interval, the lower limit offset component of the corresponding quality index is generated based on the difference between the lower bound of the corresponding target interval and the lower bound of the quality confidence interval. When the upper bound of any quality confidence interval is greater than the upper bound of the corresponding target interval, the upper limit offset component of the corresponding quality index is generated based on the difference between the upper bound of the quality confidence interval and the upper bound of the corresponding target interval. The lower limit offset component and the upper limit offset component are incorporated into the mass offset.

8. The method for partitioned calcination of high-purity alumina according to claim 4, characterized in that, The graph time series prediction model is trained using historical calcination samples, which include historical partition data, historical partition perturbation weight matrix, historical control variables, historical discharge quality data, and future partition temperature field labels, material thermal history labels, and material layer heat flux labels corresponding to each historical control cycle. During training, the loss function is composed of the prediction error of the future zone temperature field, the prediction error of the material thermal history, the prediction error of the material layer heat flux, and the prediction error of the discharge quality relative to the historical discharge quality data. The predicted discharge quality is generated based on the temperature field of each zone, the material thermal history, and the material layer heat flux. The trained model parameter version is associated with and stored with the corresponding sampling period, furnace type number and product specification number; When step S3 is executed, the model parameter version that matches the current sampling period, current furnace type number, and current product specification number is called.

9. The method for partitioned calcination of high-purity alumina according to claim 6, characterized in that, The rolling solution also includes partition response lag compensation, specifically including: The response lag time of each zone is determined based on the time difference between the time of change of historical control variables and the time of temperature response in each zone. The current material residence time is determined based on the feeding rate and material layer thickness in the data collected from the partitions, as well as the partition length associated with the furnace type number and partition number. The control lead time is determined based on the response lag time and the current material residence time. The solution time windows for the zone temperature setpoint, the fuel flow rate, and the hot air flow rate, as well as the effective time of the control quantity of the corresponding actuator, are adjusted according to the control advance amount, so that the corresponding actuator completes the control quantity adjustment before the material enters the zone to be adjusted corresponding to the control advance amount.

10. The method for partitioned calcination of high-purity alumina according to claim 6, characterized in that, When the underburn risk index and the overburn risk index both exceed their respective preset risk thresholds, conflict handling control quantities are generated in the order of first, second and third order. The preset risk thresholds are thresholds associated with the current product specification number and control cycle. The first order is to constrain the predicted α phase ratio and the predicted scorch reduction within the corresponding target interval; The second sequence is to constrain the set value of the grain stabilization zone temperature to not exceed the upper boundary of the grain growth initiation temperature range. The grain growth initiation temperature range is a preset temperature range associated with the current product specification number and furnace type number, and is used as a variable boundary correction amount to participate in the generation of feasible control quantities. When the first order and the second order cannot be satisfied simultaneously, the equipment safety boundary associated with the current furnace type number is taken as an unrelaxable constraint, and preset relaxation variables and penalty weights are set according to the order of priority of the first order over the second order, so as to perform feasible relaxation on the corresponding constraints in the first order and the second order and generate feasible control quantities. The feasible control quantities are constrained by the equipment safety boundary and preset allowable deviations. The preset allowable deviations include the predicted α phase ratio allowable deviation, the predicted ignition allowable deviation, the predicted specific surface area allowable deviation, the predicted particle size allowable deviation, the predicted whiteness allowable deviation, the grain stability zone temperature allowable deviation, and the control quantity adjustment allowable deviation associated with the current product specification number, furnace type number, and control cycle. After satisfying the first order and the second order, or after generating the feasible control quantity, the remaining mass offset is formed based on the boundary differences that have not been eliminated in the mass offset. The third step is to generate a feed rate adjustment amount and a material layer thickness correction amount based on the remaining mass offset, and to constrain the feed rate adjustment amount and the material layer thickness correction amount within the feed rate adjustment range and material layer thickness correction range associated with the current furnace type number. The conflict handling control variable is used as the priority constraint and variable boundary correction variable of the rolling optimization objective function.

Images