Search

CN-121977365-A - Multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method and system

CN121977365ACN 121977365 ACN121977365 ACN 121977365ACN-121977365-A

Abstract

The invention relates to the technical field of flue gas treatment and waste heat utilization in a nonferrous metal smelting process, and discloses a multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method and a system, wherein the method comprises the steps of collecting original data in real time, and generating a multi-source flue gas time sequence data set through processing; the method comprises the steps of establishing a multi-kiln flue gas mixing mechanism model, merging a data driving model to carry out deviation correction to form a prediction model, establishing a comprehensive multi-objective optimization function, defining safety constraint conditions, calculating the opening of each baffle, the rotating speed of an induced draft fan and the optimal setting value of a shunt valve in real time through a prediction control algorithm of the prediction model, and realizing stable flue gas parameters of an acid making inlet and efficient dynamic allocation of waste heat recovery. The system comprises a data preprocessing module, a prediction model construction module and a constraint condition construction module. The invention realizes high efficiency, stability and low carbon emission of smelting flue gas, and improves the energy utilization efficiency and production benefit of the whole metallurgical process.

Inventors

  • WANG HUA
  • HU JIE
  • YANG SHILIANG

Assignees

  • 昆明理工大学

Dates

Publication Date
20260505
Application Date
20251231

Claims (10)

  1. 1. The multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method is characterized by comprising the following steps of: Based on a multi-source smoke time sequence data set, a multi-kiln smoke mixing mechanism model based on material conservation and energy conservation is established, and a data driving model is fused to carry out deviation correction, so that smoke flow, temperature and smoke flow are formed A concentration prediction model; based on the flow rate, temperature and the like of the flue gas The concentration prediction model is used for constructing a comprehensive multi-objective optimization function and defining safety constraint conditions through flue gas flow, temperature and temperature The concentration prediction model predicts the optimal set value of each baffle opening, the rotational speed of the induced draft fan and the shunt valve of the control algorithm calculation in real time, realize the stable and high-efficient dynamic allocation of acid making inlet flue gas parameters and waste heat recovery.
  2. 2. The multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method according to claim 1, wherein flue gas flow, temperature and temperature are formed A process for a concentration prediction model comprising the steps of: Taking a multi-source smoke time sequence data set as input, and inputting the real-time flow rate of each kiln outlet, The concentration, temperature and smoke component data are substituted into a node mixing calculation frame constructed according to the principle of material conservation and energy conservation according to the actual topological connection relation of a smoke pipe network, and the node mixing calculation frame calculates the total flow and weighted average of the mixed smoke at each confluent node in real time The concentration and the mixing temperature obtained based on the specific heat and flow weighting of the flue gas of each branch generate a set of theoretical mixing parameter sequences which cover all key nodes of a full pipe network and are based on a physical law; The calculated full-network theoretical mixed parameter sequence is compared with the actual measured parameter sequence of the corresponding node in the multi-source smoke time sequence data set point by point to generate a theoretical-actual deviation time sequence curve of the full node; the method comprises the steps of taking a theoretical-actual deviation curve and corresponding raw working condition data of fuel, oxygen concentration and load states of each kiln as training samples, and sending the training samples into a nonlinear relation mapping network for learning; The method comprises the steps of inputting working condition data collected in real time into a node hybrid computing frame and a data driving compensator in parallel, outputting a current theoretical prediction reference value by the node hybrid computing frame, outputting a prediction deviation compensation quantity based on real-time working condition identification by the data driving compensator, fusing the node hybrid computing frame and the data driving compensator on line by a dynamic weight adapting mechanism, wherein the weight deviates to the theoretical reference value when the working condition is stable and the physical rule is dominant, dynamically increases the weight of the compensation quantity when the working condition is severely fluctuated and a historical similar deviation mode appears, and finally generating the flue gas flow, the temperature and the flue gas flow which simultaneously conform to the physical rule and can adaptively correct the system error The concentration is fused with a prediction model.
  3. 3. The multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method according to claim 1, wherein the process of realizing stable flue gas parameters of an acid making inlet and efficient dynamic allocation of waste heat recovery is realized, comprises the following steps: the current collected furnace working condition data and the setting value of an actuating mechanism are taken as initial states and input into the formed flue gas flow, temperature and actuating mechanism The concentration fusion prediction model is used for superposing simplified description of dynamic response characteristics of the actuating mechanism, and the flue gas flow, temperature and temperature of the acid making inlet are controlled in a plurality of continuous control time periods in the future under different candidate regulation strategies Performing parallel rolling prediction on the evolution track of the concentration key parameters, and outputting a group of multi-parameter prediction track families covering the future time domain; receiving multi-parameter prediction track families, combining all defined safety constraint conditions, carrying out constraint compliance verification on each prediction track, and hard removing tracks violating hard constraint, carrying out punishment weighting on the rest feasible tracks according to the degree of approaching soft constraint boundaries; The method comprises the steps of taking an output feasible candidate solution set as an initial searching starting point, substituting future multi-period prediction parameter values corresponding to each candidate solution into a constructed calculation formula of the comprehensive performance index, adopting a directional searching strategy in a converged feasible domain, iteratively evaluating new candidate set value combinations along the direction capable of enabling the comprehensive performance index to be lowered until an optimal solution enabling the value of the comprehensive performance index to be minimum or meeting a cut-off condition is found, and outputting each actuator set value corresponding to the next control period in the optimal solution.
  4. 4. The multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method according to claim 3, wherein the comprehensive performance index is as follows : Wherein, the The weight coefficient of each optimization target is optimized Balancing the requirements of each target; As actual flue gas The concentration of the water in the water is higher, In order to achieve the target concentration of the substance, For the actual flow rate of the flue gas, Is the target flow; For the actual flue gas temperature, Is the target temperature; Is the first The adjustment amount of the actuator.
  5. 5. A multi-kiln flue gas multi-objective collaborative intelligent dynamic allocation method according to claim 3, wherein a plurality of constraints are defined in the optimization process, and the main constraints include: Negative pressure constraint of each kiln flue and collecting flue : Throughput constraints for acid making systems Constraint on capacity of waste heat boiler : Concentration operation window constraints And Concentration operation window constraints : Physical limits and speed constraints for each actuator : 。
  6. 6. The multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method according to claim 3, wherein the acid making inlet is characterized by that Concentration, objective function is: Wherein, the As actual flue gas The concentration of the water in the water is higher, Is the target concentration; The flow of the flue gas at the acid making inlet has the objective function of: Wherein, the For the actual flow rate of the flue gas, Is the target flow; The inlet temperature of the waste heat boiler is as follows: Wherein, the For the actual flue gas temperature, Is the target temperature; The objective function is: Wherein, the Is the first The adjustment amount of the actuator.
  7. 7. A multi-kiln flue gas multi-target collaborative intelligent dynamic allocation method according to claim 3, wherein the process of outputting a set of multi-parameter prediction trajectory families covering the future time domain comprises the steps of: The method comprises the steps of taking currently collected furnace working condition data, current set values of an executing mechanism and real-time target values of key parameters of an acid making inlet as inputs, obtaining deviation vectors between the real-time values and the target values of the key parameters, taking the direction and the amplitude of the deviation vectors as guidance, combining current working points and limits of the executing mechanism, and generating a future multi-period executing mechanism action preparation sequence; The method comprises the steps of combining candidate regulation strategies with a flue gas fusion prediction model and an actuating mechanism dynamic response model to carry out future multi-period prediction, wherein a hierarchical decoupling architecture is adopted in the calculation process, namely, for each candidate regulation strategy, a calculation thread is independently distributed, and basic state prediction of an initial period is executed in parallel; Collecting single strategy prediction tracks output by all calculation threads in parallel, performing time alignment and format unification processing, checking the integrity of each single strategy prediction track, eliminating tracks with incomplete data caused by calculation abnormality, then aggregating all effective single strategy prediction tracks according to strategy numbers to form a structured data set, adding descriptive metadata into the data set, packaging and outputting the data set as a final multi-parameter prediction track family.
  8. 8. The multi-kiln flue gas multi-objective collaborative intelligent dynamic allocation method according to claim 7, wherein the process of generating a future multi-period actuator action preparation sequence comprises the steps of: Matching the deviation vector with the influence coefficient table to generate a preliminary adjustment direction vector for indicating the adjustment direction of each actuator for collaborative correction of the deviation, and obtaining the relative priority ordering of the adjustment quantity required in each direction; For each actuator, according to the adjusting direction, based on the current setting value, taking the maximum single step change limit value as a boundary to the target direction to generate a plurality of equally spaced discrete heuristic setting values, including the option of keeping unchanged; The method comprises the steps of taking a single-step feasible action set as an action candidate of a first period, taking a current state as a root node, sequentially applying each first period action candidate to a prediction model from the root node to obtain a prediction state of the end of the first period, generating the single-step feasible action set in the prediction state as an action candidate of a second period, expanding the single-step feasible action set from time to form a multi-layer decision tree, retaining a plurality of branch paths with optimal comprehensive deviation reduction effect when each layer is expanded, cutting out the rest branches, and traversing to a preset period depth, wherein all the retained complete paths from the root node to the leaf node are finally generated multi-period actuator action preparation sequences.
  9. 9. The multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method according to claim 1, further comprising the steps of collecting flow, temperature and flow of each kiln flue gas in real time, Concentration of, Raw data of concentration, pressure and dust content are processed by removing abnormal values, repairing missing data, time alignment and standardization to generate a multi-source smoke time sequence data set.
  10. 10. A multi-kiln flue gas multi-target cooperative intelligent dynamic allocation system applied to the multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method according to any one of claims 1 to 9, characterized in that the multi-kiln flue gas multi-target cooperative intelligent dynamic allocation system comprises: The data preprocessing module is used for collecting the flow rate and the temperature of the flue gas of each kiln in real time, Concentration of, Raw data of concentration, pressure and dust content are processed by removing abnormal values, repairing missing data, time alignment and standardization to generate a multi-source smoke time sequence data set; The prediction model construction module is used for establishing a multi-kiln flue gas mixing mechanism model based on material conservation and energy conservation based on the multi-source flue gas time sequence data set, and fusing the data driving model to carry out deviation correction so as to form flue gas flow, temperature and energy conservation A concentration prediction model; Constraint condition construction module for constructing constraint conditions based on flue gas flow, temperature and The concentration prediction model is used for constructing a comprehensive multi-objective optimization function and defining safety constraint conditions through flue gas flow, temperature and temperature The concentration prediction model predicts the optimal set value of each baffle opening, the rotational speed of the induced draft fan and the shunt valve of the control algorithm calculation in real time, realize the stable and high-efficient dynamic allocation of acid making inlet flue gas parameters and waste heat recovery.

Description

Multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method and system Technical Field The invention relates to the technical field of flue gas treatment and waste heat utilization in the nonferrous metal smelting process, in particular to a multi-kiln flue gas multi-target cooperative intelligent dynamic allocation method and system, and relates to flue gas flow, temperature and sintering machine for a plurality of smelting kilns (such as a flash furnace, a side-blowing furnace, an oxygen bottom-blowing furnace, a roasting furnace, a sintering machine and the like)An intelligent dynamic allocation method and system for performing multi-objective collaborative optimization on key parameters such as concentration and the like. Specifically, a multi-objective collaborative optimization method based on intelligent dynamic allocation is provided, and the flue gas flow, the temperature and the flue gas flow can be realized on the premise of ensuring the efficient and stable operation of the smelting processAccurate regulation and control of concentration. Background In the smelting process of nonferrous metals such as copper, lead, zinc and the like, a production mode of parallel operation of a plurality of furnaces is adopted, for example, a 'flash furnace + side blowing furnace + converter' combined smelting system and a plurality of roasting furnaces or sintering machines are adopted to form a parallel acid supply system. The mode not only improves the production efficiency and the resource utilization rate, but also brings technical challenges, especially the flow rate, the temperature and the utilization rate of the flue gasFluctuation of parameters such as concentration. Because of the differences of the kiln in aspects of mineral mixture ratio, fuel use, oxygen concentration, stability of the furnace condition and the like, the smoke of each branch presents obvious fluctuation on a plurality of parameters, and the operation stability of a downstream system is directly affected, so that the complex technical problem is caused. Firstly, the air inlet condition of the acid making system can be greatly changed due to the fluctuation of the flue gas of a plurality of kilns, in particularConcentration instability.Fluctuations in concentration not only affect the reaction efficiency of the catalyst, but may also lead to reduced acid production in severe cases, even to system downtime. Secondly, the fluctuation of the flue gas flow can make the design processing capacity of the acid making system not be effectively utilized, and especially when the flow is low, the catalyst and the processing equipment of the system cannot exert due efficiency, and the frequent load adjustment further aggravates the thermal stress and the fault risk of the equipment. In addition, the existing allocation mode is highly dependent on manual experience, and an operator manually adjusts the opening degree of a baffle, the frequency of a draught fan and the like of each branch according to the kiln load and the air inlet parameters of an acid plant, but the adjustment mode lacks a unified material and energy conservation model, does not have systematic optimization strategies, can only consider a certain single index, and cannot realize flow, temperature and temperature under multiple constraintsOptimization and coordination of concentration. Therefore, the existing allocation mode is low in efficiency, and global collaborative scheduling cannot be achieved on the basis of multi-objective optimization. The existing control strategy is often limited to the adjustment of a single kiln or a single collection flue, and cannot be used for controlling the flow-temperature of multi-source flue gasAnd (3) performing global optimization under multiple targets and constraint conditions such as concentration-safe negative pressure. Lack of flue gas flow, temperature and temperature for multiple kilnsIntelligent dynamic allocation of parameters such as concentration and the like causes insufficient close coordination among links and low resource utilization efficiency. Therefore, an intelligent dynamic allocation technology capable of achieving multi-objective collaborative optimization is urgently needed to improve the energy utilization efficiency and environmental benefit of the smelting process. In the first prior art, application number 202510692616.8 discloses a desulfurization device for improving the flue gas purification efficiency and an operation method thereof, a dual-system flue gas dynamic allocation mechanism is constructed by utilizing a flue gas allocation flue, a flue gas blocking valve and a gas exchange device, when any system has reduced desulfurization efficiency and excessive pollutant emission due to maintenance and shutdown of a slurry circulating pump, poor slurry quality, excessive sulfur content at an inlet and the like, a control system can automatically open the corresponding flue gas blocking v