Method and system for optimizing heat transfer structure of stepped air duct of rotary kiln for extracting lithium from lithium ore

Abstract

The invention discloses a lithium ore lithium extraction rotary kiln stepped air duct heat transfer structure optimization method and system, and relates to the technical field of intelligent control. Multi-source sensing data in a rotary kiln stepped air duct are collected in real time, geometric parameters of the stepped air duct are combined, and coupling characteristics are extracted from the multi-source sensing data by using a principal component analysis method; constructing a dynamic heat transfer resistance coefficient matrix on the basis of the coupling characteristics and the multi-source sensing data, inputting the coupling characteristics and the dynamic heat transfer resistance coefficient matrix into a pre-trained gradient lifting regression tree model, outputting a real-time effective heat transfer area proportion of each air duct layer, and calculating the real-time effective heat transfer area proportion of each air duct layer on the basis of the real-time effective heat transfer area proportion. Step air duct parameters are optimized by adopting a self-adaptive particle swarm algorithm, and the optimized step air duct parameters are used for adjusting the step air duct parameters in the rotary kiln in real time through a PID (Proportion Integration Differentiation) controller to form closed-loop control; and the heat transfer efficiency in the calcining process can be greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, in particular to a lithium ore lithium extraction rotary kiln ladder air duct heat transfer structure optimization method and system. BACKGROUND

[0002] The rotary kiln is a key equipment in the process of lithium ore lithium extraction, which uses high temperature to heat the ore and make it react chemically to extract lithium components. However, due to the unevenness of airflow distribution and heat transfer in the rotary kiln, it often leads to over-high temperature in some areas and over-low temperature in some areas, which cannot achieve uniform heat transfer, thereby affecting the calcination efficiency of lithium ore.

[0003] Therefore, the technology of ladder air duct can be generally introduced to assist in airflow guiding and heat transfer, and its geometric structure directly affects the airflow and heat transfer efficiency. However, the traditional ladder air duct of rotary kiln is usually fixed in design, lacking flexibility, which leads to the inability to adjust in time according to the airflow changes in actual operation process; that is, the ladder air duct of the current rotary kiln usually adopts fixed structure design and does not dynamically adjust according to real-time working conditions.

[0004] Therefore, the present application proposes a lithium ore lithium extraction rotary kiln ladder air duct heat transfer structure optimization method and system. SUMMARY

[0005] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a lithium ore lithium extraction rotary kiln ladder air duct heat transfer structure optimization method and system, which can greatly improve the heat transfer efficiency in the calcination process.

[0006] To achieve the above-mentioned purpose, a lithium ore lithium extraction rotary kiln ladder air duct heat transfer structure optimization method is proposed, which comprises the following steps:

[0007] Step one: real-time acquisition of multi-source sensing data in the ladder air duct of the rotary kiln;

[0008] Step two: combined with the geometric parameters of the ladder air duct, using principal component analysis to extract coupling features from the multi-source sensing data, and based on the coupling features and the multi-source sensing data, a dynamic heat transfer resistance coefficient matrix is constructed;

[0009] Step three: input the coupling features and the dynamic heat transfer resistance coefficient matrix into the pre-trained gradient boosting regression tree model, and output the real-time effective heat transfer area proportion of each air duct level;

[0010] Step four: based on the real-time effective heat transfer area proportion, the adaptive particle swarm optimization algorithm is used to optimize the ladder air duct parameters;

[0011] Step five: the optimized ladder air duct parameters are adjusted in real time through the PID controller to adjust the ladder air duct parameters in the rotary kiln, forming a closed loop control.

[0012] The real-time collection of multi-source sensing data in the ladder duct of the rotary kiln includes the following steps:

[0013] Step 11: Install physical sensors for collecting each process parameter and material filling rate respectively at each ladder layer;

[0014] Step 12: All physical sensors access the database background through the industrial Internet of Things communication network, collect corresponding process parameters or material filling rates according to a unified preset sampling frequency, and then send the process parameters or material filling rates to the database background;

[0015] Step 13: In the process of collecting process parameters and material filling rates in the database background, a unified timestamp is added to the data of each physical sensor; each timestamp of process parameters, material filling rates and geometric parameters is uniformly aggregated into a multi-dimensional data matrix, and all multi-dimensional data matrices are stored in a structured manner as multi-source sensing data in time sequence.

[0016] The way of uniformly aggregating into a multi-dimensional data matrix and storing all multi-dimensional data matrices in a structured manner as multi-source sensing data is:

[0017] Process parameters and material filling rates from each physical sensor are arranged in a multi-dimensional data matrix in time sequence, each row of the multi-dimensional data matrix represents a physical sensor, each column represents a wind channel level of the physical sensor in the ladder duct, and the following marker is used: represents the measurement value of the i-th physical sensor at the j-th wind channel level at time t;

[0018] The multi-dimensional data matrix under each timestamp is saved in the database background in time sequence.

[0019] The method for extracting coupled features from multi-source sensing data by combining the geometric parameters of the ladder duct using principal component analysis includes the following steps:

[0020] Step 211: Read the geometric parameters of the ladder duct from the calcination control background in advance, and construct all geometric parameters into a geometric parameter vector;

[0021] Step 212: The combination of each wind channel level and each physical sensor is represented as a candidate parameter, and for each candidate parameter and geometric parameter vector, an extended feature vector is constructed for each moment; all geometric parameters are combined into a key parameter set;

[0022] Step 213: All candidate parameters are combined into an extended feature matrix at each moment;

[0023] Step 214: standardize each row of the extended feature matrix Y(t) to obtain a standardized extended feature matrix, denoted as ;

[0024] Step 215: collect a plurality of time instants of the standardized extended feature matrix from the historical operation period of the stepped air duct to form a historical three-dimensional matrix; wherein N is the number of collected time instants;

[0025] Step 216: based on the historical three-dimensional matrix, calculate the global mean and covariance matrix;

[0026] Step 217: perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues to , and corresponding eigenvectors to , all of which form a principal component matrix ;

[0027] Step 218: for the standardized extended feature matrix at time t, use the fixed global model to project the data, and the projection result of each candidate parameter reflects its air duct level in the global principal component space;

[0028] Step 219: in the projection result of each candidate parameter, analyze the load of the first component corresponding to the extended vector in each principal component, and preset a load threshold value. If the absolute values of the corresponding loads in a preset number of principal components are greater than the load threshold value, the candidate parameter is supplemented to the key parameter set. The finally generated key parameter set constitutes the coupling feature.

[0029] The construction of the dynamic heat transfer resistance coefficient matrix based on the coupling feature and the multi-source sensing data includes the following steps:

[0030] Step 221: set the dynamic heat transfer resistance coefficient matrix as R(t);

[0031] Step 222: for the jth air duct level, read the type of the physical sensor contained in the jth air duct level from the key parameter set of the coupling feature, and then combine the material filling rate of the jth air duct level and the geometric parameters of the stepped air duct to form a key parameter;

[0032] Step 223: describe the motion characteristics of the airflow in the stepped air duct through the fluid dynamics equation, and then combine the heat transfer model to describe the heat transfer according to the temperature distribution, to obtain the heat transfer resistance coefficient of any two air duct levels;

[0033] Step 224: Collect the actual heat transfer resistance coefficients between each air duct level during the historical operation period of the ladder air duct, and use regression analysis or neural networks to fit the parameters in the calculation equation of the heat transfer resistance coefficient.

[0034] The input of the pre-trained gradient boosting regression tree model includes the coupling feature and the dynamic heat transfer resistance coefficient matrix, and the output includes the real-time effective heat transfer area proportion of each air duct level.

[0035] Step 31: Collect the coupling feature vector and the heat transfer resistance coefficient matrix at each time from the historical operation period of the ladder air duct to form a training input sample, and collect the real-time effective heat transfer area proportion of each air duct level to form a training label.

[0036] Step 32: Construct a gradient boosting regression tree model containing multiple decision trees.

[0037] The structure of the gradient boosting regression tree model is composed of multiple decision trees.

[0038] The structure of each decision tree is composed of multiple branch nodes and leaf nodes. Each branch node is split according to the dimension of the input feature, and the goal is to make the mean value of each subset after splitting as similar as possible in the output label, i.e. the effective heat transfer area proportion. Each leaf node contains the final regression value, i.e. the predicted value of the model for the input feature.

[0039] Each decision tree reduces the prediction error by fitting the residual of the previous decision tree. The output of each decision tree is weighted and summed with the output of the previous decision tree to obtain the final prediction result.

[0040] The input of the gradient boosting regression tree model is the combination of the coupling feature vector and the dynamic heat transfer resistance coefficient matrix, and the output of the gradient boosting regression tree model is a continuous value representing the real-time effective heat transfer area proportion of each air duct level.

[0041] Step 33: Train the gradient boosting regression tree model using the training input sample and the training label.

[0042] The training method of the gradient boosting regression tree model is as follows:

[0043] Initialize the gradient boosting regression tree model, i.e. initialize the gradient boosting regression tree model to a constant value, and select the average value of the effective heat transfer area proportion in the training data as the initial value.

[0044] For each training input sample, calculate the current residual of the gradient boosting regression tree model, which is the difference between the actual value and the predicted value.

[0045] Each round of training fits the current residual by building a new decision tree, which optimizes the residual so that the output value of the gradient boosting regression tree model is close to the actual value of the training label;

[0046] The training of each decision tree depends on the error of the last round of training. The decision tree finds the best split point by dividing the training input samples, so that the output error on each branch is minimized;

[0047] The output of the final gradient boosting regression tree model is the weighted sum of all decision trees.

[0048] Step 34: When the gradient boosting regression tree model training is completed, real-time target input samples are formed by collecting real-time coupling features and dynamic heat transfer resistance coefficient matrix, and the target input samples are input into the gradient boosting regression tree model to obtain the real-time effective heat transfer area ratio of each air duct level output by the gradient boosting regression tree model.

[0049] The adaptive particle swarm optimization algorithm is used to optimize the stepped air duct parameters based on the real-time effective heat transfer area ratio, including the following steps:

[0050] Step 41: Construct a target optimization function according to the real-time effective heat transfer area ratio, and select the optimized stepped air duct parameters;

[0051] The optimized stepped air duct parameters include gas injection valve opening G and guide vane angle θ;

[0052] Step 42: Design a fitness function based on the target optimization function;

[0053] Step 43: Initialize the particle swarm of the adaptive particle swarm optimization algorithm;

[0054] The initialization method of the particle swarm of the adaptive particle swarm optimization algorithm is:

[0055] Generate a set of random particles, each particle representing a set of stepped air duct parameters;

[0056] Pre-set a motion range for each parameter of each particle, so that the particle is limited within the motion range during each generation update;

[0057] Use the particle update formula to update the position of each particle.

[0058] Step 44: Apply the adaptive particle swarm optimization algorithm to optimize the position of the particle swarm, gradually approach the optimal solution, and obtain the optimized stepped air duct parameters.

[0059] The way to adjust the stepped air duct parameters in the rotary kiln in real time by the PID controller based on the optimized stepped air duct parameters to form a closed-loop control is:

[0060] The optimized step air duct parameters are input into the PID controller, and the optimized step air duct parameters are used as adjustment target values. The PID controller adjusts the step air duct parameters in the rotary kiln to the adjustment target values ​​according to the adjustment target values.

[0061] A lithium ore extraction rotary kiln step duct heat transfer structure optimization system is proposed, which includes a multi-source data collection module, a feature extraction module, a heat transfer area ratio output module, and an air duct parameter optimization module; wherein each module is electrically connected;

[0062] The multi-source data collection module collects multi-source sensor data in the stepped air duct of the rotary kiln in real time and sends the multi-source sensor data to the feature extraction module;

[0063] The feature extraction module uses principal component analysis to extract coupling features from multi-source sensor data based on the geometric parameters of the stepped air duct. It then constructs a dynamic heat transfer resistance coefficient matrix based on the coupling features and multi-source sensor data. The coupling features and the dynamic heat transfer resistance coefficient matrix are then sent to the heat transfer area ratio output module.

[0064] The heat transfer area ratio output module inputs the coupling characteristics and dynamic heat transfer resistance coefficient matrix into the pre-trained gradient boosting regression tree model, outputs the real-time effective heat transfer area ratio of each duct level, and sends the real-time effective heat transfer area ratio to the duct parameter optimization module;

[0065] The air duct parameter optimization module uses an adaptive particle swarm algorithm to optimize the step air duct parameters based on the real-time effective heat transfer area ratio. The optimized step air duct parameters are adjusted in real time in the rotary kiln through a PID controller to form a closed-loop control.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] The application extracts coupling features from the multi-source sensing data by real-time acquisition of the multi-source sensing data in the rotary kiln ladder air duct, combined with the geometric parameters of the ladder air duct, using principal component analysis method, and based on the coupling features and the multi-source sensing data, a dynamic heat transfer resistance coefficient matrix is constructed, the coupling features and the dynamic heat transfer resistance coefficient matrix are input into the pre-trained gradient boosting regression tree model, and the real-time effective heat transfer area proportion of each air duct level is output, based on the real-time effective heat transfer area proportion, the adaptive particle swarm optimization algorithm is used to optimize the ladder air duct parameters, the optimized ladder air duct parameters are used to adjust the ladder air duct parameters in the rotary kiln in real time through the PID controller, forming a closed loop control; the coupling features are extracted from the multi-source sensing data by the principal component analysis method, combined with the dynamic heat transfer resistance coefficient matrix, a feature model related to the air duct parameters is constructed, the adaptive particle swarm optimization algorithm is used, based on the real-time feedback data, the ladder air duct parameters in the rotary kiln are automatically adjusted to optimize the heat transfer efficiency, so that the airflow distribution and heat transfer are more uniform, and the heat transfer efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 It is the internal heat transfer type rotary kiln and the kiln structure diagram;

[0069] Figure 2 It is the flow chart of the heat transfer structure optimization method of the lithium ore lithium extraction rotary kiln ladder air duct in embodiment 1 of the application.

[0070] Figure 3 It is the module connection relationship diagram of the heat transfer structure optimization system of the lithium ore lithium extraction rotary kiln ladder air duct in embodiment 2 of the application. DETAILED DESCRIPTION

[0071] The technical solutions of the application will be described below in conjunction with the embodiments, obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0072] In the traditional process of lithium extraction rotary kiln, the external heating mode is indirect heat transfer through baking the outer cylinder of the kiln body. The heat needs to penetrate through the thick steel plate, usually more than 50 mm, to be transferred to the internal materials, resulting in long heat transfer path, large thermal resistance, high energy consumption, and heavy equipment with high manufacturing and maintenance costs. Therefore, the technology of designing multi-layer stepped flow guide duct on the inner wall of the kiln body is introduced, which makes the high-temperature flue gas directly flow through the 8-12 mm thick thin-walled duct and exchanges heat with the materials, changing the traditional "outer wall heat conduction" mode to "embedded radiation-convection combined heat transfer". This structure increases the effective heat transfer area by 50%-100%, and the turbulence disturbance formed by the step difference enhances the uniformity of gas-solid heat exchange, which increases the heat exchange efficiency by more than 50%, reduces the thickness of the steel plate to realize the lightweight of the equipment, and saves energy by more than 40%. For example Figure 1 The inner heat transfer type rotary kiln and the kiln structure diagram are shown; the diagram shows the cylinder of the stepped duct of the inner heat transfer type rotary kiln, and the cross-sectional example diagram of the inlet and outlet of the stepped duct.

[0073] However, the geometric parameters such as the inclination, spacing, and wall thickness of the stepped duct are closely coupled with process parameters such as airflow pressure, flue gas flow rate, local temperature, oxygen concentration, carbon dioxide concentration, and material filling rate. If the parameters are not matched properly, it may cause uneven airflow distribution, local thermal stress concentration, and even duct structure deformation. For example, too large step spacing will weaken the turbulence effect, and too small inclination will easily cause ash accumulation and blockage; in addition, the thin-walled structure requires real-time dynamic adjustment of flue gas flow to balance the heat transfer efficiency and structural reliability due to its sensitivity to temperature fluctuations. Therefore, through multi-parameter collaborative optimization, combined with heat transfer performance prediction and structural mechanics analysis, the energy utilization efficiency is maximized under the premise of ensuring the long-term stable operation of the equipment, which is the core challenge of industrial application of the stepped duct technology.

[0074] Embodiment 1

[0075] As shown in Figure 2 The lithium extraction rotary kiln stepped duct heat transfer structure optimization method includes the following steps:

[0076] Step one: real-time collection of multi-source sensing data in the stepped duct of the rotary kiln;

[0077] Step two: using principal component analysis to extract coupled features from the multi-source sensing data combined with the geometric parameters of the stepped duct, and based on the coupled features and the multi-source sensing data, a dynamic heat transfer resistance coefficient matrix is constructed;

[0078] Step three: input the coupled features and the dynamic heat transfer resistance coefficient matrix into the pre-trained gradient boosting regression tree model, and output the real-time effective heat transfer area ratio of each duct level;

[0079] Step four: based on the real-time effective heat transfer area ratio, the adaptive particle swarm optimization algorithm is used to optimize the parameters of the stepped air duct;

[0080] Step five: the optimized parameters of the stepped air duct are adjusted in real time through the PID controller to adjust the parameters of the stepped air duct in the rotary kiln, forming a closed loop control.

[0081] In the embodiments of the present application, the real-time collection of multi-source sensing data in the stepped air duct of the rotary kiln includes the following steps:

[0082] Step 11: install physical sensors for collecting each process parameter and material filling rate in each stepped layer respectively;

[0083] Specifically, the material filling rate refers to the ratio between the volume occupied by the material such as ore particles in the rotary kiln and the available space volume in the kiln;

[0084] During the operation of the rotary kiln, the material continuously enters from the feed inlet and moves through the kiln, and part of the material has been converted or extracted into lithium and other components by the time it reaches the outlet. The remaining ore material is then discharged. Due to the changes in feed quantity and output, the material filling rate will change, so the material filling rate is real-time variable;

[0085] The sensor for collecting the material filling rate can be collected in real time by a magnetic sensor. The principle of the magnetic sensor for collecting the material filling rate is that the material in the rotary kiln generally contains magnetic components, and the magnetic sensor can detect the distribution of the magnetic components in the material. When the material passes through the magnetic sensor, the signal strength of the sensor changes. By analyzing the change of the signal, the accumulation and filling degree of the material can be understood.

[0086] The physical sensors and their uses include:

[0087] The temperature sensor is used to measure the local temperature, the pressure sensor is used to monitor the gas flow pressure, the gas flow speed sensor is used to detect the gas flow speed, and the gas concentration sensor is used to detect the oxygen, carbon dioxide and other gas concentrations in some key areas to reflect the combustion state;

[0088] All physical sensors are selected from industrial-grade high-temperature anti-interference devices, which have high sampling frequency such as once per second or faster to meet the real-time monitoring requirements. The air duct level and number of each physical sensor are recorded in the system to form a sensor set.

[0089] Step 12: all physical sensors access the database background through the industrial Internet of Things communication network, collect corresponding process parameters or material filling rates according to the unified preset sampling frequency, and then send the process parameters or material filling rates to the database background;

[0090] In a preferred embodiment, for real-time process parameters or material filling rate, preliminary filtering and denoising preprocessing can also be performed by embedded multi-dimensional filtering algorithms such as Kalman filtering or low-pass filtering, and adaptive parameter adjustment according to predefined process stages.

[0091] Step 13: During the process of collecting process parameters and material filling rate in the database background, a unified timestamp is attached to each physical sensor data; for each timestamp of process parameters, material filling rate and geometric parameters, a unified aggregation into a multi-dimensional data matrix is performed, and all multi-dimensional data matrices are stored in a time sequence as multi-source sensor data.

[0092] Specifically, the unified aggregation into a multi-dimensional data matrix and the storage of all multi-dimensional data matrices in a time sequence as multi-source sensor data is as follows:

[0093] The process parameters and material filling rate from each physical sensor are arranged in a time sequence to form a multi-dimensional data matrix, each row of which represents a physical sensor, each column of which represents the air duct level of the physical sensor in the ladder air duct, and the following notation is used: represents the measurement value of the i-th physical sensor at the j-th air duct level at time t;

[0094] The multi-dimensional data matrix at each timestamp is saved in the database background in a time sequence.

[0095] In a further preferred embodiment, in order to ensure data accuracy and long-term stability, it is also necessary to calibrate the process parameters and material filling rate regularly, which can be achieved by using a correction model such as linear regression or neural network correction algorithm constructed from historical data and real-time data to correct sensor drift and sudden change data.

[0096] Further, the extraction of coupled features from multi-source sensor data by combining the geometric parameters of the ladder air duct using principal component analysis includes the following steps:

[0097] Step 211: The geometric parameters of the ladder air duct are read in advance from the roasting control background, and all geometric parameters are constructed into a geometric parameter vector;

[0098] The geometric parameters include the inclination angle of the flow direction and the heat transfer efficiency, the distance between the ladder layers in the air duct, the thickness of the air duct wall affecting the heat conduction effect and the thermal stability of the equipment;

[0099] Step 212: The combination of each air duct level and each physical sensor is represented as a candidate parameter, and for each candidate parameter and geometric parameter vector, an extended feature vector is constructed for each moment; all geometric parameters are combined into a key parameter set;

[0100] Specifically, the way of constructing the extended feature vector is:

[0101] The geometric parameter vector is marked as ;

[0102] The extended feature vector corresponding to the candidate parameter combination of the i th physical sensor in the j th air duct level group is ; ; so that in subsequent analysis, not only the fluctuations of the candidate parameters themselves are considered, but also the geometric conditions are introduced as context information;

[0103] Step 213: The extended feature vectors of all candidate parameters at each time are combined into an extended feature matrix;

[0104] The form of the extended feature matrix Y(t) is:

[0105] , where is the number of all geometric parameters, M is the total number of all candidate parameters, and K is used to represent ;

[0106] Step 214: Each row in the extended feature matrix Y(t) is normalized to obtain a normalized extended feature matrix, marked as ;

[0107] Step 215: From the historical operation period of the stepped air duct, a plurality of time points of the normalized extended feature matrix are collected to form a historical three-dimensional matrix; where N is the number of collected time points;

[0108] Step 216: Based on the historical three-dimensional matrix, the global mean and covariance matrix are calculated;

[0109] Specifically, the calculation formula of the global mean is:

[0110] ;

[0111] The calculation formula of the covariance matrix C is:

[0112] ;

[0113] Step 217: The covariance matrix C is subjected to eigenvalue decomposition to obtain eigenvalues to , and corresponding eigenvectors to , all of which constitute a principal component matrix ;

[0114] Step 218: standardizing the extended feature matrix of time t , using a fixed global model to project the data, and the projection result of each candidate parameter reflects its wind tunnel level in the global principal component space;

[0115] Specifically, the global model is:

[0116] ;

[0117] Wherein, is By multiplying with a full 1 row vector, and then copying the generated mean vector M times to obtain a matrix with the same size as Align each candidate parameter vector with the global mean to eliminate the effect of the mean; thus Z(t) is the first to center each candidate parameter's extended feature vector, and then use the fixed principal component matrix Project these centralized vectors into the principal component space, and T represents transposition; and each column of the obtained Z(t) is the low-dimensional representation of the candidate parameter in the principal component space, so as to capture the main coupling features of the candidate parameter under fixed geometric conditions, and ensure uniform and stable semantics at different time points;

[0118] Step 219: in the projection result of each candidate parameter, analyze the load of the first component of the corresponding extended vector in each principal component, and preset a load threshold value, if the absolute value of the corresponding load in the preset number of principal components is greater than the load threshold value, the candidate parameter is supplemented to the key parameter set; The final generated key parameter set constitutes the coupling feature;

[0119] The analysis of the load of the first component of the corresponding extended vector in each principal component refers to the absolute value of each component in each feature vector, and the candidate parameter with larger load indicates that its candidate parameter value contributes significantly to the overall variation in the principal component, so that the candidate parameter is considered to have higher importance under the current geometric condition;

[0120] For example, if the screening result shows that the temperature sensor of the candidate parameter (i1, j1) at the wind tunnel level j1 and the air flow velocity sensor of the candidate parameter (i2, j2) at the wind tunnel level j2 both have higher load during operation, the key parameter set ;

[0121] Thus, from the numerous data collection wind tunnel levels and numerous types of physical sensors, the combination of the wind tunnel level most affected by the fixed geometric parameters and the physical sensors is selected, which reduces the difficulty of subsequent data processing.

[0122] In the embodiments of the present application, the constructing the dynamic heat transfer resistance coefficient matrix based on the coupling features and the multi-source sensing data comprises the following steps:

[0123] Step 221: setting the dynamic heat transfer resistance coefficient matrix as R(t), wherein R(t) is a matrix related to time t, reflecting the heat transfer resistance of different air duct levels in the rotary kiln;

[0124] Specifically, the element of the matrix R(t) represents the heat transfer resistance coefficient between the j1th air duct level and the j2th air duct level at time t;

[0125] Step 222: for the jth air duct level, reading the type of physical sensor contained in the jth air duct level from the key parameter set of the coupling features, and combining the material filling rate of the jth air duct level and the geometric parameters of the stepped air duct to form the key parameters;

[0126] Step 223: describing the motion characteristics of the airflow in the stepped air duct through the fluid dynamics equation, and combining the heat transfer model to describe the heat transfer according to the temperature distribution, to obtain the heat transfer resistance coefficient of any two air duct levels;

[0127] Specifically, the heat transfer model is, for example, the Fourier heat conduction equation;

[0128] And the heat transfer resistance coefficient of any two air duct levels can be expressed as:

[0129] wherein, contains the influence of fluid mechanics and heat conduction, and the specific form and parameters are fitted according to historical data; and respectively represent the key parameters of the j1th air duct level and the j2th air duct level, is a correction factor that needs to be fitted;

[0130] Step 224: collecting the actual heat transfer resistance coefficient between each air duct level during the historical operation of the stepped air duct, and using regression analysis, neural network or other optimization algorithms to fit the parameters in the calculation equation of the heat transfer resistance coefficient;

[0131] In further preferred embodiments, since the operation process of the rotary kiln has dynamic nature, the heat transfer resistance coefficient matrix needs to be dynamically adjusted over time; therefore, new heat transfer resistance coefficients can also be dynamically calculated according to the process parameters collected by each physical sensor of each air duct level, to realize the correction of the heat transfer resistance coefficient matrix according to the current working conditions; this correction method can also be statistically and fitted based on historical data using regression analysis, neural network or other optimization algorithms.

[0132] Further, the inputting the coupling feature vector and the dynamic heat transfer resistance coefficient matrix into the pre-trained gradient boosting regression tree model and outputting the real-time effective heat transfer area proportion of each air duct level comprises the following steps:

[0133] Step 31: Collecting the coupling feature vector and the heat transfer resistance coefficient matrix at each time point during the historical operation of the stepped air duct to form a training input sample, and collecting the real-time effective heat transfer area proportion of each air duct level to form a training label;

[0134] Specifically, the coupling feature vector is the parameter value of the key parameter of each air duct level at each time point during the historical operation extracted through the above-mentioned process of extracting the coupling feature;

[0135] The heat transfer resistance coefficient matrix is obtained through the above-mentioned process of extracting the heat transfer resistance coefficient matrix;

[0136] The real-time effective heat transfer area proportion represents the heat transfer capacity of each air duct level at the time point, and the actual value of the effective heat transfer area proportion is generally obtained through experimental data or numerical simulation.

[0137] Step 32: Constructing a gradient boosting regression tree model comprising a plurality of decision trees;

[0138] Specifically, the structure of the gradient boosting regression tree model is composed of a plurality of decision trees;

[0139] The structure of each decision tree is composed of a plurality of branch nodes and leaf nodes, each branch node is split according to a certain dimension of the input feature, and the target is to make the mean value of each subset after splitting in the output label, i.e., the effective heat transfer area proportion, as similar as possible, and each leaf node contains a final regression value, i.e., the prediction value of the model for a certain input feature;

[0140] Each decision tree gradually reduces the prediction error by fitting the residual of the previous decision tree, and the output of each decision tree is weighted and summed with the output of the previous decision tree to obtain the final prediction result;

[0141] The input of the gradient boosting regression tree model is the combination of the coupling feature vector and the dynamic heat transfer resistance coefficient matrix, and the output of the gradient boosting regression tree model is a continuous value representing the real-time effective heat transfer area proportion of each air duct level;

[0142] Step 33: Training the gradient boosting regression tree model using the training input sample and the training label;

[0143] Specifically, the training method of the gradient boosting regression tree model is as follows:

[0144] Initialize the gradient boosting regression tree model, that is, initialize the gradient boosting regression tree model to a constant value, and select the average value of the effective heat transfer area proportion in the training data as the initial value;

[0145] For each training input sample, calculate the current residual of the gradient boosting regression tree model, which is the difference between the actual value and the predicted value;

[0146] Each round of training fits the current residual by building a new decision tree, which optimizes the residual so that the output value of the gradient boosting regression tree model approaches the actual value of the training label;

[0147] The training of each decision tree depends on the error of the last round of training, and the decision tree finds the best split point by dividing the training input samples to minimize the output error on each branch;

[0148] The output of the final gradient boosting regression tree model is the weighted sum of all decision trees.

[0149] Step 34: When the gradient boosting regression tree model training is completed, real-time target input samples are formed by collecting real-time coupling features and dynamic heat transfer resistance coefficient matrix, the target input samples are input into the gradient boosting regression tree model, and the real-time effective heat transfer area proportion of each air duct level output by the gradient boosting regression tree model is obtained;

[0150] It should be noted that first, the extracted coupling feature vector can compress a few important principal components from the highly correlated multi-dimensional features, greatly reducing the dimension of the input data, and second, the dynamic heat transfer resistance coefficient matrix integrates the influence of air duct geometry and physical characteristics on heat transfer, and can consider the heat transfer relationship between different air duct levels without separately inputting detailed parameters of each physical characteristic of the air duct level. Through the combination of the coupling feature vector and the dynamic heat transfer resistance coefficient matrix, the complex coupling relationship inside the system can be captured, that is, the coupling feature vector captures the main trend of the sensor data, and the dynamic heat transfer resistance coefficient matrix can combine the geometry of the air duct level, airflow and heat transfer state, thereby improving the prediction accuracy of the real-time effective heat transfer area proportion.

[0151] Further, the adaptive particle swarm optimization algorithm is used to optimize the stepped air duct parameters based on the real-time effective heat transfer area proportion, which includes the following steps:

[0152] Step 41: Construct a target optimization function according to the real-time effective heat transfer area proportion, and select the optimized stepped air duct parameters;

[0153] In an embodiment of the present application, the expression of the target optimization function is: ; wherein J is the number of all air duct levels; is the effective heat transfer area ratio of the jth air duct level predicted by the gradient boosting regression tree model; and the optimization objective of the target optimization function is to maximize the function value of the target optimization function;

[0154] The optimized stepped air duct parameters in the embodiments of the present application include a gas injection valve opening G and a guide vane angle θ.

[0155] The opening of the gas injection valve directly affects the flow rate and distribution of the gas in the kiln, thereby affecting the gas flow dynamics and heat transfer.

[0156] The guide vane angle affects the direction of the gas flow, thereby affecting the distribution and transfer effect of heat in the kiln.

[0157] Step 42: design a fitness function based on the target optimization function;

[0158] In the embodiments of the present application, the design of the fitness function is as follows:

[0159] The fitness function is expressed as follows:

[0160] wherein, and are the initial gas injection valve opening and guide vane angle, respectively, representing the initial or reference state of the current rotary kiln, is a penalty coefficient for controlling the amplitude of changes in the gas injection valve opening and the guide vane angle during the optimization process to avoid excessive adjustment;

[0161] The fitness function is a standard for the particle swarm optimization algorithm to evaluate the pros and cons of the solution represented by each particle. Therefore, the above fitness function aims to maximize the effective heat transfer area ratio while limiting the changes in the gas injection valve opening and the guide vane angle, so that the parameter adjustment during the optimization process is not too drastic, ensuring stable operation of the rotary kiln.

[0162] Step 43: initialize the particle swarm of the adaptive particle swarm optimization algorithm;

[0163] In the embodiments of the present application, the initialization of the particle swarm of the adaptive particle swarm optimization algorithm is as follows:

[0164] Generate a set of random particles, each particle representing a set of stepped air duct parameters, i.e., each particle represents a possible solution;

[0165] ​A movement range is preset for each parameter of each particle, so that the particle is limited in a reasonable movement range when being updated in each generation, in the embodiment, the gas injection valve opening degree is set in a reasonable range, such as between 0% and 100%; the guide vane angle is set in a range of 0 to 90 degrees;

[0166] The position of each particle is updated by using a particle updating formula as follows:

[0167] wherein, and denote the gas injection valve opening degree at t+1 and t respectively, and denotes the velocity updating value of the particle in the gas injection valve opening degree, and denote the guide vane angle at t+1 and t respectively, denotes the velocity updating value in the guide vane angle direction.

[0168] Step 44: applying the adaptive particle swarm algorithm to optimize the position of the particle swarm, gradually approaching the optimal solution, and obtaining the optimized stepped air duct parameters;

[0169] In the embodiment of the application, the process of applying the adaptive particle swarm algorithm to optimize the position of the particle swarm and gradually approaching the optimal solution is as follows:

[0170] The velocity and position of each particle are updated based on the current fitness function, the individual optimal position and the global optimal position according to the following formula:

[0171] The velocity updating formula is as follows:

[0172]

[0173] ,

[0174] wherein, is an inertia weight, controlling the influence of the particle's own velocity, and are acceleration constants, determining the degree of approaching the individual optimal position and the global optimal position, is the optimal position of the individual gas injection valve opening degree, is the optimal position of the individual guide vane angle; and are random numbers generated randomly, ensuring the randomness of the search; and are the global optimal position of the particle's gas injection valve opening degree and the global optimal position of the guide vane angle respectively;

[0175] It should be noted that the individual optimal position is the best solution found by each particle in the history of optimization, that is, the position with the optimal fitness that the particle has experienced before; and the global optimal position is the position of the optimal solution in the particle swarm, which guides the search process of the particle swarm based on the overall experience of the particle swarm; the updating method of the global optimal position is: in each iteration, each particle in the particle swarm will compare with the individual optimal position of other particles, if the fitness value of the individual optimal position of a particle is better than the current global optimal fitness, the global optimal position is updated to the individual optimal position of the particle;

[0176] It is easy to understand that in the above speed updating formula, the first formula, that is, and , are inertia terms, which can be regarded as the "inertia" of the particle to continue to move according to its own historical speed, so as to keep the original movement trend of the particle, thereby avoiding the particle completely stopping or losing the motive power of movement; and the second formula, that is, and , reflect how the particle moves to its individual optimal position, and the third formula, that is, and , reflect the process of how the particle moves to the global optimal position of the whole group.

[0177] Further, the way of adjusting the optimized ladder air duct parameters in the rotary kiln in real time through the PID controller to form a closed loop control is:

[0178] The optimized ladder air duct parameters are input into the PID controller, and the optimized ladder air duct parameters are taken as the adjustment target value, and the PID controller adjusts the ladder air duct parameters in the rotary kiln to the adjustment target value according to the adjustment target value.

[0179] Embodiment 2

[0180] As shown in Figure 3 , the lithium ore lithium extraction rotary kiln ladder air duct heat transfer structure optimization system comprises a multi-source data collection module, a feature extraction module, a heat transfer area ratio output module and an air duct parameter optimization module; wherein each module is connected through electrical connection.

[0181] The multi-source data collection module collects real-time multi-source sensing data in the rotary kiln ladder air duct, and sends the multi-source sensing data to the feature extraction module;

[0182] The feature extraction module extracts coupling features from the multi-source sensing data using principal component analysis method combined with the geometric parameters of the ladder air duct, and constructs a dynamic heat transfer resistance coefficient matrix based on the coupling features and the multi-source sensing data, and sends the coupling features and the dynamic heat transfer resistance coefficient matrix to the heat transfer area ratio output module.

[0183] The heat transfer area ratio output module inputs the coupling feature and the dynamic heat transfer resistance coefficient matrix into a pre-trained gradient boosting regression tree model, outputs the real-time effective heat transfer area ratio of each air duct level, and sends the real-time effective heat transfer area ratio to the air duct parameter optimization module;

[0184] The air duct parameter optimization module optimizes the stepped air duct parameters based on the real-time effective heat transfer area ratio using an adaptive particle swarm algorithm, adjusts the stepped air duct parameters in the rotary kiln in real time through a PID controller to form a closed loop control.

[0185] In addition, the part of the above technical solution in the embodiments of the present application that is consistent with the implementation principle of the corresponding technical solution in the prior art is not described in detail to avoid excessive repetition.

[0186] The specific embodiments described above further illustrate the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0187] The above preset parameters or preset thresholds are set by a person skilled in the art according to actual conditions or obtained through a large amount of data simulation.

[0188] The above embodiments are only used to illustrate the technical method of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A method for optimizing the heat transfer structure of a rotary kiln for lithium extraction from lithium ore, characterized in that, The method comprises the following steps: Step 1: Real-time acquisition of multi-source sensing data in the rotary kiln ladder air duct; Step 2: Combined with the geometric parameters of the ladder air duct, the principal component analysis method is used to extract the coupling features from the multi-source sensing data, and based on the coupling features and the multi-source sensing data, a dynamic heat transfer resistance coefficient matrix is constructed; Step 3: The coupling features and the dynamic heat transfer resistance coefficient matrix are input into the pre-trained gradient boosting regression tree model, and the real-time effective heat transfer area proportion of each air duct level is output; Step 4: Based on the real-time effective heat transfer area proportion, the adaptive particle swarm optimization algorithm is used to optimize the ladder air duct parameters; Step 5: The optimized ladder air duct parameters are adjusted through the PID controller to adjust the ladder air duct parameters in the rotary kiln in real time, forming a closed loop control.

2. The method of claim 1, wherein the method is characterized by: The real-time acquisition of multi-source sensing data in the rotary kiln ladder air duct comprises the following steps: Step 11: Install physical sensors corresponding to each process parameter and material filling rate in each ladder layer respectively; Step 12: All physical sensors access the database background through the industrial Internet of Things communication network, and collect the corresponding process parameters or material filling rate according to the unified preset sampling frequency, and then send the process parameters or material filling rate to the database background; Step 13: In the process of collecting process parameters and material filling rate in the database background, a unified time stamp is added to each physical sensor data; the process parameters, material filling rate and geometric parameters of each time stamp are uniformly aggregated into a multi-dimensional data matrix, and all multi-dimensional data matrices are structured and stored as multi-source sensing data in time sequence.

3. The method of claim 2, wherein the method further comprises: The way of uniformly aggregating into a multi-dimensional data matrix and structuring and storing all multi-dimensional data matrices as multi-source sensing data in time sequence is: The process parameters and the material fill rate from each physical sensor are arranged in a time series into a multidimensional data matrix, wherein each row represents a physical sensor and each column represents a wind tunnel level of the ladder wind tunnel, and using a marker; denotes the measured value of the i-th physical sensor at the j-th wind tunnel level at time t. Save the multi-dimensional data matrix under each time stamp in the database background in time sequence.

4. The method of claim 3, wherein the method further comprises: The combination of the geometric parameters of the ladder air duct and the extraction of the coupling features from the multi-source sensing data using the principal component analysis method comprises the following steps: Step 211: Pre-read the geometric parameters of the ladder air duct from the roasting control background, and construct the geometric parameter vector with all the geometric parameters; Step 212: The combination of each air duct level and each physical sensor is represented as a candidate parameter, and for each candidate parameter and geometric parameter vector, an extended feature vector is constructed for each moment; all geometric parameters form a key parameter set; Step 213: All candidate parameters are aggregated into an extended feature matrix at each moment; Step 214: standardize each row of the extended feature matrix Y(t) to obtain a standardized extended feature matrix, denoted as ; Step 215: Collecting a plurality of standardized extended feature matrices at different time points from historical operation periods of the stairway air duct , to form a historical three-dimensional matrix; wherein N is the number of time points collected; Step 216: Based on the historical three-dimensional matrix, the global mean and covariance matrix are calculated; Step 217: eigen-decomposition of the covariance matrix to obtain eigenvalues to and corresponding eigenvectors to , all eigenvectors form the principal component matrix ; Step 218: Normalized extended feature matrix for time t Data projection is performed using a fixed global model, and the projection result of each candidate parameter reflects its wind tunnel level in the global principal component space. Step 219: In the projection result of each candidate parameter, the load of the first component of the corresponding extended vector in each principal component is analyzed, and a load threshold is preset. If the absolute value of the load corresponding to the preset number of principal components is greater than the load threshold, the candidate parameter is supplemented to the key parameter set; the finally generated key parameter set constitutes the coupling features.

5. The method of claim 4, wherein the method further comprises: The construction of the dynamic heat transfer resistance coefficient matrix based on the coupling features and the multi-source sensing data comprises the following steps: Step 221: Set the dynamic heat transfer resistance coefficient matrix as R(t); Step 222: For the jth air duct level, read the type of physical sensor contained in the jth air duct level from the key parameter set of the coupling feature, and combine the material filling rate of the jth air duct level and the geometric parameters of the stepped air duct to form a key parameter; Step 223: Describe the motion characteristics of the airflow in the stepped air duct by the fluid dynamics equation, and then combine the heat transfer model to describe the heat transfer according to the temperature distribution, to obtain the heat transfer resistance coefficient of any two air duct levels; Step 224: Collect the actual heat transfer resistance coefficient between each air duct level during the historical operation period of the stepped air duct, and use regression analysis or neural network to fit the parameters in the calculation equation of the heat transfer resistance coefficient.

6. The method of claim 5, wherein the method further comprises: The input of the gradient boosting regression tree model includes the coupling feature and the dynamic heat transfer resistance coefficient matrix, and the output is the real-time effective heat transfer area ratio of each air duct level. Step 31: Collect the coupling feature vector and the heat transfer resistance coefficient matrix at each time during the historical operation period of the stepped air duct to form a training input sample, and collect the real-time effective heat transfer area ratio of each air duct level to form a training label; Step 32: Construct a gradient boosting regression tree model containing multiple decision trees; Step 33: Train the gradient boosting regression tree model using the training input sample and the training label; Step 34: When the gradient boosting regression tree model is trained, input the real-time coupling feature and the dynamic heat transfer resistance coefficient matrix into the gradient boosting regression tree model to obtain the real-time effective heat transfer area ratio of each air duct level output by the gradient boosting regression tree model.

7. The method of claim 6, wherein the method further comprises: The structure of the gradient boosting regression tree model is composed of multiple decision trees; The structure of each decision tree is composed of multiple branch nodes and leaf nodes, each branch node is split according to the dimension of the input feature, and each leaf node contains the final regression value, which is the prediction value of the model for the input feature; Each decision tree reduces the prediction error by fitting the residual of the previous decision tree, and the output of each decision tree is weighted and summed with the output of the previous decision tree to obtain the final prediction result; The input of the gradient boosting regression tree model is the combination of the coupling feature vector and the dynamic heat transfer resistance coefficient matrix, and the output of the gradient boosting regression tree model is a continuous value representing the real-time effective heat transfer area ratio of each air duct level.

8. The method of claim 7, wherein the method further comprises: The training method of the gradient boosting regression tree model is as follows: Initialize the gradient boosting regression tree model, that is, initialize the gradient boosting regression tree model to a constant value, and select the average value of the effective heat transfer area ratio in the training data as the initial value; For each training input sample, calculate the current residual of the gradient boosting regression tree model, which is the difference between the actual value and the predicted value; Each round of training fits the current residual by constructing a new decision tree, which optimizes the residual; The training of each decision tree depends on the error of the previous round of training, and the decision tree finds the best split point by dividing the training input sample, so that the output error of each branch is minimized; The output of the final gradient boosting regression tree model is a weighted sum of all decision trees.

9. The method of claim 8, wherein the method further comprises: The step of optimizing the stepped air duct parameters based on the real-time effective heat transfer area ratio using the adaptive particle swarm optimization algorithm comprises the following steps: Step 41: constructing a target optimization function according to the real-time effective heat transfer area ratio, and selecting optimized stepped air duct parameters; The optimized stepped air duct parameters include the gas injection valve opening G and the guide vane angle θ. Step 42: designing a fitness function based on the target optimization function; Step 43: initializing the particle swarm of the adaptive particle swarm optimization algorithm; Step 44: optimizing the position of the particle swarm using the adaptive particle swarm optimization algorithm to obtain the optimized stepped air duct parameters.

10. The method of claim 9, wherein the method further comprises: The way of initializing the particle swarm of the adaptive particle swarm optimization algorithm is: Generate a group of random particles, each particle representing a set of stepped air duct parameters; Pre-set a movement range for each parameter of each particle, so that the particle is limited within the movement range when updated in each generation; Update the position of each particle using the particle update formula.

11. The method of claim 10, wherein the method further comprises: The way of adjusting the stepped air duct parameters in the rotary kiln in real time through the PID controller to form a closed-loop control is: Input the optimized stepped air duct parameters into the PID controller, and take the optimized stepped air duct parameters as the adjustment target value, and the PID controller adjusts the stepped air duct parameters in the rotary kiln to the adjustment target value according to the adjustment target value.

12. A lithium ore lithium extraction rotary kiln ladder draft channel heat transfer structure optimization system for implementing the lithium ore lithium extraction rotary kiln ladder draft channel heat transfer structure optimization method of any one of claims 1-11, characterized in that, It comprises a multi-source data collection module, a feature extraction module, a heat transfer area ratio output module, and an air duct parameter optimization module; wherein each module is connected through electrical connection; The multi-source data collection module collects real-time multi-source sensing data in the stepped air duct of the rotary kiln, and sends the multi-source sensing data to the feature extraction module; The feature extraction module uses principal component analysis to extract coupling features from the multi-source sensing data in combination with the geometric parameters of the stepped air duct, and constructs a dynamic heat transfer resistance coefficient matrix based on the coupling features and the multi-source sensing data, and sends the coupling features and the dynamic heat transfer resistance coefficient matrix to the heat transfer area ratio output module; The heat transfer area ratio output module inputs the coupling features and the dynamic heat transfer resistance coefficient matrix into the pre-trained gradient boosting regression tree model, outputs the real-time effective heat transfer area ratio of each air duct level, and sends the real-time effective heat transfer area ratio to the air duct parameter optimization module; The air duct parameter optimization module optimizes the stepped air duct parameters based on the real-time effective heat transfer area ratio using the adaptive particle swarm optimization algorithm, adjusts the stepped air duct parameters in the rotary kiln in real time through the PID controller, and forms a closed-loop control.

Images