Artificial intelligence-based lithium concentration prediction method and device

Abstract

The lithium concentration prediction method based on artificial intelligence according to one embodiment includes a cleaning step of obtaining a plurality of first variables by preprocessing a plurality of initial variables used in an electrodialysis apparatus for producing lithium, a preprocessing step of preprocessing the plurality of first variables by inputting the plurality of first variables to a Boosting series artificial intelligence model, and obtaining a final variable, and a prediction step of predicting the lithium concentration by inputting the final variable to the Boosting series artificial intelligence model.

Description

Technical Field

[0001] This invention relates to a method and apparatus for predicting lithium concentration based on artificial intelligence. More specifically, this invention relates to a method and apparatus for predicting lithium concentration based on artificial intelligence, which utilizes artificial intelligence to predict lithium concentration with minimal variables. Background Technology

[0002] Bipolar electrodialysis (BPED) is an electro-separation process used to separate or concentrate specific ions from a solution using ion exchange materials. BPED has attracted considerable attention as a particularly important water separation and regeneration technology.

[0003] BPED operational data can include various data collected during the operation of the technology. Examples of BPED operational data include current and voltage data, substance concentration data, temperature and pressure data, and time data.

[0004] To improve experience-based BPED and increase lithium (Li) production, a model is needed to predict the lithium concentration produced based on operational data. The prediction results can then be used for process control or optimization. Summary of the Invention

[0005] (a) Technical problems to be solved One embodiment of the present invention aims to provide an artificial intelligence-based lithium concentration prediction method and apparatus. It applies a series of Boosting artificial intelligence models during the preprocessing process, performs a first preprocessing of variables based on the importance of each variable according to the permutation importance, and performs a second preprocessing of variables based on the mean absolute percentage error (MAPE), thereby effectively predicting lithium concentration with the fewest variables.

[0006] (II) Technical Solution In some embodiments, an artificial intelligence-based lithium concentration prediction method includes: a cleaning step, which preprocesses multiple initial variables used in an electrodialysis device for lithium production to obtain multiple first variables; a preprocessing step, which inputs the multiple first variables into a Boosting series artificial intelligence model for preprocessing and obtains final variables; and a prediction step, which inputs the final variables into the Boosting series artificial intelligence model to predict the lithium concentration.

[0007] The cleaning steps may include a data cleaning step, used to remove missing values, remove data from non-operation days, and remove data for processes other than the BPED process.

[0008] The cleaning step may further include: detecting and removing at least one initial variable from the initial variables whose correlation with each other is above a certain threshold through correlation analysis.

[0009] The Boosting series of artificial intelligence models can be selected from a group consisting of AdaBoost, Random Forest, Catboost, Gradient Boosting Model, Lightweight Gradient Boosting Machine (Light GBM), and Extreme Gradient Boosting (XGBoost).

[0010] The preprocessing step includes a first preprocessing step, which is used to input the first variable into the Boosting series of artificial intelligence models to calculate the variable importance of each of the first variables, and obtain multiple second variables through preprocessing based on the variable importance. The variable importance can be permutation importance.

[0011] The first preprocessing step may include the step of removing first variables whose permutation importance is less than a preset second benchmark.

[0012] The preprocessing step further includes a second preprocessing step, which is used to input the second variable into the Boosting series artificial intelligence model to obtain multiple third variables that meet the prediction fit requirements above the first benchmark with a minimum number of variables. The second preprocessing step may include a step of calculating the prediction fit using the mean absolute percentage error (MAPE).

[0013] The second preprocessing step may further include a step of calculating the variable importance of each of the acquired third variables.

[0014] The second preprocessing step may further include the following steps: comparing the goodness of fit when only the variable with the highest importance among the third variables is input into the Boosting series of artificial intelligence models with the predicted goodness of fit; if the predicted goodness of fit is greater, then the third variable is determined as the final variable.

[0015] The third variable can be selected from any one of the following: conductivity, circulation flow rate, and circulation pressure for any one of lithium sulfate aqueous solution, sulfuric acid, and lithium hydroxide.

[0016] In some embodiments, an artificial intelligence-based lithium concentration prediction device includes: a cleaning unit for obtaining a plurality of first variables by preprocessing a plurality of initial variables used in an electrodialysis device for lithium production; a preprocessing unit for inputting the plurality of first variables into a Boosting series artificial intelligence model for preprocessing and obtaining a final variable; and a concentration prediction unit for inputting the final variable into the Boosting series artificial intelligence model to predict the lithium concentration.

[0017] The cleaning unit can remove missing values, remove data from non-operation days, and remove data for processes other than the BPED process.

[0018] The cleaning unit can detect and remove at least one initial variable whose correlation with each other is above a certain threshold through correlation analysis.

[0019] The Boosting series of artificial intelligence models can be selected from a group consisting of AdaBoost, Random Forest, Catboost, Gradient Boosting Model, Lightweight Gradient Boosting Machine (Light GBM), and Extreme Gradient Boosting (XGBoost).

[0020] The preprocessing unit includes a first preprocessing unit, which is used to input the first variable into the Boosting series of artificial intelligence models to calculate the variable importance of each of the first variables, and obtain multiple second variables based on the variable importance through the first preprocessing. The variable importance can be obtained by permutation importance.

[0021] The first preprocessing unit can remove first variables whose arrangement importance is less than a preset second benchmark from the first variables.

[0022] The preprocessing unit further includes a second preprocessing unit, used to input the second variable into the Boosting series artificial intelligence model to obtain multiple third variables that meet the prediction fit requirements above the first benchmark with a minimum number of variables. The second preprocessing unit can calculate the prediction fit using the mean absolute percentage error (MAPE).

[0023] The second preprocessing unit can calculate the variable importance for each of the acquired third variables.

[0024] The second preprocessing unit can compare the goodness of fit when only the variable with the highest importance among the third variables is input into the Boosting series of artificial intelligence models with the predicted goodness of fit. If the predicted goodness of fit is greater, then the third variable is determined as the final variable.

[0025] The third variable can be selected from any one of the following: conductivity, circulation flow rate, and circulation pressure for any one of lithium sulfate aqueous solution, sulfuric acid, and lithium hydroxide.

[0026] (III) Beneficial Effects According to an embodiment of the present invention, an artificial intelligence-based lithium concentration prediction method and apparatus apply a Boosting series of artificial intelligence models during the preprocessing process. The variables are preprocessed for the first time based on the importance of each variable according to the permutation importance, and the variables are preprocessed for the second time based on the mean absolute percentage error (MAPE). Thus, lithium concentration can be effectively predicted using the fewest variables. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of an electrodialysis device according to an embodiment of the present invention.

[0028] Figure 2 This is a schematic diagram of one end of an electrodialysis device according to an embodiment of the present invention.

[0029] Figure 3 This is a block diagram of an artificial intelligence-based lithium concentration prediction device according to an embodiment of the present invention.

[0030] Figure 4 and Figure 5 This is a flowchart of an artificial intelligence-based lithium concentration prediction method according to an embodiment of the present invention.

[0031] Figure 6 This is a view illustrating the preprocessing steps of an artificial intelligence-based lithium concentration prediction method according to an embodiment of the present invention.

[0032] Figure 7 is a graph illustrating the effect of an artificial intelligence-based lithium concentration prediction method according to an embodiment of the present invention.

[0033] Figure 8 This is a view used to describe a computing device according to an embodiment of the present invention. Detailed Implementation

[0034] In the following description, embodiments of the present invention will be described in detail with reference to the accompanying drawings to facilitate implementation by those skilled in the art. However, the present invention can be implemented in many different ways and is not limited to the described embodiments. Furthermore, for the sake of clear description in the drawings, parts irrelevant to the description have been omitted, and similar reference numerals have been used for similar parts throughout the specification.

[0035] Throughout the specification and claims, when a portion is described as "comprising" a particular element, unless specifically stated otherwise, it indicates that other elements may be further included, not that other elements are excluded. Ordinal terms such as "first," "second," etc., may be used to describe various elements, but the elements are not limited by these terms. These terms are used only to distinguish one element from others.

[0036] The terms “...unit”, “...device”, “module”, etc. used in this specification may refer to a unit capable of performing at least one function or action described in this specification, which may be implemented by hardware or circuitry, software, or a combination of hardware or circuitry and software.

[0037] Various embodiments of the present invention will now be described with reference to the accompanying drawings.

[0038] Figure 1 This is a schematic diagram of an electrodialysis device according to an embodiment of the present invention. Figure 2 This is a schematic diagram of one end of an electrodialysis device according to an embodiment of the present invention.

[0039] exist Figure 1 and Figure 2 In this context, the electrodialysis equipment can be a bipolar electrodialysis (BPED) device. That is, the electrodialysis equipment can be a bipolar membrane electrodialysis device. An electrodialysis device (BPED) can be a device that converts an aqueous solution of lithium sulfate into lithium hydroxide and sulfuric acid. Here, lithium sulfate is Li₂SO₄, lithium hydroxide is LiOH, and sulfuric acid is H₂SO₄. An electrodialysis device (BPED) can also be an aqueous solution treatment device that simultaneously performs water splitting / ion separation using an electrodialysis membrane within an electric field.

[0040] Reference Figure 1 and Figure 2 Electrodialysis equipment (BPED) may include cation exchange membranes (CEM), anion exchange membranes (AEM), and bipolar membranes (BPM).

[0041] Cation exchange membranes (CEMs) have internal anionic groups, thus allowing only cations (e.g., Li) to pass through. +Anion exchange membranes (AEMs) allow only anions (e.g., SO42-) to pass through their internal cation groups. 2- The bipolar membrane (BPM) is composed of a cation exchange membrane and an anion exchange membrane, with a water splitting catalyst sandwiched between them. The BPM can split water within an electric field, thereby generating hydrogen ions (H+). + ), hydroxide ions (OH-) - ).

[0042] In other words, a bio-dilation device (BPED) can simultaneously perform water splitting (decomposition into H+) using an electrodialysis membrane (cation exchange membrane, anion exchange membrane, bipolar membrane) within an electric field. + OH - ) / Ion separation (Li + SO4 2- Aqueous solution treatment equipment for ion separation.

[0043] exist Figure 1 In a BPED electrodialysis unit, through a three-stage process (Press) comprising stages 1 to 3, the lithium sulfate (LS) solution can transfer Li and SO4 ions to the lithium hydroxide (LH) solution and the sulfuric acid (H2SO4) solution. For example, in a BPED unit, deionized water (DI water) can be contacted with the LS solution in a counter-flow manner and converted into LH and sulfuric acid solutions. Here, the LS solution is lithium sulfate (Li2SO4), and the LH solution is lithium hydroxide (LiOH).

[0044] Levels 1 through 3 can respectively include a salt chamber, an acid chamber, an alkali chamber, a salt tank, an acid tank, and an alkali tank.

[0045] In each stage, the salt chamber can be supplied with lithium sulfate (Li₂SO₄) to produce desalted water after the reaction. The acid chamber can be supplied with deionized water (DI water) to produce sulfuric acid (H₂SO₄) after the reaction. The alkali chamber can be supplied with deionized water (DI water) to produce lithium hydroxide (LiOH) after the reaction.

[0046] Deionized water produced is stored in a salt tank. Sulfuric acid produced is stored in an acid tank. Lithium hydroxide produced is stored in an alkali tank.

[0047] The production capacity of a BPED (Biodialysis Electrodialysis) unit depends on the discharge flow rates of sulfuric acid and lithium hydroxide. Production capacity is also dependent on the control of the water (H₂O) inflow rate, lithium sulfate inflow rate, rectifier current and voltage, and the management of pH, conductivity, circulation flow rate, and circulation pressure in each chamber. The solution inflow rate, rectifier current, voltage, pH, conductivity, circulation flow rate, and circulation pressure, corresponding to the control and management elements, can be detected by sensors installed in each chamber. These control and management elements correspond to the variables that determine the concentrations of lithium hydroxide and sulfuric acid produced.

[0048] Figure 3 This is a block diagram of an artificial intelligence-based lithium concentration prediction device according to an embodiment of the present invention.

[0049] An AI-based lithium concentration prediction device 100 predicts lithium (Li) production concentration by performing specialized preprocessing on operational data, including control and management elements of the electrodialysis equipment (BPED). The AI-based lithium concentration prediction device 100 utilizes the Boosting series of AI models in both operational data preprocessing and lithium concentration prediction, reducing noise and computational load, quantifying the derivation of key factors, and improving the fitting accuracy. In other words, the AI-based lithium concentration prediction device 100 executes an AI-based lithium concentration prediction method, which uses the Boosting series of AI models to predict lithium (Li) production concentration through specialized preprocessing of operational data. A detailed description of the AI-based lithium concentration prediction method will be provided below. Figures 4 to 6 To elaborate.

[0050] Reference Figure 3 The lithium concentration prediction device 100 based on artificial intelligence may include a cleaning unit 110, a first pretreatment unit 120, a second pretreatment unit 130, and a concentration prediction unit 140.

[0051] The first preprocessing unit 120 and the second preprocessing unit 130 can be collectively referred to as preprocessing units. Preprocessing units can preprocess multiple initial variables and obtain final variables. Cleaning unit 110 obtains multiple first variables by preprocessing multiple initial variables used in the lithium-ion electrodialysis (BPED) equipment. Cleaning unit 110 removes missing values, outliers, duplicate data, data from non-operating days (sensor data from non-operating days), and data related to processes other than BPED by performing data cleaning preprocessing on initial variables including all variables of the BPED process. Cleaning unit 110 can also remove synchronously changing data as duplicate data based on correlation analysis of the data.

[0052] The cleaning unit 110 can obtain a first variable that is reduced in number relative to the initial number of variables through data cleaning.

[0053] The first preprocessing unit 120 can input the first variable into the Boosting series artificial intelligence model to calculate the variable importance of each of the first variables.

[0054] Boosting-based AI models are AI models that utilize the Boosting algorithm. Boosting is a type of machine learning algorithm that combines weak learners to build a strong learner. Boosting methods can be used for classification and regression problems. The Boosting algorithm trains the models sequentially. At each stage, the Boosting algorithm adds a new model to correct the errors of the previous models. The Boosting algorithm combines all the models to generate the final prediction.

[0055] The Boosting family of AI models can be selected from a group consisting of AdaBoost, Random Forest, Catboost, Gradient Boosting Model, Lightweight Gradient Boosting Machine (Light GBM), and Extreme Gradient Boosting (XGBoost). Extreme Gradient Boosting (XGBoost) is the preferred Boosting family of AI models.

[0056] Adaptive Boosting (AdaBoost) is one of the early Boosting algorithms. It works by assigning weights to each data point and assigning more weights to misclassified data points during training.

[0057] Gradient Boosting Machine (GBM) is a method of training a model with the goal of reducing residuals (errors). The residuals at each stage are calculated using the gradient of the loss function.

[0058] Extreme Gradient Boosting (XGBoost) is an extended version of GBM, which includes features such as regularization, parallel processing, and missing value handling.

[0059] Lightweight Gradient Boosting Machine (Light GBM) is a boosting algorithm specifically designed for large datasets, characterized by fast training speed and high memory efficiency. Like Extreme Gradient Boosting (XGBoost), it includes regularization and parallel processing features.

[0060] Catboost is a boosting algorithm specifically designed for categorical data, offering automatic categorical feature transformation and faster training speed.

[0061] The Boosting series of AI models, used in regression problems such as production concentration prediction, involves preprocessing data, training the model, and performing performance evaluation and feature importance analysis. If a satisfactory model is obtained through this process, it can be applied to real-world environments to predict concentrations in real time and for production control and optimization.

[0062] In this process, the Boosting series of artificial intelligence models can identify which variables have the greatest impact on production concentration prediction by analyzing the importance of features or variables.

[0063] In other words, the first preprocessing unit 120 analyzes the importance of variables to the first variable through the Boosting series of artificial intelligence models, and can identify which variables in the first variable have a greater impact on the prediction of lithium production concentration.

[0064] The first preprocessing unit 120 can use permutation importance to identify the importance of variables. Permutation importance is a methodology that estimates feature importance by determining how much the model's performance metrics (accuracy, F1 score, coefficient of determination (R^2), etc.) decrease when a certain feature is excluded from the artificial intelligence model.

[0065] The first preprocessing unit 120 can obtain multiple second variables based on the calculated variable importance through first preprocessing. For example, the first preprocessing unit 120 can remove first variables whose permutation importance is lower than a preset specific benchmark from the first variables. That is, the first preprocessing unit 120 can, based on the permutation importance of each first variable, assume that the lower the variable importance, the lower the impact on the performance of the artificial intelligence model, and reduce data noise by removing variables whose importance is lower than the specific benchmark, thereby improving the lithium concentration prediction performance of the artificial intelligence model. The second variables may include the variables remaining after removing variables whose permutation importance is lower than the specific benchmark (second benchmark) from the first variables.

[0066] The second preprocessing unit 130 can input the second variable into the Boosting series artificial intelligence model to obtain multiple third variables that meet the prediction fit above the first benchmark with a minimum number of variables.

[0067] Boosting-based AI models utilize preprocessed data to generate models and evaluate their performance, thereby improving the model's predictive performance. In one embodiment, the second preprocessing unit 130 uses a second variable to evaluate the model's performance and obtains a third variable that, based on the evaluation results, demonstrates a predictive fit above a specific benchmark. For example, the second preprocessing unit 130 can use evaluation metrics such as MAE (mean absolute error), RMSE (root mean square error), and MAPE (mean absolute percentage error).

[0068] MAE (Mean Absolute Error) is the average of the absolute values ​​of all prediction errors. RMSE (Root Mean Square Error) is the square root of the average of the squares of all prediction errors. MAPE (Mean Absolute Percentage Error) is the average of the absolute values ​​of all prediction errors divided by the actual values. In other words, it is the percentage of the difference between the actual and predicted values ​​divided by the actual value. The result can be expressed as a percentage. MAPE is a commonly used value in regression problems / models as a tool for evaluating prediction fit. Because it represents the proportion of relative error, it allows for a direct interpretation of the prediction results.

[0069] The second preprocessing unit 130 can use the mean absolute percentage error to calculate the prediction fit of the artificial intelligence model using the second variable, and obtain a third variable from a set of multiple variables whose calculated prediction fit satisfies a preset specific benchmark (first benchmark) and is included in a set consisting of the fewest possible variables.

[0070] The concentration prediction unit 140 can calculate the variable importance of each of the acquired third variables. The concentration prediction unit 140 can compare the prediction fit when only the variable with the highest variable importance among the third variables is input into the Boosting AI model with the prediction fit when all third variables are input. If the prediction fit when all third variables are input is greater than the prediction fit when only the variable with the highest variable importance is input, then the concentration prediction unit 140 can determine the third variable as the final variable and input it into the Boosting AI model to predict the lithium concentration.

[0071] The third variable can be selected from any one of the following: conductivity, circulation flow rate, and circulation pressure for any one of lithium sulfate aqueous solution, sulfuric acid, and lithium hydroxide.

[0072] Figure 4 and Figure 5 This is a flowchart of an artificial intelligence-based lithium concentration prediction method according to an embodiment of the present invention. Figure 6This is a view illustrating the preprocessing steps of an AI-based lithium concentration prediction method according to an embodiment of the present invention. The AI-based lithium concentration prediction method can utilize an AI-based lithium concentration prediction device 100 (see...). Figure 3 )implement.

[0073] exist Figure 4 In this process, the AI-based lithium concentration prediction device 100 obtains a first variable by preprocessing the initial variables used in the biodialysis equipment (BPED) for lithium production (step S100). The initial variable may include BPED operating data. The BPED operating data may include data on control and management elements that determine the production results of the biodialysis equipment (BPED). In the AI ​​model, the operating data may include features or variables.

[0074] For example, the AI-based lithium concentration prediction device 100 can obtain the first variable in Table 2 by preprocessing the initial variables in Table 1 using data cleaning. Data cleaning may include removing missing values, outliers, duplicate data, data from non-operational days (sensor data from non-operational days), and data for processes other than BPED.

[0075] Table 1 Table 1 lists a total of 525 initial variables. These initial variables include missing values, outliers, duplicate data, and data with the same correlation. Figure 6 This is a view showing the variables removed through correlation analysis during data cleaning. Figure 6 In the diagram, the adjustment values ​​for the first, third to fifth, seventh to ninth, and eleventh and twelfth cooling water valves represent the cooling water valve adjustment values ​​for the corresponding sensors. The first to twelfth temperature values ​​represent the temperature values ​​for the corresponding first to twelfth sensors. In other words, Figure 6 This illustrates the correlation between cooling water valve adjustment values ​​and temperature values. (Refer to...) Figure 6 The adjustment values ​​of the first cooling water valve, the third to fifth cooling water valves, the seventh to ninth cooling water valves, the eleventh and twelfth cooling water valves, and the first to twelfth temperature values ​​represent the cooling water valve adjustment values ​​and temperature values ​​measured by the corresponding sensors within the same device.

[0076] In other words, within the same device, the correlation between the temperature value of the same sensor and the cooling water valve adjustment value is 1, thus they are essentially duplicate variables. For example, the correlation between the first cooling water valve adjustment value measured by the first sensor and the first temperature value is 1. Therefore, the AI-based lithium concentration prediction device 100 can remove any one of the variables—the first cooling water valve adjustment value, the third to fifth cooling water valve adjustment values, the seventh to ninth cooling water valve adjustment values, the eleventh and twelfth cooling water valve adjustment values, or the first temperature value, the third to fifth temperature values, the seventh to ninth temperature values, and the eleventh and twelfth temperature values—through data cleaning preprocessing.

[0077] In addition, the AI-based lithium concentration prediction device 100 can eliminate process-related variables other than the BPED process (e.g., tank levels corresponding to subsequent BPED processes, a total of 86).

[0078] Table 2 In Table 2, there are a total of 400 first variables. The first variables include those remaining after some variables have been removed from the initial variables through data cleaning preprocessing. That is, 125 variables can be removed during the cleaning step. The AI-based lithium concentration prediction device 100 can input the first variables into a Boosting series AI model to calculate the importance of each first variable and perform preprocessing to remove inferior factors based on the calculated variable importance, thereby obtaining the second variables (step S200). In one embodiment, the AI-based lithium concentration prediction device 100 can calculate the permutation importance of each of the 400 first variables. For example, the AI-based lithium concentration prediction device 100 can obtain the second variables in Table 3, which includes the top 100 variables from the 400 first variables in Table 2 based on permutation importance. Here, the top 100 are arbitrarily set numbers that can be flexibly determined.

[0079] Table 3 The AI-based lithium concentration prediction device 100 obtains the main variables that yield a prediction fit above a specific benchmark for lithium concentration with the minimum number of variables input into the Boosting series AI model (step S300). Based on the 100 second variables in Table 2, the AI-based lithium concentration prediction device 100 obtains the 4 third variables in Table 4. These third variables, being the fewest variables input into the AI ​​model to predict lithium concentration with a prediction fit above a specific benchmark, can be referred to as the main variables.

[0080] Table 4 The AI-based lithium concentration prediction device 100 starts with 100 second variables and, by eliminating them one by one and inputting them into a Boosting series of AI models, can compare the mean absolute percentage error (MAPE) for each number of variables. The mean absolute percentage error can be used as a benchmark for judging the goodness of fit. When the prediction goodness of fit based on the mean absolute percentage error is above the first benchmark (e.g., above 95%, with a MAPE benchmark below 5), and the number of primary variables is minimized, the AI-based lithium concentration prediction device 100 can identify variables that satisfy the second benchmark of importance. For example, the AI-based lithium concentration prediction device 100 can determine that the case with four final primary variables in Table 4 represents the fewest important variables with a high goodness of fit. The AI-based lithium concentration prediction device 100 can compare the training and evaluation results of the AI ​​model based on the selected four primary variables with the training and evaluation results of the AI ​​model based on one primary variable (e.g., the third conductivity value of the third sensor, which has the highest importance). The AI-based lithium concentration prediction device 100 ultimately determines the four main variables based on the best evaluation results (prediction fit) using the four main variables as a benchmark.

[0081] The AI-based lithium concentration prediction device 100 can predict the lithium concentration produced by BPED by inputting the acquired main variables into the Boosting series of AI models (step S400). The AI-based lithium concentration prediction device 100 can predict the lithium concentration using the four main variables in Table 4.

[0082] In one embodiment, the third variable (primary variable) may be selected from any one of the conductivity, influent flow rate, circulating flow rate, and circulating pressure for any one of lithium sulfate aqueous solution, sulfuric acid, and lithium hydroxide in the multi-stage process.

[0083] For example, the third conductivity value could be the conductivity of lithium hydroxide in the first-stage base chamber (base) measured by the third sensor, the tenth flow rate value could be the circulating flow rate of sulfuric acid in the second-stage acid chamber (acid) measured by the tenth sensor, the eighth flow rate value could be the circulating flow rate of the first-stage electrode solution measured by the eighth sensor, and the eleventh conductivity value could be the conductivity of lithium hydroxide in the third-stage base chamber (base) measured by the eleventh sensor. Figure 5 Specifically shown Figure 4 One embodiment. Figure 5 An example of the preprocessing of BPED runtime data and lithium concentration prediction process using the Boosting series of artificial intelligence models is shown in sequence.

[0084] exist Figure 5First, the AI-based lithium concentration prediction method may include a cleaning step (step S110) of preparing and cleaning the dataset of BPED running data. The AI-based lithium concentration prediction method obtains the first variable by cleaning the initial variables (step S120).

[0085] Subsequently, the AI-based lithium concentration prediction method can preprocess the first variable using Extreme Gradient Boosting (XGBoost), a type of AI model within the Boosting family. The AI-based lithium concentration prediction method inputs the first variable into Extreme Gradient Boosting (XGBoost) (step S210). The AI-based lithium concentration prediction method identifies the individual variable importance of the first variable using permutation importance (step S220). The AI-based lithium concentration prediction method obtains the second variable by removing the inferior factors of permutation importance (step S230).

[0086] Subsequently, the AI-based lithium concentration prediction method inputs the second variable into the Limiting Gradient Boosting (XGBoost) algorithm to obtain performance feedback based on the goodness of fit (step S310). The AI-based lithium concentration prediction method identifies factors using goodness-of-fit methods such as MAPE, searching for variables with a goodness of fit greater than 90% and the fewest number of variables satisfying the condition of variable importance (step S320). The AI-based lithium concentration prediction method derives the main factors satisfying the above conditions as the fewest variables affecting the process and quantifies them (step S330).

[0087] Subsequently, the AI-based lithium concentration prediction method trains and evaluates the AI ​​model using the derived principal variables (step S410). The AI-based lithium concentration prediction method can verify the influence of the principal factors (principal variables) based on the prediction fit of the evaluation results (step S420). The influence of the principal factors can be verified through the curves in Figure 7.

[0088] Figure 7 is a graph illustrating the effect of an artificial intelligence-based lithium concentration prediction method according to an embodiment of the present invention. Figure 7 is a graph showing the change in lithium concentration during operation. Figure 7a This shows a comparison example when one main variable is applied. Figure 7b An embodiment of the invention is shown, applying all four main variables.

[0089] from Figure 7aIt can be seen that the actual waveform of the lithium concentration change curve (Real) differs from the predicted waveform (Pred). In the predicted (Pred) scenario, the lithium concentration remains constant at approximately 18 g / L, while in the actual (Real) scenario, it exhibits various variations between approximately 16 g / L and approximately 19 g / L during operation. The mean absolute percentage error (MAPE) is also 5.406, indicating that the goodness of fit is less than 95%.

[0090] On the other hand, from Figure 7b It can be seen that the actual curve waveform (Real) of lithium concentration change is similar to the predicted waveform (Pred). Specifically, the predicted (Pred) and actual (Real) lithium concentrations are similar, varying from approximately 16 g / L to approximately 19 g / L during operation. The mean absolute percentage error (MAPE) is 2.021, therefore the goodness of fit is approximately 98%, exceeding the baseline value.

[0091] Figure 8 This is a view used to describe a computing device according to an embodiment of the present invention.

[0092] Reference Figure 8 The artificial intelligence-based lithium concentration prediction method and apparatus according to the embodiments can be implemented using computing device 900.

[0093] The computing device 900 may include at least one of a processor 910, memory 930, user interface input device 940, user interface output device 950, and storage device 560 that communicate via a bus 920. The computing device 900 may further include a network interface 970 electrically connected to a network 90. ​​The network interface 970 can transmit or receive signals with other entities via the network 90.

[0094] The processor 910 can be implemented as various types such as MCU (Microcontroller Unit), AP (Application Processor), CPU (Central Processing Unit), GPU (Graphics Processing Unit), NPU (Neural Processing Unit), etc., and can be any semiconductor device that executes commands stored in memory 930 or storage device 960. The processor 910 can be configured to implement... Figure 1 The aforementioned functions and methods are related to Figure 7.

[0095] The memory 930 and storage device 960 may include various forms of volatile or non-volatile storage media. For example, the memory may include ROM (Read-Only Memory) 931 and RAM (Random Access Memory) 932. In this embodiment, the memory 930 may be located inside or outside the processor 910, and the memory 930 may be connected to the processor 910 by various known means.

[0096] In some embodiments, at least some components or functions of the AI-based lithium concentration prediction method and apparatus according to the embodiments can be implemented by a program or software running in a computing device 900, and the program or software can be stored in a computer-readable medium.

[0097] In some embodiments, at least some components or functions of the AI-based lithium concentration prediction method and apparatus according to the embodiments can be implemented using the hardware or circuitry of the computing device 900, or can be implemented using separate hardware or circuitry that can be electrically connected to the computing device 900.

[0098] Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited to the above embodiments. Various modifications and improvements made by those skilled in the art using the basic concepts of the present invention as defined in the claims also fall within the scope of the present invention.

[0099] [Explanation of reference numerals in the attached figures] 100: Artificial Intelligence-Based Lithium Concentration Prediction Device 110: Cleaning Unit 120: First Preprocessing Unit 130: Second Preprocessing Unit 140: Concentration Prediction Unit

Claims

1. An artificial intelligence-based method for predicting lithium concentration, comprising: The cleaning step involves preprocessing multiple initial variables used in the electrodialysis equipment for lithium production to obtain multiple first variables; The preprocessing step involves inputting the multiple first variables into the Boosting series artificial intelligence model for preprocessing and obtaining the final variables; as well as The prediction step involves inputting the final variable into the Boosting series of artificial intelligence models to predict the lithium concentration.

2. The lithium concentration prediction method based on artificial intelligence according to claim 1, wherein, The cleaning steps include a data cleaning step, used to remove missing values, remove data from non-operation days, and remove data for processes other than electrodialysis.

3. The lithium concentration prediction method based on artificial intelligence according to claim 2, wherein, The cleaning step further includes: detecting and removing at least one initial variable from the initial variables whose correlation with each other is above a benchmark value through correlation analysis.

4. The lithium concentration prediction method based on artificial intelligence according to claim 1, wherein, The Boosting series of artificial intelligence models are selected from a group consisting of Adaptive Boosting (AdaBoost), Random Forest, Category Feature Boosting (Catboost), Gradient Boosting Model, Lightweight Gradient Boosting Machine, and Extreme Gradient Boosting (XGBoost).

5. The lithium concentration prediction method based on artificial intelligence according to claim 1, wherein, The preprocessing step includes a first preprocessing step, used to input the first variable into a Boosting series of artificial intelligence models to calculate the variable importance of each of the first variables, and to obtain multiple second variables based on the variable importance through preprocessing. The importance of the variables is determined using permutation importance.

6. The lithium concentration prediction method based on artificial intelligence according to claim 5, wherein, The first preprocessing step includes the step of removing first variables whose permutation importance is less than a preset second benchmark.

7. The lithium concentration prediction method based on artificial intelligence according to claim 6, wherein, The preprocessing step further includes a second preprocessing step, used to input the second variable into the Boosting series artificial intelligence model to obtain multiple third variables that satisfy the prediction fit above the first benchmark with a minimum number of variables. The second preprocessing step includes calculating the prediction fit using the mean absolute percentage error.

8. The lithium concentration prediction method based on artificial intelligence according to claim 7, wherein, The second preprocessing step further includes the step of calculating the variable importance of each of the acquired third variables.

9. The lithium concentration prediction method based on artificial intelligence according to claim 8, wherein, The second preprocessing step further includes the following steps: comparing the goodness of fit when only the variable with the highest importance among the third variables is input into the Boosting series of artificial intelligence models with the predicted goodness of fit. If the prediction fit is greater, then the third variable is determined as the final variable.

10. The lithium concentration prediction method based on artificial intelligence according to claim 1, wherein, The final variables are selected from any one of the following: conductivity, circulation flow rate, and circulation pressure for any one of lithium sulfate aqueous solution, sulfuric acid, and lithium hydroxide.

11. An artificial intelligence-based lithium concentration prediction device, comprising: A cleaning unit is used to obtain multiple first variables by preprocessing multiple initial variables used in an electrodialysis device for lithium production; The preprocessing unit is used to input the plurality of first variables into the Boosting series artificial intelligence model for preprocessing and to obtain the final variables; as well as The concentration prediction unit is used to input the final variable into the Boosting series artificial intelligence model to predict the lithium concentration.

12. The artificial intelligence-based lithium concentration prediction device according to claim 11, wherein, The cleaning unit removes missing values, removes data from non-operating days, and removes data for processes other than electrodialysis.

13. The artificial intelligence-based lithium concentration prediction device according to claim 12, wherein, The cleaning unit uses correlation analysis to detect and remove at least one initial variable whose correlation with each other is above a benchmark value.

14. The artificial intelligence-based lithium concentration prediction device according to claim 11, wherein, The Boosting series of artificial intelligence models are selected from a group consisting of Adaptive Boosting (AdaBoost), Random Forest, Category Feature Boosting (Catboost), Gradient Boosting Model, Lightweight Gradient Boosting Machine, and Extreme Gradient Boosting (XGBoost).

15. The artificial intelligence-based lithium concentration prediction device according to claim 11, wherein, The preprocessing unit includes a first preprocessing unit, used to input the first variable into a Boosting series artificial intelligence model to calculate the variable importance of each of the first variables, and obtain multiple second variables based on the variable importance through the first preprocessing. The importance of the variables is determined using permutation importance.

16. The artificial intelligence-based lithium concentration prediction device according to claim 15, wherein, The first preprocessing unit removes the first variable whose arrangement importance is less than a preset second benchmark from the first variables.

17. The artificial intelligence-based lithium concentration prediction device according to claim 16, wherein, The preprocessing unit further includes a second preprocessing unit, used to input the second variable into the Boosting series artificial intelligence model to obtain multiple third variables that satisfy the prediction fit above the first benchmark with a minimum number of variables. The second preprocessing unit calculates the prediction fit using the mean absolute percentage error.

18. The artificial intelligence-based lithium concentration prediction device according to claim 17, wherein, The second preprocessing unit calculates the variable importance for each of the acquired third variables.

19. The artificial intelligence-based lithium concentration prediction device according to claim 18, wherein, The second preprocessing unit compares the goodness of fit when only the variable with the highest importance among the third variables is input into the Boosting series of artificial intelligence models with the predicted goodness of fit. If the predicted goodness of fit is greater, then the third variable is determined as the final variable.

20. The artificial intelligence-based lithium concentration prediction device according to claim 11, wherein, The final variables are selected from any one of the following: conductivity, circulation flow rate, and circulation pressure for any one of lithium sulfate aqueous solution, sulfuric acid, and lithium hydroxide.

Images