Production process of cobalt salt high-purity MO source electronic chemicals

Abstract

The invention provides a production process of cobalt salt high-purity MO source electronic chemicals, relates to the technical field of chemical production processes, and realizes automation and intelligence of the production process by optimizing links such as pretreatment, dissolution and chemical reaction, intelligent crystallization separation, drying and vacuum sublimation of raw materials. The core technology comprises an improved long short-term memory (LSTM) algorithm used for adjusting dissolution conditions in real time, a Bayesian optimization algorithm used for optimizing the crystallization process, a self-adaptive dynamic programming algorithm used for optimizing energy consumption, and a support vector machine used for intelligent anomaly detection. The digital twin technology simulates and optimizes the production process, ensures the high purity and stability of the product, optimizes the reaction conditions, crystallization process, drying and vacuum sublimation steps in each process link, and ensures the high purity of the cobalt salt product.

Description

Technical Field

[0001] The invention relates to the technical field of chemical production processes, and in particular to a cobalt salt high-purity MO source electronic chemical production process. Background Art

[0002] According to a method for reducing the content of impurity Fe in high-purity cobalt disclosed in Chinese patent publication number "CN103966627B", high-purity cobalt is prepared by selectively masking the impurity Fe in the cobalt solution after electrolytic impurity removal and purification. The high-purity cobalt sample is analyzed by GD-MS, and the purity is higher than 99.999% high-purity cobalt, and the impurity Fe element content is lower than 0.5ppm. In the process, the method of electrolytic impurity removal and iron removal by masking iron ions simplifies the high-purity cobalt preparation process, avoids the use of ion exchange and hydrolysis precipitation iron removal methods with large metal loss and complex operation, thereby reducing the production cost of high-purity cobalt; phosphate or phosphoric acid is used to mask the impurity iron ions. Compared with other masking agents, phosphate or phosphoric acid can form a highly stable complex with iron ions within the solution pH range required by cobalt electrolysis, thereby increasing the precipitation potential of impurity iron and facilitating the precipitation of the main metal cobalt.

[0003] According to a method for preparing a high-purity cobalt solution by stirring extraction disclosed in Chinese patent publication number "CN105154673B", a cobalt solution is prepared in a high-level extraction tank; sulfonated kerosene and P507 extractant are added to a low-level extraction tank, and sodium hydroxide is added for stirring and saponification, and then the solution is pumped into a high-level extraction tank for stirring and extraction, and the cobalt in the cobalt solution is extracted into an organic phase; hydrochloric acid is added to stir and wash the loaded organic phase, and back-extract to prepare a cobalt-rich solution; a macroporous adsorption resin is added for deep purification, and after concentration, electrolytic deposition is performed, and the electrodeposition of cobalt is vacuum smelted to obtain a high-purity cobalt ingot, and the purity of cobalt is above 5N8 for more than 70 elements according to glow discharge mass spectrometry analysis.

[0004] Cobalt salt is one of the important raw materials for the production of MO source electronic chemicals. MO source is widely used in electronic materials, and its purity and quality directly determine the performance of downstream products. However, in the traditional cobalt salt production process, due to the lack of intelligent data feedback and regulation mechanism, the product purity is unstable and the production efficiency is low, which cannot meet the efficient and intelligent production needs in modern industry. Summary of the invention

[0005] Technical issues solved

[0006] In view of the shortcomings of the prior art, the present invention provides a cobalt salt high-purity MO source electronic chemical production process, which solves the following problems:

[0007] In the traditional cobalt salt production process, due to the lack of intelligent data feedback and control mechanisms, the product purity is unstable and the production efficiency is low, which cannot adapt to the efficient and intelligent production needs in modern industry.

[0008] Technical Solution

[0009] To achieve the above objectives, the present invention is implemented by the following technical scheme: a cobalt salt high-purity MO source electronic chemical production process, the process comprising the following steps:

[0010] Sp1. Raw material pretreatment: purify, dry and screen the high-purity cobalt salt raw materials to ensure that the purity of the raw materials meets the predetermined requirements;

[0011] Sp2, dissolution and chemical reaction: the pretreated cobalt salt raw material is added to the reactor and dissolved under controllable temperature and pressure conditions. During the dissolution process, the pH value, temperature and reaction rate of the reaction solution are monitored in real time through the reaction parameter monitoring module, and the improved long short-term memory algorithm is used to predict the optimal reaction conditions based on historical process data and dynamically adjust the dissolution conditions; the specific formula of the improved long short-term memory algorithm is as follows:

[0012]

[0013] h t : hidden state at time t; x t : Input multidimensional data, including temperature, pressure, reactant concentration and other parameters; W x ,W h ,W c : Weight matrix, representing the connection between different input and hidden layers; i t : Input gate, which controls the impact of new input on the state; c t : Cell state, recording long-term memory information; f: Activation function, here we choose the adaptively adjusted ReLU function; b h ,b c : bias term;

[0014] Sp3, intelligent crystallization separation: After the dissolution is completed, the reaction mixture is transferred to an intelligent crystallization separation system, which collects multidimensional data of cooling rate, stirring speed and solvent amount, analyzes the multidimensional data in combination with a Bayesian optimization algorithm, optimizes crystallization conditions, dynamically adjusts crystallization parameters, and generates high-purity cobalt salt crystals; the objective function of the Bayesian optimization algorithm is:

[0015]

[0016] μ(x): predicted mean under input condition x, representing the reaction output under current conditions; σ(x): predicted variance under input condition x, representing uncertainty; κ: exploration parameter, controlling the balance between mean and variance; x: process parameter;

[0017] Sp4, drying and vacuum sublimation: drying the high-purity cobalt salt crystals, the drying process is monitored by a data verification adjustment system, and the drying temperature and time are adjusted in real time; vacuum sublimation is performed after drying, and during the vacuum sublimation process, an adaptive dynamic programming algorithm is used to optimize the sublimation parameters according to energy consumption and output effect. The sublimation parameters are as follows:

[0018] Sublimation temperature (T): should be set according to the phase transition temperature of the cobalt salt to ensure effective sublimation while preventing thermal decomposition.

[0019] Vacuum degree (P): controls the pressure inside the reactor and affects the sublimation rate.

[0020] Time (t): Duration of the sublimation process, which needs to be dynamically adjusted to achieve the desired purity and yield.

[0021] Condensation temperature (T cloud ): The condensation temperature of the collected vapor affects the collection efficiency of the product.

[0022] Gas flow rate (F): During the sublimation process, the flow rate of the gas will affect the mass transfer efficiency;

[0023] The adaptive dynamic programming algorithm dynamically adjusts the control parameters by defining an optimization objective function to minimize energy consumption and maximize output effect. The control parameters are: x = [T, P, t, T cloud ,F], and define the optimization objective function as:

[0024] J(x)=α·E(x)-β·P yield (x)

[0025] E(x): A function representing energy consumption, which depends on the control parameters x, P yield (x): A function representing the output effect (yield of high-purity cobalt salt), which depends on the control parameter x. α and β are weight coefficients, both of which have values ​​between 0 and 1, and are used to balance the impact of energy consumption and output effect.

[0026] The state transfer equation is:

[0027]

[0028] V(x,t): the state value at time t, x t : control parameter selected at time t, γ: discount factor used to evaluate the impact of future states;

[0029] Sp5. Virtual production optimization based on digital twins: Before adjusting the actual production process, a virtual production line model is constructed through digital twin technology to simulate and analyze the impact of different process parameters on the purity and physical and chemical properties of cobalt salts, and the parameter settings are optimized in combination with the random forest regression algorithm to ensure the accuracy of the final production process;

[0030] Sp6, intelligent anomaly detection and adaptive adjustment: use deep learning algorithms to perform real-time analysis on multi-dimensional data collected during the production process, and combine support vector machines and anomaly detection algorithms to predict abnormal situations that may occur in the process; when an anomaly is detected, automatically adjust production parameters to avoid the occurrence of anomalies, ensuring the continuity and stability of the production process; the formula of the support vector machine anomaly detection algorithm: given a training data set: where x i is the multidimensional process parameter, y i For the corresponding category (normal or abnormal), the decision boundary formula of the support vector machine is:

[0031]

[0032] α i : Lagrange multiplier; K(x,x i ): kernel function, used to map data to high-dimensional space; b: bias anomaly detection algorithm sets a threshold τ, and when f(x)<τ, it is judged as abnormal;

[0033] Sp7, Data verification and feedback control: Multi-point sensors are used to collect process parameters throughout the entire production process, including temperature, pressure, material microstructure changes, and reaction kinetics. Adaptive dynamic programming and Bayesian optimization algorithms are used for real-time data analysis, and feedback is used to adjust various parameters in the production process to ensure the high purity and stability of the final product.

[0034] Preferably, in the raw material pretreatment step, purification includes removing impurity ions in the raw materials, and the drying process adopts a vacuum drying method to control the humidity of the raw materials below 10 ppm, and the particle size of the cobalt salt is controlled to 10-30 microns through a screening system.

[0035] Preferably, in the dissolution and chemical reaction steps, the material of the reactor is corrosion-resistant alloy steel, the temperature of the dissolution reaction is controlled at room temperature -200°C, the pressure is controlled at 1.5-2.5MPa, the pH value during the reaction is predicted by the improved LSTM model, and the amount of acid and alkali agent added is adjusted according to real-time feedback.

[0036] Preferably, in the intelligent crystallization separation step, the cooling rate of the crystallization system is controlled at 0.5-1°C / minute, the stirring speed is controlled at 200-300rpm, and the cooling rate and the amount of solvent are dynamically adjusted according to the multidimensional data.

[0037] Preferably, in the drying and vacuum sublimation steps, the temperature of the drying process is controlled at 50-100°C, and the vacuum sublimation temperature is controlled at 200-600°C.

[0038] Preferably, in the virtual production optimization step based on digital twins, the digital twin model combines the real-time data of the production line to perform parameter optimization, and uses the random forest regression algorithm to analyze the impact of process parameters such as cooling rate and reaction time on product purity, so as to predict potential problems in production in advance.

[0039] Preferably, in the intelligent anomaly detection and adaptive adjustment step, the anomaly detection algorithm monitors the fluctuations of temperature, pressure, and solubility parameters, and uses a support vector machine model to automatically identify process anomalies that may cause a decrease in product purity, and triggers an adaptive adjustment module when an anomaly is detected to automatically optimize the parameters.

[0040] Preferably, in the data verification and feedback control step, the data acquisition system monitors the process parameters in real time through multi-point sensors installed in the reaction kettle, crystallization equipment, drying and vacuum sublimation equipment.

[0041] Preferably, the system for producing a cobalt salt high-purity MO source electronic chemical product comprises the following modules:

[0042] Raw material pretreatment module, used to purify, dry and screen high-purity cobalt salt raw materials to ensure the purity of raw materials;

[0043] The dissolution and chemical reaction module includes a reactor, a temperature and pressure control device, and an improved long-short-term memory algorithm unit to monitor and adjust the dissolution conditions in real time;

[0044] Intelligent crystallization separation module, including cooling control unit, stirring control unit and Bayesian optimization algorithm unit, used to adjust crystallization conditions according to multi-dimensional data;

[0045] A drying and vacuum sublimation module, including a drying device and an adaptive dynamic programming algorithm unit, for controlling the drying temperature and sublimation parameters;

[0046] Digital twin virtual production optimization module, which simulates the actual production process through digital twin technology and combines the random forest regression algorithm to optimize parameters;

[0047] Intelligent anomaly detection and adjustment module, combining deep learning and support vector machine algorithms, analyzes multi-dimensional data in real time and detects production anomalies, automatically adjusting process parameters;

[0048] The data verification and feedback control module collects process data during the production process through multi-point sensors, and combines Bayesian optimization and adaptive dynamic programming algorithms to perform real-time feedback control.

[0049] Preferably, the digital twin virtual production optimization module includes the following parts: State equation: Digital twin technology simulates the dynamic evolution of physical processes through differential equations, and the state equation describes the change of system state over time:

[0050]

[0051] x(t) represents the state variable vector of the production process, including temperature, pressure, reaction rate, and cooling rate; u(t) represents the control input vector of the system, including adjustable parameters such as cooling rate, stirring rate, and energy input; t is time; f(x,u,t) represents the nonlinear function of the physical process, which is modeled based on the physical characteristics of the system; output equation: The output of the system is some observable variables (including product purity and energy consumption), expressed as:

[0052] y(t)=g(x(t),u(t),t)

[0053] y(t) is the system output, which represents the quality characteristics of the product; g(x,u,t) is the output function, which is usually a linear or nonlinear function, depending on the physical relationship in the actual process; the regression model of the random forest regression algorithm: random forest regression uses multiple decision tree models h m (x) Make predictions:

[0054]

[0055] represents the predicted value, such as the purity of saw salt crystals in the production process; M is the number of decision trees in the random forest; h m (x) represents the regression prediction of the input feature x by the mth decision tree model; the decision tree splitting rule is as follows: each decision tree uses a binary split method to select a feature x at each node j. i and its threshold t j , with minimizing the mean square error (MSE) as the objective function:

[0056]

[0057] y i is the true value of the i-th sample; n is the number of samples in the current node; t jIt is the optimal feature threshold for splitting, which is selected by minimizing MSE; Feature importance analysis: The random forest algorithm can evaluate the impact of each production parameter on the final output result. The importance of the feature is evaluated by calculating the contribution of each feature's split point in the decision tree to the MSE:

[0058]

[0059] I(f i ) is the feature f i The importance of is the feature f i The resulting MSE reduction; -M is the total number of decision trees.

[0060] Beneficial Effects

[0061] The present invention provides a cobalt salt high-purity MO source electronic chemical production process. It has the following beneficial effects:

[0062] 1. The present invention ensures the high purity of cobalt salt products by optimizing the reaction conditions, crystallization process, drying and vacuum sublimation steps in each process link. In particular, through the application of intelligent algorithms such as LSTM algorithm, Bayesian optimization and adaptive dynamic programming, the purity and quality consistency of the product are significantly improved, the generation of impurities is reduced, and the problems of unstable product purity and large batch quality differences in the prior art are solved.

[0063] 2. The present invention effectively optimizes energy consumption by intelligently controlling parameters such as heating, cooling rate, and stirring speed, especially in the drying and vacuum sublimation stages, and reduces the energy consumption per unit product through adaptive dynamic programming. Compared with the large amount of energy waste in traditional processes, this solution achieves energy conservation and consumption reduction, greatly reduces production costs, and solves the problem of excessive energy consumption in the prior art.

[0064] 3. Through real-time data feedback and algorithm optimization, the present invention significantly shortens all links in the production process (such as reaction speed, crystallization time, drying and sublimation parameters). In particular, in the reaction and crystallization process, the optimized cooling rate and stirring speed greatly improve the process efficiency. This solution solves the problem of long production cycle and low efficiency in the prior art.

[0065] 4. The present invention realizes real-time monitoring and adaptive adjustment of the production process by introducing multiple intelligent algorithms, ensuring the intelligence and controllability of the process and reducing the necessity of human intervention. This solves the drawbacks of the existing technology of inaccurate process control and reliance on manual adjustment, and improves the automation level of the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1A microstructure comparison diagram between the finished product of the present invention and the finished product of the prior art;

[0067] Figure 2 It is a product quality curve diagram during the drying and vacuum sublimation process of the present invention;

[0068] Figure 3 It is a curve diagram of energy consumption in the drying and vacuum sublimation process of the present invention;

[0069] Figure 4 is a bar graph of the best crystallization parameters of the present invention;

[0070] Figure 5 It is a temperature change curve diagram of the reaction process of the present invention;

[0071] Figure 6 It is a curve diagram of pressure change during the reaction process of the present invention;

[0072] Figure 7 It is a curve diagram of pH value change during the reaction process of the present invention. DETAILED DESCRIPTION

[0073] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Specific embodiment one:

[0075] As shown in the figure, a cobalt salt high-purity MO source electronic chemical production process includes the following steps:

[0076] Sp1. Raw material pretreatment: Purify, dry and screen the high-purity cobalt salt raw materials to ensure that the purity of the raw materials meets the predetermined requirements. Purification includes removing impurity ions in the raw materials. The drying process adopts vacuum drying method to control the humidity of the raw materials below 10ppm, and the particle size of the cobalt salt is controlled within 10-30 microns through the screening system;

[0077] Sp2, dissolution and chemical reaction: the pretreated cobalt salt raw materials are added to the reactor and dissolved under controllable temperature and pressure conditions. The material of the reactor is corrosion-resistant alloy steel. The temperature of the dissolution reaction is controlled at room temperature -200°C, and the pressure is controlled at 1.5-2.5MPa. The pH value during the reaction process is predicted by the improved LSTM model, and the amount of acid and alkali agent added is adjusted according to real-time feedback. During the dissolution process, the pH value, temperature, and reaction rate of the reaction solution are monitored in real time through the reaction parameter monitoring module. The improved long short-term memory algorithm is used to predict the optimal reaction conditions based on historical process data and dynamically adjust the dissolution conditions. The specific formula of the improved long short-term memory algorithm is as follows:

[0078]

[0079] h t : hidden state at time t; x t : Input multidimensional data, including temperature, pressure, reactant concentration and other parameters; W x ,W h ,W c : Weight matrix, representing the connection between different input and hidden layers; i t : Input gate, which controls the impact of new input on the state; c t : Cell state, recording long-term memory information; f: Activation function, here we choose the adaptively adjusted ReLU function; b h ,b c : bias term;

[0080] Sp3, intelligent crystallization separation: after the dissolution is completed, the reaction mixture is transferred to the intelligent crystallization separation system, the cooling rate of the crystallization system is controlled at 0.5-1 ° C / minute, the stirring speed is controlled at 200-300rpm, and the cooling rate and solvent amount are dynamically adjusted according to the multidimensional data. The crystallization separation system collects multidimensional data of cooling rate, stirring speed and solvent amount, analyzes the multidimensional data in combination with the Bayesian optimization algorithm, optimizes the crystallization conditions, dynamically adjusts the crystallization parameters, and generates high-purity cobalt salt crystals; the objective function of the Bayesian optimization algorithm is:

[0081]

[0082] μ(x): predicted mean under input condition x, representing the reaction output under current conditions; σ(x): predicted variance under input condition x, representing uncertainty; κ: exploration parameter, controlling the balance between mean and variance; x: process parameter;

[0083] Sp4, drying and vacuum sublimation: drying the high-purity cobalt salt crystals, the drying process is monitored by a data verification adjustment system, and the drying temperature and time are adjusted in real time; vacuum sublimation is performed after drying, and during the vacuum sublimation process, an adaptive dynamic programming algorithm is used to optimize the sublimation parameters according to energy consumption and output effect. The sublimation parameters are as follows: sublimation temperature (T), vacuum degree (P), time (t), condensation temperature (T cloud ), gas flow (F);

[0084] The adaptive dynamic programming algorithm dynamically adjusts the control parameters by defining an optimization objective function to minimize energy consumption and maximize output effect. The control parameters are: x = [T, P, t, T cloud ,F], and define the optimization objective function as:

[0085] J(x)=α·E(x)-β·P yield (x)

[0086] E(x): A function representing energy consumption, which depends on the control parameters x, P yield (x): A function representing the output effect (yield of high-purity cobalt salt), which depends on the control parameter x. α and β are weight coefficients, both of which have values ​​between 0 and 1, and are used to balance the impact of energy consumption and output effect.

[0087] The state transfer equation is:

[0088]

[0089] V(x,t): the state value at time t, x t : control parameter selected at time t, γ: discount factor used to evaluate the impact of future states;

[0090] Sp5. Virtual production optimization based on digital twins: Before adjusting the actual production process, a virtual production line model is constructed through digital twin technology to simulate and analyze the impact of different process parameters on the purity and physical and chemical properties of cobalt salts, and the parameter settings are optimized in combination with the random forest regression algorithm to ensure the accuracy of the final production process. The digital twin model optimizes parameters in combination with the real-time data of the production line, and uses the random forest regression algorithm to analyze the impact of process parameters such as cooling rate and reaction time on product purity, and predict potential problems in production in advance;

[0091] The digital twin virtual production optimization module includes the following parts: State equation: Digital twin technology simulates the dynamic evolution of physical processes through differential equations. The state equation describes the change of system state over time:

[0092]

[0093] x(t) represents the state variable vector of the production process, including temperature, pressure, reaction rate, and cooling rate; u(t) represents the control input vector of the system, including adjustable parameters such as cooling rate, stirring rate, and energy input; t is time; f(x,u,t) represents the nonlinear function of the physical process, which is modeled based on the physical characteristics of the system; output equation: The output of the system is some observable variables (including product purity and energy consumption), expressed as:

[0094] y(t)=g(x(t),u(t),t)

[0095] y(t) is the system output, which represents the quality characteristics of the product; g(x,u,t) is the output function, which is usually a linear or nonlinear function, depending on the physical relationship in the actual process; the regression model of the random forest regression algorithm: random forest regression uses multiple decision tree models h m (x) Make predictions:

[0096]

[0097] represents the predicted value, such as the purity of saw salt crystals in the production process; M is the number of decision trees in the random forest; h m (x) represents the regression prediction of the input feature x by the mth decision tree model; the decision tree splitting rule is as follows: each decision tree uses a binary split method to select a feature x at each node j. i and its threshold t j , with minimizing the mean square error (MSE) as the objective function:

[0098]

[0099] y i is the true value of the i-th sample; n is the number of samples in the current node; t j It is the optimal feature threshold for splitting, which is selected by minimizing MSE; Feature importance analysis: The random forest algorithm can evaluate the impact of each production parameter on the final output result. The importance of the feature is evaluated by calculating the contribution of each feature's split point in the decision tree to the MSE:

[0100]

[0101] I(f i ) is the feature f i The importance of is the feature f i The amount of MSE reduction caused; -M is the total number of decision trees;

[0102] Sp6, intelligent anomaly detection and adaptive adjustment: use deep learning algorithms to perform real-time analysis on multi-dimensional data collected during the production process, and combine support vector machines and anomaly detection algorithms to predict abnormal situations that may occur in the process; the anomaly detection algorithm monitors temperature, pressure, and solubility parameter fluctuations, and uses the support vector machine model to automatically identify process anomalies that may cause product purity to decrease, and trigger the adaptive adjustment module when an anomaly is detected to automatically optimize the parameters. When an anomaly is detected, the production parameters are automatically adjusted to avoid the anomaly and ensure the continuity and stability of the production process; the formula of the support vector machine anomaly detection algorithm: given a training data set: where x i is the multidimensional process parameter, y i For the corresponding category (normal or abnormal), the decision boundary formula of the support vector machine is:

[0103]

[0104] α i : Lagrange multiplier; K(x,x i ): kernel function, used to map data to high-dimensional space; b: bias anomaly detection algorithm sets a threshold τ, and when f(x)<τ, it is judged as abnormal;

[0105] Sp7. Data verification and feedback control: Multi-point sensors are used to collect process parameters throughout the entire production process, including temperature, pressure, material microstructure changes, and reaction kinetics. Adaptive dynamic programming and Bayesian optimization algorithms are used for real-time data analysis, and feedback is used to adjust various parameters in the production process to ensure the high purity and stability of the final product. The data acquisition system monitors process parameters in real time through multi-point sensors installed in reactors, crystallization equipment, drying and vacuum sublimation equipment.

[0106] LSTM algorithm: Input data includes temperature, pressure, pH value, reactant concentration, and historical reaction data monitored in real time during the reaction process. When training the model, sequence data is used, and the data is divided into training set and test set. The mean square error (MSE) is used as the loss function for optimization. After the output is calculated through forward propagation, the gradient is calculated through the chain rule in the back propagation, and the weight matrix is ​​adjusted. The optimization process uses the Adam optimizer to increase the convergence speed. Finally, the hyperparameters (such as learning rate, batch size, etc.) are adjusted to achieve a better model output.

[0107] High-purity cobalt salt raw materials are processed through three sub-steps: purification, drying, and screening. During the purification process, ion exchange or membrane filtration technology is used to remove impurity ions to ensure that the purity of the raw materials meets the required standards. The drying process uses a vacuum drying method to keep the humidity below 10ppm to prevent moisture from affecting subsequent reactions. Ion exchange or membrane filtration technology is usually used to purify and remove impurity ions from raw materials. These technologies effectively remove unwanted ions through selective adsorption or filtration to ensure the high purity of raw materials, thereby improving the efficiency of subsequent reactions and product quality. Drying uses a vacuum drying method to effectively remove moisture from raw materials and control the humidity below 10ppm. The vacuum environment lowers the boiling point of water, allowing moisture to evaporate at a lower temperature, reducing heat damage to raw materials, while avoiding the negative impact of moisture on the reaction. The particle size of cobalt salt is controlled at 10-30 microns, which helps to ensure the uniformity and solubility of raw materials during the reaction. The material of the reactor is corrosion-resistant alloy steel that can withstand high temperature and high pressure environment, avoiding corrosion of the material and damage to the reactor caused by chemical reactions during the reaction. Temperature control at room temperature -200℃ and pressure at 1.5-2.5MPa helps to enhance the solubility and reaction rate of cobalt salt. This condition can promote full contact of reactants and improve the yield of products. The improved long short-term memory (LSTM) model is used to monitor the changes in pH value during the reaction in real time. By training the LSTM model with historical data, the amount of acid and alkali agent added can be optimized, thereby dynamically adjusting the reaction conditions to ensure that the reaction is carried out within the optimal pH range and improve the purity of the product. The uniformity of particle size can reduce local overheating or overcooling during the reaction and improve the reaction rate and efficiency. The screening system controls the particle size of cobalt salt at 10-30 microns to ensure its uniformity in dissolution and reaction, thereby improving the reaction efficiency. The pretreated cobalt salt raw materials are added to the reactor, and the material of the reactor is corrosion-resistant alloy steel to adapt to high temperature and high pressure environments. This step controls the temperature at room temperature -200℃ and the pressure at 1.5-2.5MPa. By real-time monitoring of pH value, temperature and reaction rate, an improved long short-term memory (LSTM) algorithm is used to predict the reaction conditions. The LSTM model is trained using historical data, and its core formula is continuously optimized through a feedback mechanism to ensure that the reaction process is carried out under optimal conditions. The key control node of this part is the dynamic adjustment of the pH value to ensure the integrity of the reaction and the purity of the product. The dissolved reaction mixture enters the crystallization separation system to control the cooling rate, stirring speed and solvent amount. Controlling the cooling rate at 0.5-1℃ / min and the stirring speed at 200-300rpm can effectively promote the formation of cobalt salt crystals and affect the particle size and morphology of the crystals. A slower cooling rate helps to form large and uniform crystals. Based on real-time analysis of multidimensional data, the cooling rate and solvent amount can be dynamically adjusted to adapt to changes in reaction conditions.This flexibility can further optimize the crystallization process and improve the quality and purity of the final product. During this process, the Bayesian optimization algorithm analyzes multidimensional data in real time to determine the best crystallization conditions. The Bayesian optimization algorithm uses historical data to build a model and dynamically adjust various parameters to ensure the generation of high-purity cobalt salt crystals. The key control points are the cooling rate and stirring speed, which directly affect the particle size and morphology of the crystals. The high-purity cobalt salt crystals are dried and vacuum sublimated. The temperature of the drying process is controlled at 50-100°C to ensure moisture removal.

[0108] Adaptive dynamic programming algorithm: Dynamically adjust sublimation parameters by monitoring energy consumption and output quality. This algorithm can combine real-time data to optimize the vacuum sublimation process, improve efficiency, and ensure that the impurity content of the final product is less than 50ppb. The adaptive dynamic programming algorithm optimizes sublimation parameters by monitoring energy consumption and output effects in real time. The core of this step is dynamic adjustment to maximize output quality and reduce energy consumption. The virtual production optimization module based on digital twins simulates the actual production process by building a virtual production line model. Digital twin technology uses differential equations to establish state equations to describe the change of system state over time, and combines random forest regression algorithms to optimize production parameters. Model outputs include product purity and energy consumption, which are dynamically adjusted through real-time data feedback. The key control nodes are cooling speed and reaction time, which affect the purity and stability of the final product. During the production process, data is analyzed in real time by deep learning algorithms, and support vector machines are used for anomaly detection. The algorithm establishes decision boundaries based on training data sets and determines whether there are anomalies by setting thresholds. When anomalies are detected, the system automatically adjusts production parameters to avoid quality degradation. The key control point is the monitoring of temperature, pressure and solubility to ensure the continuity of the production process. The data acquisition system monitors the process parameters in real time through multi-point sensors, including temperature, pressure, material microstructure changes, etc. The system combines Bayesian optimization and adaptive dynamic programming algorithms to analyze the data and perform feedback control to ensure the high purity and stability of the product. The key control node is the real-time adjustment of process parameters to adapt to changes in production.

[0109] Bayesian optimization algorithm: In the process, Bayesian optimization is used to optimize the cooling rate and stirring speed. First, a probability model is established, usually using Gaussian process regression, combined with historical data for training. After each experiment, the model updates and calculates the current optimal solution, and selects the next set of process parameters for experimentation by maximizing the expected improvement (EI). This iteration gradually finds the optimal process conditions that can improve product purity and reaction efficiency.

[0110] Adaptive dynamic programming algorithm: This algorithm is used to dynamically adjust the time and temperature of drying and vacuum sublimation in the process. By constructing a state space model, the current state (such as energy consumption, output, product quality) is evaluated with the future state, and the current control input is optimized based on the update of the value function. The algorithm uses reinforcement learning to adjust parameters based on reward signals (such as increased output and reduced energy consumption) to ensure the optimal operation of the production process.

[0111] Support vector machine algorithm: The support vector machine constructs the boundary through the training set and selects the maximum interval hyperplane as the decision boundary. The decision logic of the boundary is based on the selection of the support vector, that is, the sample closest to the boundary is processed. The training process uses the Lagrange multiplier method to optimize the interval to ensure classification accuracy. When it is detected that the sample exceeds the boundary, the system triggers the abnormal adjustment mechanism and automatically adjusts the process parameters (such as temperature and pressure) to restore to the normal production state to ensure stable product quality. In this way, timely response and adjustment to abnormalities in the production process can be achieved. Specific embodiment 2:

[0113] As shown in the figure, a system for the production process of cobalt salt high-purity MO source electronic chemicals includes the following modules:

[0114] Raw material pretreatment module, used to purify, dry and screen high-purity cobalt salt raw materials to ensure the purity of raw materials;

[0115] The dissolution and chemical reaction module includes a reactor, a temperature and pressure control device, and an improved long-short-term memory algorithm unit to monitor and adjust the dissolution conditions in real time;

[0116] Intelligent crystallization separation module, including cooling control unit, stirring control unit and Bayesian optimization algorithm unit, used to adjust crystallization conditions according to multi-dimensional data;

[0117] The drying and vacuum sublimation module includes a drying device and a vacuum pump, which are used to reduce the pressure in the reactor to achieve a vacuum state; a sublimation reactor, which contains high-purity cobalt salt to be sublimated and provides an environment for controlling temperature and pressure; a condenser, which collects the vapor generated during the sublimation process and cools it into a solid to increase the yield; a temperature control system, which ensures the temperature in the reactor is stable and optimizes the sublimation effect; a pressure monitoring instrument, which monitors the vacuum degree in real time for timely adjustment; and a data acquisition system, which is used to record and analyze the parameters in the sublimation process. These devices work together to achieve an efficient and controllable vacuum sublimation process, and an adaptive dynamic programming algorithm unit, which is used to control the drying temperature and sublimation parameters;

[0118] Digital twin virtual production optimization module, which simulates the actual production process through digital twin technology and combines the random forest regression algorithm to optimize parameters;

[0119] Intelligent anomaly detection and adjustment module, combining deep learning and support vector machine algorithms, analyzes multi-dimensional data in real time and detects production anomalies, automatically adjusting process parameters;

[0120] The data verification and feedback control module collects process data during the production process through multi-point sensors, and combines Bayesian optimization and adaptive dynamic programming algorithms to perform real-time feedback control.

[0121] The raw material pretreatment module includes ion exchange equipment, vacuum drying equipment, laser particle size analyzer, and screening equipment. First, the cobalt salt raw materials are removed of impurities through the ion exchange equipment. The ion exchange equipment is connected to multiple purification filter elements by a fluid pipeline system. The filter element material selects high-efficiency resin to ensure the removal of water-soluble impurities. The treated raw materials enter the vacuum drying equipment. The drying equipment includes a vacuum pump, a heating plate, and a temperature sensor. The drying temperature and vacuum degree are accurately controlled by a PID controller. After that, a laser particle size analyzer is used to detect the size of the material particles and the particle size is separated by a screening device. The screening device is a vibrating screen, and its motor drives the screen to screen particles of different particle sizes to ensure that the cobalt salt particle size is within the range of 10-30 microns.

[0122] Dissolution and chemical reaction module reactor, temperature and pressure control system, pH sensor, solution agitator, acid and alkali injection system, LSTM algorithm embedded controller. The reactor is made of corrosion-resistant alloy steel and is equipped with temperature sensor and pressure sensor for real-time detection of the reaction environment. The LSTM algorithm runs in the embedded controller, receives temperature, pressure, pH value and reaction rate data, and controls the injection of acid and alkali. The agitator adjusts the stirring speed through the frequency converter to adapt to different stages in the reaction process. The LSTM algorithm optimizes the reaction efficiency and product quality by adjusting the reaction conditions in real time. The acid and alkali injection system controls the precise injection amount of the acid and alkali through an electric pump, and the LSTM algorithm dynamically adjusts the injection rate based on real-time feedback to ensure that the dissolution reaction is carried out under optimal conditions.

[0123] Intelligent crystallization separation module cooling control system, stirring control system, solvent circulation system, crystallization tank, multi-dimensional data acquisition system, Bayesian optimization algorithm processor. The cooling control system includes a heat exchanger and a cooling pipe. The heat exchanger adjusts the flow rate and temperature of the cooling water according to real-time data through a PLC (programmable logic controller). The stirring control system is driven by a variable frequency motor, and the stirring speed is collected by a sensor and dynamically adjusted according to the Bayesian optimization algorithm. The solvent circulation system circulates the solution from the crystallization tank back to the cooling system through a pump to achieve temperature control and uniform crystallization of the solution. The multi-dimensional data acquisition system collects the cooling rate, stirring speed, solution viscosity and temperature in real time. The data is analyzed by the Bayesian optimization algorithm processor to optimize the crystallization conditions. The optimization algorithm adjusts the cooling and stirring parameters to keep the crystallization process under optimal conditions and ensure the crystallization of high-purity cobalt salts.

[0124] Drying and vacuum sublimation module vacuum drying equipment, energy management system, temperature sensor, adaptive dynamic programming algorithm control unit. The drying equipment includes a vacuum chamber and a heating element. The temperature sensor monitors the temperature in the chamber, and the energy management system maintains the set temperature by controlling the power of the heating element.

[0125] Digital twin-based virtual production optimization module digital twin server, high-precision data collector, simulation environment generator, random forest algorithm processor, virtual process control terminal. The digital twin server obtains real-time data from various production equipment through the high-precision data collector, and builds a virtual production line model through the simulation environment generator. The random forest algorithm processor runs in the server, processes parameters such as temperature, pressure, cooling rate, etc. collected from various devices, analyzes their impact on product purity, and optimizes parameter combinations. The virtual process control terminal is responsible for adjusting the actual process conditions according to the optimized parameters, and simulating the impact of different process settings on the purity of cobalt salts in the digital twin model. Virtual process control is synchronized with the actual production process in real time to ensure that potential problems are discovered and adjusted in advance in actual production.

[0126] Intelligent anomaly detection and adaptive adjustment module Deep learning algorithm server, real-time monitoring system (temperature, pressure, solution sensor), support vector machine (SVM) processor, automatic adjustment system. The deep learning algorithm server collects key parameters in the production process through the real-time monitoring system, and the support vector machine algorithm running in the server builds a decision boundary based on historical data and current data. When certain parameters in production (such as temperature, pressure or solution concentration) exceed the decision boundary, the SVM processor triggers an alarm and makes adaptive adjustments through the automatic adjustment system. The automatic adjustment system adjusts production conditions in a timely manner to ensure process stability by controlling equipment such as valves, heating elements or cooling systems. The SVM algorithm dynamically updates its decision boundary to adapt to different production batches and process changes to achieve continuous process optimization.

[0127] Data verification and feedback control module multi-point sensor network, data analysis server, feedback control unit, Bayesian optimization algorithm processor, adaptive dynamic programming control unit. The multi-point sensor network is installed in the reactor, crystallization equipment, drying and vacuum sublimation equipment. The sensors collect process parameters such as temperature, pressure, microstructure, and transmit the data to the data analysis server. The Bayesian optimization algorithm and adaptive dynamic programming control unit running in the server analyze the data and provide real-time feedback to adjust the production conditions. The feedback control unit controls key equipment such as heating, cooling, and stirring through the PLC system to ensure that the process parameters remain within the optimal range. The data verification system can also compare the actual data with the process standards in real time. If deviations are found, they will be adjusted immediately, and the operation of each step will be optimized through the adaptive dynamic programming controller.

[0128] According to the attached Figure 1 The 3D graph shows the difference in the microstructure of the finished product between the prior art and the improved technology. The microstructure of the prior art is more random and uneven, while the microstructure of the improved technology is more uniform and optimized. The iso-section graph shows the uniformity of the internal structure of the finished product, reflecting the advantages of the optimized process. Specific embodiment three:

[0130] A system for the production process of cobalt salt high-purity MO source electronic chemicals. The production line using this design solution includes the following three product process series:

[0131] The steps for the organic cobalt salt are as follows:

[0132] Feeding and synthesis reaction: The weighed cobalt sulfate heptahydrate is slowly added to the reactor by manual feeding, water is added and stirred to dissolve, and then the liquid alkali (32%) is quantitatively measured by a flow meter and the feeding speed is controlled by a pneumatic valve, and slowly added to the reactor with condensation reflux through a closed pipeline to react at room temperature and pressure. After the reaction is completed, wait for the system temperature to cool naturally to below 50°C, and the measured mixed organic acid, mixed ab diketone, and mixed keto acid are transported to the reactor through a pneumatic diaphragm pump through a pipeline in proportion and batches, and a promoter is added to cause a synthesis reaction. The system reacts for 8-12 hours, the conversion rate is greater than or equal to 99%, and the yield is 98%. G1-1 feeding waste gas and G1-2 reaction waste gas will be generated.

[0133] Efficient separation: After the synthesis reaction is completed, the obtained solid-liquid mixture is pumped into a cooling tank through a pipeline pump to cool to 20-40°C, and then pumped into a high-efficiency diaphragm filter press through a pipeline pump for solid-liquid separation. The solid phase is a crude organic cobalt salt product, which is collected by an automatic conveyor belt, and the liquid phase is a sodium sulfate solution, which is pumped into a precipitation tank through a pipeline pump. This process takes 0.5 hours. No material will be entrained under normal working conditions.

[0134] Drying finished product: After separation, the solid is transported to the feed port of the flash dryer by an automatic conveyor belt, and the separated solid is subjected to rotary flash drying at 160-200°C. This process takes 1 hour, and the steam is collected by a spray device, which will produce G1-3 drying waste gas.

[0135] Material packaging: The powder mixing tank mixes products produced in different batches at room temperature to ensure uniform and stable product quality. The dried materials are conveyed into the crusher through a conveyor belt for crushing and packaging. The working time is 1 hour, which will produce G1-4 crushing and packaging waste gas.

[0136] Multiple-effect evaporation: After separation, the sodium sulfate solution is pumped into a precipitation tank through a pipeline, and an external precipitant is added to precipitate cobalt ions. The solution is then pumped into a high-efficiency diaphragm filter press through a pipeline for solid-liquid separation to obtain G2-1 filter residue. The separated liquid is subjected to multiple-effect evaporation at 95-135°C using a multiple-effect evaporator to obtain by-products (including but not limited to sodium sulfate). The steam is collected using a spray device to generate G2-2 evaporation waste gas.

[0137] Material packaging: The powder mixing tank mixes products from different batches at room temperature to ensure uniform and stable product quality. The dried sodium sulfate by-product is conveyed into the crusher through a conveyor belt for crushing and packaging. The working time is 1 hour, which will produce G2-3 crushing and packaging waste gas.

[0138] The steps of organic cobalt salt are as follows: material and synthesis reaction: weighed cobalt hydroxide is slowly added into the reactor by manual feeding, water is added to stir and disperse, the system temperature is raised to below 40°C, and the measured mixed organic acid, mixed ab diketone, and mixed keto acid are transported to the reactor through a pneumatic diaphragm pump through a pipeline according to proportion and batch, and a promoter is added to cause a synthesis reaction. The system reacts for 6 hours, the conversion rate is 99%, and the yield is 98%. G1-1-1 feeding waste gas and G1-2 reaction waste gas will be generated.

[0139] Efficient separation: After the synthesis reaction is completed, the obtained solid-liquid mixture is pumped into a cooling tank through a pipeline pump to be cooled to 30 degrees, and then pumped into a high-efficiency diaphragm filter press through a pipeline pump for solid-liquid separation. The solid phase is the crude organic cobalt salt product, which is collected by an automatic conveyor belt, and the liquid phase is the recyclable reaction liquid, which is pumped into a collection tank through a pipeline pump and can be used as water for the next batch of reactions. This process takes 0.5 hours. No materials will be entrained under normal working conditions.

[0140] Drying finished product: After separation, the solid is transported to the feed port of the flash dryer by an automatic conveyor belt, and the separated solid is subjected to rotary flash drying at 140°C. This process takes 1 hour, and the steam is collected using a spray device, which will produce G1-3 drying waste gas.

[0141] Material packaging: The powder mixing tank mixes products produced in different batches at room temperature to ensure uniform and stable product quality. The dried materials are conveyed into the crusher through a conveyor belt for crushing and packaging. The working time is 1 hour, which will produce G1-4 crushing and packaging waste gas.

[0142] The steps for 5n high purity organic cobalt salt are as follows:

[0143] Heating furnace feeding: The products measured according to the ratio are added into the heating zone of the device, which will produce G2-1 feeding waste gas.

[0144] System vacuum: Start the mechanical pump to evacuate the system, and observe the system pressure through a thermocouple vacuum gauge, which will produce G2-2 vacuum exhaust gas.

[0145] Ar carrier gas: When the pressure reaches the limit of the mechanical pump, fill in Ar gas, turn off the Ar gas and continue to evacuate, repeat three times. Use the gas flow meter to control the flow of added gas, and continue to fill in Ar gas until the sublimation is completed, which will produce G2-3 carrier gas waste.

[0146] Sublimation deposition: Heat the heating zone II of the device, and quickly stabilize it to a temperature where the purified material can be deposited but the impurities cannot be deposited. Then slowly heat the heating zone I. When some impurities sublime, stabilize it for a period of time, and then continue heating until the product begins to sublime. Maintain this temperature until the product is completely sublimated.

[0147] Ar carrier gas cooling: Stop heating and continue to fill the sublimation furnace with Ar gas until the temperature drops to room temperature, which will produce G2-4 carrier gas waste.

[0148] Collect finished products: Open the sublimation device and collect the products in the heating zone. The flight mass spectrometer detects the purity of the single component to ensure that the product purity reaches 5N. The unqualified products will repeat the above steps for further purification until the product purity reaches 5N, and this step will be terminated, which will produce G2-5 waste residue.

[0149] Material packaging: The purified materials are packaged, the working time is 1 hour, and G2-6 packaging waste gas will be generated.

[0150] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a reference structure" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0151] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A cobalt salt high-purity MO source electronic chemical production process, characterized in that: The process comprises the following steps: Sp1. Raw material pretreatment: purify, dry and screen the high-purity cobalt salt raw materials to ensure that the purity of the raw materials meets the predetermined requirements; Sp2, dissolution and chemical reaction: the pretreated cobalt salt raw material is added to the reactor and dissolved under controllable temperature and pressure conditions. During the dissolution process, the pH value, temperature and reaction rate of the reaction solution are monitored in real time through the reaction parameter monitoring module, and the improved long short-term memory algorithm is used to predict the optimal reaction conditions based on historical process data and dynamically adjust the dissolution conditions; the specific formula of the improved long short-term memory algorithm is as follows: h t : hidden state at time t; x t : Input multidimensional data, including temperature, pressure, reactant concentration and other parameters; W x ,W h ,W c : Weight matrix, representing the connection between different input and hidden layers; i t : Input gate, which controls the impact of new input on the state; c t : Cell state, recording long-term memory information; f: Activation function, here we choose the adaptively adjusted ReLU function; b h ,b c : bias term; Sp3, intelligent crystallization separation: After the dissolution is completed, the reaction mixture is transferred to an intelligent crystallization separation system, which collects multidimensional data of cooling rate, stirring speed and solvent amount, analyzes the multidimensional data in combination with a Bayesian optimization algorithm, optimizes crystallization conditions, dynamically adjusts crystallization parameters, and generates high-purity cobalt salt crystals; the objective function of the Bayesian optimization algorithm is: μ(x): predicted mean under input condition x, representing the reaction output under current conditions; σ(x): predicted variance under input condition x, representing uncertainty; κ: exploration parameter, controlling the balance between mean and variance; x: process parameter; Sp4, drying and vacuum sublimation: drying the high-purity cobalt salt crystals, the drying process is monitored by a data verification adjustment system, and the drying temperature and time are adjusted in real time; vacuum sublimation is performed after drying, and during the vacuum sublimation process, an adaptive dynamic programming algorithm is used to optimize the sublimation parameters according to energy consumption and output effect. The sublimation parameters are as follows: sublimation temperature (T), vacuum degree (P), time (t), condensation temperature (T cloud ), gas flow (F); The adaptive dynamic programming algorithm dynamically adjusts the control parameters by defining an optimization objective function to minimize energy consumption and maximize output effect. The control parameters are: x = [T, P, t, T cloud ,F], and define the optimization objective function as: J(x)=α·E(x)-β·P yield (x) E(x): A function representing energy consumption, which depends on the control parameters x, P yield (x): A function representing the output effect (yield of high-purity cobalt salt), which depends on the control parameter x. α and β are weight coefficients, both of which have values ​​between 0 and 1, and are used to balance the impact of energy consumption and output effect. The state transfer equation is: V(x,t): the state value at time t, x t : control parameter selected at time t, γ: discount factor used to evaluate the impact of future states; Sp5. Virtual production optimization based on digital twins: Before adjusting the actual production process, a virtual production line model is constructed through digital twin technology to simulate and analyze the impact of different process parameters on the purity and physical and chemical properties of cobalt salts, and the parameter settings are optimized in combination with the random forest regression algorithm to ensure the accuracy of the final production process; Sp6, intelligent anomaly detection and adaptive adjustment: use deep learning algorithms to perform real-time analysis on multi-dimensional data collected during the production process, and combine support vector machines and anomaly detection algorithms to predict abnormal situations that may occur in the process; when an anomaly is detected, automatically adjust production parameters to avoid the occurrence of anomalies, ensuring the continuity and stability of the production process; the formula of the support vector machine anomaly detection algorithm: given a training data set: where x i is the multidimensional process parameter, y i For the corresponding category (normal or abnormal), the decision boundary formula of the support vector machine is: α i : Lagrange multiplier; K(x,x i ): kernel function, used to map data to high-dimensional space; b: bias anomaly detection algorithm sets a threshold τ, and when f(x)<τ, it is judged as abnormal; Sp7, Data verification and feedback control: Multi-point sensors are used to collect process parameters throughout the entire production process, including temperature, pressure, material microstructure changes, and reaction kinetics. Adaptive dynamic programming and Bayesian optimization algorithms are used for real-time data analysis, and feedback is used to adjust various parameters in the production process to ensure the high purity and stability of the final product.

2. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 1, characterized in that: In the raw material pretreatment step, purification includes removing impurity ions in the raw materials, and the drying process adopts a vacuum drying method to control the humidity of the raw materials below 10 ppm, and the particle size of the cobalt salt is controlled within 10-30 microns through a screening system.

3. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 1, characterized in that: In the dissolution and chemical reaction steps, the material of the reactor is corrosion-resistant alloy steel, the temperature of the dissolution reaction is controlled at room temperature -200°C, and the pressure is controlled at 1.5-2.5MPa. The pH value during the reaction process is predicted by the improved LSTM model, and the amount of acid and alkali agent added is adjusted according to real-time feedback.

4. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 1, characterized in that: In the intelligent crystallization separation step, the cooling rate of the crystallization system is controlled at 0.5-1°C / minute, the stirring speed is controlled at 200-300rpm, and the cooling rate and the amount of solvent are dynamically adjusted according to multidimensional data.

5. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 1, characterized in that: In the drying and vacuum sublimation steps, the temperature of the drying process is controlled at 50-100°C, and the vacuum sublimation temperature is controlled at 200-600°C.

6. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 1, characterized in that: In the digital twin-based virtual production optimization step, the digital twin model combines the real-time data of the production line to optimize parameters, and uses the random forest regression algorithm to analyze the impact of process parameters such as cooling rate and reaction time on product purity, so as to predict potential problems in production in advance.

7. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 1, characterized in that: In the intelligent anomaly detection and adaptive adjustment step, the anomaly detection algorithm monitors the fluctuations of temperature, pressure, and solubility parameters, and uses the support vector machine model to automatically identify process anomalies that may cause a decrease in product purity, and triggers the adaptive adjustment module when an anomaly is detected to automatically optimize the parameters.

8. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 1, characterized in that: In the data verification and feedback control step, the data acquisition system monitors the process parameters in real time through multi-point sensors installed in the reactor, crystallization equipment, drying and vacuum sublimation equipment.

9. A system for producing a cobalt salt high-purity MO source electronic chemical product according to any one of claims 1 to 8, characterized in that: The system includes the following modules: Raw material pretreatment module, used to purify, dry and screen high-purity cobalt salt raw materials to ensure the purity of raw materials; The dissolution and chemical reaction module includes a reactor, a temperature and pressure control device, and an improved long-short-term memory algorithm unit to monitor and adjust the dissolution conditions in real time; Intelligent crystallization separation module, including cooling control unit, stirring control unit and Bayesian optimization algorithm unit, used to adjust crystallization conditions according to multi-dimensional data; A drying and vacuum sublimation module, including a drying device and an adaptive dynamic programming algorithm unit, for controlling the drying temperature and sublimation parameters; Digital twin virtual production optimization module, which simulates the actual production process through digital twin technology and combines the random forest regression algorithm to optimize parameters; Intelligent anomaly detection and adjustment module, combining deep learning and support vector machine algorithms, analyzes multi-dimensional data in real time and detects production anomalies, automatically adjusting process parameters; The data verification and feedback control module collects process data during the production process through multi-point sensors, and combines Bayesian optimization and adaptive dynamic programming algorithms to perform real-time feedback control.

10. The process for producing high-purity MO source electronic chemicals from cobalt salt according to claim 6, characterized in that: The digital twin virtual production optimization module includes the following parts: State equation: Digital twin technology simulates the dynamic evolution of physical processes through differential equations. The state equation describes the change of system state over time: x(t) represents the state variable vector of the production process, including temperature, pressure, reaction rate, and cooling rate; u(t) represents the control input vector of the system, including adjustable parameters such as cooling rate, stirring rate, and energy input; t is time; f(x,u,t) represents the nonlinear function of the physical process, which is modeled based on the physical characteristics of the system; output equation: The output of the system is some observable variables (including product purity and energy consumption), expressed as: y(t)=g(x(t),u(t),t) y(t) is the system output, which represents the quality characteristics of the product; g(x,u,t) is the output function, which is usually a linear or nonlinear function, depending on the physical relationship in the actual process; the regression model of the random forest regression algorithm: random forest regression uses multiple decision tree models h m (x) Make predictions: represents the predicted value, such as the purity of saw salt crystals in the production process; M is the number of decision trees in the random forest; h m (x) represents the regression prediction of the input feature x by the mth decision tree model; the decision tree splitting rule is as follows: each decision tree uses a binary split method to select a feature x at each node j. i and its threshold t j , with minimizing the mean square error (MSE) as the objective function: y i is the true value of the i-th sample; n is the number of samples in the current node; t j It is the optimal feature threshold for splitting, which is selected by minimizing MSE; Feature importance analysis: The random forest algorithm can evaluate the impact of each production parameter on the final output result. The importance of the feature is evaluated by calculating the contribution of each feature's split point in the decision tree to the MSE: I(f i ) is the feature f i The importance of is the feature f i The resulting MSE reduction; -M is the total number of decision trees.

Images