A purification process of waste lithium battery electrolyte

Abstract

The application relates to a purification process of waste lithium battery electrolyte, and relates to the technical field of battery recycling.The process comprises the following steps: mixing and standing dehydrated electrolyte (with a water content of less than or equal to 20 ppm) with a non-water antisolvent in an inert atmosphere to form a light upper phase and a lithium-rich heavy phase; collecting the interface height and the lithium ion concentration in the lower layer in real time, constructing a time sequence and calculating a layered process characteristic value; inputting the characteristic value into an endpoint prediction model, and determining the separation endpoint when the output layered process completion degree reaches a preset threshold; and taking the maximum lithium enrichment amount of the lithium-rich heavy phase per unit time and the minimum residual lithium concentration of the light phase as optimization objectives, iteratively optimizing the antisolvent ratio and the standing temperature, obtaining optimal parameters and executing the same. The application solves the technical problems that in the waste lithium battery electrolyte recycling process, the traditional direct heating distillation is easy to cause the high-temperature decomposition of lithium salt to generate toxic gas, and the fixed process parameters cannot adapt to the fluctuation of the characteristics of different batches of raw materials, resulting in low separation efficiency and unstable recovery rate.

Description

Technical Field

[0001] This application relates to the field of battery recycling technology, and in particular to a purification process for waste lithium battery electrolyte. Background Technology

[0002] The recycling and treatment of spent lithium-ion battery electrolytes is a crucial step in the resource-based recycling of lithium-ion batteries. Spent electrolytes typically contain lithium salts such as lithium hexafluorophosphate and carbonate organic solvents. Direct discharge causes severe environmental pollution and resource waste. Effectively separating and recovering the lithium salts from the organic solvents is the main technological approach for achieving electrolyte resource utilization. Currently, industrial methods for separating and treating spent electrolytes mainly employ direct heating distillation or rectification. The electrolyte is heated to the boiling points of each component, and by controlling temperature and pressure, the organic solvents are vaporized and condensed for recovery, while the lithium salts remain in the distillation residue.

[0003] However, this direct heating distillation method has significant shortcomings in practical applications. Lithium hexafluorophosphate is prone to thermal decomposition at high temperatures, producing toxic and harmful gases such as phosphorus pentafluoride and hydrogen fluoride, posing safety hazards to operators and equipment, and also leading to lithium salt loss and a decline in recovery quality. Furthermore, the sources of waste electrolytes are complex, with significant differences in lithium salt concentration and organic solvent composition between electrolytes from different manufacturers and different batches of retired batteries. Traditional distillation processes use fixed temperature and pressure parameters for uniform processing, making it impossible to adjust for the actual characteristics of each batch of material. This results in large fluctuations in separation efficiency between different batches, making it difficult to consistently achieve the expected lithium salt recovery rate.

[0004] Besides distillation, although various general liquid-liquid separation technologies exist in the chemical industry, they all face obstacles in directly applying them when dealing with the unique system of waste lithium battery electrolytes. On the one hand, lithium salts in waste electrolytes are highly sensitive to moisture and impurities; the environmental conditions or auxiliary reagents relied upon by conventional separation methods can easily trigger side reactions, causing irreversible loss of lithium salts or introducing new separation challenges. On the other hand, the complex solvation environment remaining in waste electrolytes makes the differences in physical properties between the target component and interfering substances insufficiently significant, making it difficult for existing methods to achieve highly selective targeted separation without sacrificing recovery quality. Therefore, how to overcome the constraints of raw material fluctuations on separation efficiency while avoiding lithium salt loss and secondary pollution, and achieve efficient and stable separation of lithium salts and organic solvents, is a pressing technical problem to be solved in this field. Summary of the Invention

[0005] This application provides a purification process for waste lithium battery electrolyte, which improves the technical problems of incomplete separation or excessively prolonged process time when using liquid-liquid stratification lithium extraction method for waste lithium battery electrolyte, which is faced with differences in the content of key components of waste liquid from different sources, uneven stratification process speed, lack of online judgment means for determining the timing of separation completion, and relying solely on experience to determine the endpoint.

[0006] This application discloses the following technical solution:

[0007] This application provides a purification process for waste lithium battery electrolyte, the method comprising:

[0008] Obtain the initial physical properties of the electrolyte to be treated, and dehydrate the electrolyte to a moisture content of ≤20ppm;

[0009] Under an inert atmosphere, the dehydrated electrolyte is mixed with a non-aqueous antisolvent according to the initial ratio and kept at the initial settling temperature to form a light phase rich in carbonate solvent on the upper layer and a lithium-rich heavy phase rich in lithium salt on the lower layer.

[0010] The interface height of the stratified pool is collected at set time intervals. At the same time, the intensity of the characteristic peak of lithium salt in the lower lithium-rich heavy phase is obtained by online spectroscopy and converted into lithium ion concentration. The interface height time series and lithium ion concentration time series are constructed.

[0011] The interface rise rate and concentration growth rate are calculated to determine the stratification process characteristic value and construct the stratification process characteristic sequence. The stratification process characteristic sequence is input into the endpoint prediction model and the stratification process completion degree is output. When the stratification process completion degree reaches the preset threshold, the separation endpoint is determined to be reached.

[0012] With the optimization objectives of maximizing the lithium enrichment in the lithium-rich heavy phase and minimizing the residual lithium concentration in the light phase, a multi-objective fitness function is constructed to iteratively optimize the antisolvent ratio and settling temperature until convergence, thereby obtaining the optimal antisolvent ratio and optimal settling temperature for the current batch and executing them.

[0013] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0014] The technical solution of this application provides a purification process for waste lithium battery electrolyte. First, by obtaining the initial physical property parameters of the electrolyte to be treated and performing deep dehydration treatment, the water content is controlled below 20ppm. This achieves accurate acquisition of key component information of different batches of raw materials and deep removal of water, providing an anhydrous and oxygen-free process basis for the stability and consistency of subsequent liquid-liquid stratification, and solving the problem of unstable separation effect caused by unclear raw material characteristics and water interference.

[0015] Furthermore, by mixing the dehydrated electrolyte with a non-aqueous antisolvent in an inert atmosphere, and then stirring and settling at a constant temperature, a light phase rich in carbonate solvent is formed on the upper layer and a heavy phase rich in lithium salt is formed on the lower layer. By utilizing the difference in solubility of lithium salt and organic solvent in the antisolvent, the two phases are efficiently separated under non-high temperature conditions, avoiding the problems of lithium salt decomposition and toxic gas release caused by high-temperature distillation.

[0016] Furthermore, by collecting the interface height of the stratified pool and the lithium ion concentration of the lower lithium-rich heavy phase in real time online, a time series was constructed. The interface rise rate and concentration increase rate were calculated and converted into a stratification process feature vector. The completion degree of the stratification process was predicted by using a machine learning model. This achieved real-time quantitative perception of the stratification process and automatic determination of the separation endpoint, solving the problem that traditional methods rely solely on experience to judge the endpoint timing, resulting in incomplete separation or excessively long process time.

[0017] Finally, by taking the dual objectives of maximizing the lithium enrichment in the heavy phase and minimizing the residual lithium concentration in the light phase per unit time, a multi-objective fitness function was constructed to iteratively optimize the antisolvent ratio and settling temperature. This enabled adaptive matching of process parameters for the characteristics of each batch of raw materials, automatically finding the optimal balance between lithium recovery efficiency and separation quality. This solved the problem that fixed process parameters could not adapt to raw material fluctuations, leading to batch-to-batch instability in recovery rate and product quality.

[0018] In summary, this application achieves efficient separation and enrichment of lithium salts under non-high-temperature conditions by performing deep dehydration and low-temperature antisolvent stratification on waste lithium battery electrolytes, using online spectroscopy to monitor the stratification process in real time and predicting the separation endpoint based on machine learning, and combining multi-objective iterative optimization to adaptively match the antisolvent ratio and settling temperature. This effectively improves the technical problems of traditional direct heating distillation, which easily leads to lithium salt decomposition, and the inability of process parameters to adapt to raw material fluctuations, resulting in low and unstable recovery efficiency. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic flow chart of a purification process for waste lithium battery electrolyte provided in an embodiment of this application;

[0021] Figure 2 A flowchart illustrating the online spectral data validity determination process for a purification process of waste lithium battery electrolyte provided in this application embodiment. Detailed Implementation

[0022] This application provides a purification process for waste lithium battery electrolyte, which solves the technical problem that in the lithium extraction process of waste lithium battery electrolyte by anti-solvent separation, the content of key components of the waste liquid to be treated fluctuates, and the process parameters such as anti-solvent dosage and operating temperature are difficult to automatically match the optimal conditions for each batch of material, resulting in inconsistent recovery efficiency and product quality between different batches.

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. It should be noted that the numerical values ​​in the embodiments are for illustrative purposes only and do not constitute a limitation on this application.

[0024] Examples, as shown in the appendix Figure 1 As shown, this application provides a purification process for waste lithium battery electrolyte, the process including the following steps:

[0025] S100: Obtain the initial physical property parameters of the electrolyte to be treated, and dehydrate the electrolyte to be treated until the water content is ≤20ppm.

[0026] In this embodiment, the spent lithium battery electrolytes come from different manufacturers and different batches of retired batteries, with significant differences in lithium salt concentration and organic solvent composition. Furthermore, the electrolytes inevitably come into contact with air and contain moisture during the recycling and dismantling process. Lithium hexafluorophosphate is extremely sensitive to water; residual moisture can trigger lithium salt hydrolysis and produce hydrogen fluoride, interfering with subsequent liquid-liquid stratification and posing safety hazards. Therefore, this step requires precise analysis of the key component indicators of each batch of electrolyte and deep dehydration to control the moisture content to trace levels, providing an anhydrous and oxygen-free environment for the stability and consistency of the subsequent stratification process.

[0027] Step S100 in the method provided in this application embodiment includes:

[0028] The initial physical properties include at least lithium salt concentration, organic solvent composition, and moisture content. The lithium salt concentration is obtained from the electrolyte sample using inductively coupled plasma atomic emission spectrometry (ICP-AES). The organic solvent composition is obtained from the electrolyte sample using gas chromatography-mass spectrometry (GC-MS). The moisture content is obtained from the electrolyte sample using Karl Fischer chromatography. Dehydration is performed using a combination of 3A molecular sieve adsorption and vacuum evaporation. The amount of 3A molecular sieve used is 2% to 10% of the electrolyte mass. Vacuum evaporation is carried out at a temperature of 30°C to 60°C and an absolute pressure of 0.1 kPa to 5 kPa. Dehydration is performed until the moisture content, as determined by Karl Fischer chromatography, is ≤20 ppm. Detailed explanation follows:

[0029] In this embodiment, the initial physical properties refer to the set of key component indicators of the electrolyte to be treated, determined by standard analytical methods before entering the purification process. These include at least lithium salt concentration, organic solvent composition, and moisture content, serving as the basic input data for subsequently determining process parameters such as the antisolvent ratio and settling temperature. Lithium salt concentration refers to the molar concentration of lithium salts, primarily lithium hexafluorophosphate, in the electrolyte, calculated by converting the lithium content determined by inductively coupled plasma atomic emission spectrometry (ICP-AES). Organic solvent composition refers to the mass fraction of carbonate solvents such as ethylene carbonate, dimethyl carbonate, and ethyl methyl carbonate in the electrolyte, calculated by peak area normalization after separating and identifying each component using gas chromatography-mass spectrometry (GC-MS).

[0030] Moisture content refers to the mass fraction of trace water dissolved in the electrolyte, determined by the Karl Fischer method. 3A molecular sieve is a crystalline aluminosilicate with a uniform microporous structure and a pore size of approximately 0.3 nanometers. It selectively adsorbs water molecules with diameters smaller than this pore size, while not adsorbing larger organic solvent molecules. Vacuum-reduced evaporation dehydration involves heating the electrolyte under sub-atmospheric pressure, causing water to preferentially vaporize and be extracted at a lower temperature, thus preventing the decomposition of lithium salts due to high temperatures.

[0031] In this step, firstly, to obtain key component information of the current batch of electrolyte to be treated to guide subsequent process parameter setting, samples need to be taken from the electrolyte. The lithium salt concentration is determined using inductively coupled plasma atomic emission spectrometry (ICP-AES), the organic solvent composition is determined using gas chromatography-mass spectrometry (GC-MS), and the moisture content is determined using the Karl Fischer method. These three analyses are performed simultaneously, and the measurements are completed as soon as possible after sampling to minimize the electrolyte's exposure time to air.

[0032] Furthermore, to achieve deep dehydration and control the moisture content to a trace level that does not affect subsequent reactions, a combined process of 3A molecular sieve adsorption dehydration and vacuum evaporation dehydration is required. The specific operation of molecular sieve adsorption dehydration involves adding 3A molecular sieves to the electrolyte to be treated at 2% to 10% of the electrolyte mass, continuously stirring under an inert atmosphere, and utilizing the microporous structure of the molecular sieve to selectively adsorb water molecules.

[0033] After a period of adsorption treatment, the molecular sieve is separated by filtration or sedimentation. Then, vacuum evaporation dehydration is performed. The electrolyte after molecular sieve adsorption is placed in a sealable container, and a vacuum is drawn to an absolute pressure of 0.1 kPa to 5 kPa. Under heating conditions of 30°C to 60°C, the water preferentially boils and vaporizes and is removed, while the carbonate solvent and lithium salt remain stable under these conditions, achieving deep dehydration. Vacuum evaporation dehydration continues until a sample is taken and the water content is determined by the Karl Fischer method to be ≤20 ppm. After dehydration, the dehydrated electrolyte is sealed and stored under an inert atmosphere for subsequent mixing with antisolvent and constant-temperature settling.

[0034] In summary, this step determined the lithium salt concentration, organic solvent composition, and moisture content using inductively coupled plasma atomic emission spectrometry, gas chromatography-mass spectrometry, and Karl Fischer method, respectively. Furthermore, a deep dehydration process combining 3A molecular sieve adsorption and vacuum evaporation was employed to control the moisture content to ≤20 ppm, thus resolving the issues of unclear characteristics of different batches of raw materials and the interference of moisture with subsequent stratification.

[0035] S200: Under an inert atmosphere, the dehydrated electrolyte is mixed with a non-aqueous antisolvent according to the initial ratio and kept at a constant temperature at the initial settling temperature to form a light phase rich in carbonate solvent on the upper layer and a lithium-rich heavy phase rich in lithium salt on the lower layer.

[0036] In this embodiment, after deep dehydration of the electrolyte, efficient separation of the carbonate solvent and lithium salt is achieved under non-high-temperature conditions through liquid-liquid layering. The physical principle of this layering process is as follows: the non-aqueous antisolvent is miscible with the carbonate solvent, but has extremely low solubility for lithium salt. When the antisolvent is added to the electrolyte, a thermodynamically unstable state occurs in the mixed system. Due to the difference in solubility, the lithium salt is repelled from the carbonate solvent and spontaneously migrates and accumulates in the antisolvent phase, forming a lithium-rich heavy phase dominated by the antisolvent; the carbonate solvent floats to the surface due to the density difference with the antisolvent, forming a light phase. Traditional direct heating distillation methods operate at high temperatures, posing risks of lithium salt decomposition and the release of toxic gases. This step achieves efficient two-phase construction through non-high-temperature liquid-liquid layering, avoiding safety issues caused by high temperatures.

[0037] Step S200 in the method provided in this application embodiment includes:

[0038] Under stirring conditions, the entire amount of the non-aqueous antisolvent is added to the dehydrated electrolyte over 10 to 60 minutes, with the system temperature controlled not to exceed 25°C during the addition process. After the addition is complete, the system temperature is raised to a second temperature value at a heating rate of 0.5°C / min to 2.0°C / min, and stirred at the second temperature value for 10 to 30 minutes at a stirring speed of 100 to 300 rpm, wherein the second temperature value is 5°C to 20°C higher than the initial settling temperature. The system temperature is then lowered from the second temperature value to the initial settling temperature at a cooling rate of 0.2°C / min to 1.0°C / min, and stirred at a stirring speed of 30 to 80 rpm during the cooling process. After reaching the initial settling temperature, stirring is stopped, and the mixture is kept at the initial settling temperature until a clear two-phase interface appears, completing the construction of the light phase and the lithium-rich heavy phase. Detailed explanation follows:

[0039] In this embodiment, the non-aqueous antisolvent refers to an anhydrous organic solvent, which is mixed with the dehydrated electrolyte under an inert atmosphere, has extremely low solubility for lithium salts, and has a significantly different density from the carbonate solvent. Its function is to promote the precipitation and enrichment of lithium salts from the carbonate solution to form a lithium-rich heavy phase. The initial settling temperature refers to the target temperature at which the mixed system is finally maintained at a constant temperature after undergoing a temperature-controlled stirring procedure. During the layering process, at this temperature, the mass transfer and interphase equilibrium of lithium salts continue to occur slowly between the two phases until a stable interface appears.

[0040] The light phase refers to the liquid phase located on the upper layer after stratification, which is mainly composed of carbonate solvent and has a low lithium salt content. The lithium-rich heavy phase refers to the liquid phase located on the lower layer after stratification, which is mainly composed of antisolvent and enriched with most of the lithium salt, and is the target product for lithium salt recovery in this process. The second temperature value is a peak temperature for heating and stirring, which is 5°C to 20°C higher than the initial settling temperature. At this temperature, the molecular diffusion rate increases, the system viscosity decreases, and the frequency of molecular collisions between the antisolvent and carbonate solvent and the mass transfer rate of lithium salt between the two phases increase.

[0041] In this step, firstly, to ensure sufficient contact between the antisolvent and the electrolyte and to avoid particle agglomeration caused by uneven local lithium salt supersaturation, the non-aqueous antisolvent is slowly and uniformly added to the dehydrated electrolyte over 10 to 60 minutes under stirring. Before adding the non-aqueous antisolvent to the dehydrated electrolyte, it is pre-cooled to the first pre-cooling temperature at a rate of 0.2 to 0.5°C / min to increase the density difference between the antisolvent and the room-temperature electrolyte, promoting subsequent stratification. During the addition process, continuous stirring is maintained, and the system temperature is controlled not to exceed 25°C to prevent local overheating that could lead to lithium salt decomposition. The non-aqueous antisolvent is selected from at least one of trifluorotoluene or methyl pentafluoropropionate, with a water content of less than 50 ppm.

[0042] Furthermore, to accelerate the mass transfer of lithium salt in the mixed system and shorten the equilibrium time through temperature fluctuations, the mixed system needs to be stirred under controlled temperature. The volume ratio of the non-aqueous antisolvent to the dehydrated electrolyte is determined by the lithium salt concentration: when the lithium salt concentration is ≤1.0 mol / L, the volume ratio is 0.8:1 to 1.2:1; when the lithium salt concentration is >1.0 mol / L, the volume ratio is 1.2:1 to 2.0:1. After the addition is complete, the system temperature is raised to the second temperature value at a heating rate of 0.5℃ / min to 2.0℃ / min, and stirred at this temperature for 10 min to 30 min at a stirring speed of 100 rpm to 300 rpm to ensure thorough mixing of the antisolvent and carbonate solvent. The system temperature was then reduced to the initial settling temperature at a cooling rate of 0.2℃ / min to 1.0℃ / min. During the cooling process, the system was stirred at a low speed of 30 rpm to 80 rpm to allow the lithium salt to migrate orderly from the carbonate solvent phase to the antisolvent phase during the slow cooling process. This also prevented the lithium salt from precipitating out in the form of fine particles due to rapid cooling, which would make it difficult for the particles to settle.

[0043] Step S200 in the method provided in this application embodiment further includes:

[0044] The non-aqueous antisolvent is selected from at least one of trifluorotoluene or methyl pentafluoropropionate, and the water content of the non-aqueous antisolvent is less than 50 ppm; the volume ratio of the non-aqueous antisolvent to the dehydrated electrolyte is determined by the lithium salt concentration, and when the lithium salt concentration is ≤1.0 mol / L, the volume ratio is 0.8:1 to 1.2:1, and when the lithium salt concentration is >1.0 mol / L, the volume ratio is 1.2:1 to 2.0:1; before adding the non-aqueous antisolvent to the dehydrated electrolyte, it is first diluted with 0. The lithium salt is pre-cooled at a rate from 0.2℃ / min to 0.5℃ / min to a first pre-cooling temperature, which is determined by the lithium salt concentration and the mass fraction of ethylene carbonate in the organic solvent composition. When the lithium salt concentration is ≤1.0 mol / L and the mass fraction of ethylene carbonate is ≤15%, the first pre-cooling temperature is 15℃ to 20℃; when the lithium salt concentration is >1.0 mol / L or the mass fraction of ethylene carbonate is >15%, the first pre-cooling temperature is 5℃ to 15℃. A detailed explanation follows:

[0045] In this embodiment, the first pre-cooling temperature refers to the temperature at which the non-aqueous antisolvent needs to be pre-cooled before being added to the dehydrated electrolyte. This pre-cooling temperature is set in segments based on the different lithium salt concentrations and ethylene carbonate mass fractions in the electrolyte. The physical basis for this is that higher lithium salt concentrations or higher ethylene carbonate content result in higher viscosity of the mixed system, slower molecular motion rates, and greater mass transfer resistance of lithium salt between the two phases. Using a lower pre-cooling temperature can effectively increase the density difference and solubility gradient driving force between the antisolvent and the electrolyte, promoting faster migration of lithium salt from the carbonate phase to the antisolvent phase. This compensates for the insufficient mass transfer rate in high-viscosity systems, thereby ensuring consistency in stratification speed and separation effect under different electrolyte component conditions.

[0046] In this step, the mixing system is cooled to the initial settling temperature and then stirred until it is kept at the initial settling temperature. During the settling process, the lithium salt continues to accumulate in the antisolvent phase, while the carbonate solvent, due to its lower density, floats to the surface to form a lighter phase. As time progresses, the interface between the two phases gradually becomes clearer. When a clear interface between the two phases that is visible to the naked eye appears, the construction of the lighter phase and the lithium-rich heavier phase is complete.

[0047] In summary, this step involves mixing a non-aqueous antisolvent with the dehydrated electrolyte under an inert atmosphere with temperature-controlled stirring. Following a temperature-controlled process of accelerated mass transfer through heating and orderly migration through cooling, the mixture is allowed to stand at a constant temperature for layering. Utilizing the differences in solubility and density between the antisolvent and the carbonate solvent, efficient separation of lithium salt and organic solvent is achieved under non-high-temperature conditions, constructing an immiscible light phase and a lithium-rich heavy phase. Compared to existing technologies, this step avoids the lithium salt decomposition and toxic gas release problems caused by high-temperature distillation. Furthermore, by setting segmented pre-cooling temperatures and volume ratios for the antisolvent, the layering process can adapt to fluctuations in lithium salt concentration and ethylene carbonate content in different batches of electrolyte, ensuring the stability and batch-to-batch consistency of the layering effect.

[0048] S300: Collects the interface height of the layered pool at set time intervals, and simultaneously obtains the characteristic peak intensity of lithium salt in the lower lithium-rich heavy phase through online spectroscopy and converts it into lithium ion concentration, constructing a time series sequence of interface height and lithium ion concentration.

[0049] In this embodiment, after the formation of a two-phase system consisting of a light phase and a lithium-rich heavy phase, the mass transfer and migration of lithium salts within the stratification pool continue, with the interface position and heavy phase lithium concentration dynamically changing over time. Changes in interface height reflect the progress of macroscopic separation between the two phases, while changes in lithium ion concentration reflect the enrichment degree of lithium salts in the heavy phase; both are crucial indicators for determining the stratification process and the separation endpoint. Traditional methods typically employ manual, timed sampling. Opening the container for sampling disrupts the slightly positive pressure inert atmosphere within the stratification pool and disturbs the stable state of the two-phase interface. Furthermore, offline detection suffers from data lag, making it difficult to capture the continuous concentration change trend during stratification in a timely manner. Therefore, this step requires the simultaneous acquisition of real-time data on interface height and lithium ion concentration using online sensors and online spectroscopy without disrupting the stratification environment. This provides a continuous and stable time-series data foundation for subsequent calculation of stratification process characteristic values. The online spectral data validity determination process for this step is as follows: Figure 2 As shown.

[0050] Step S300 in the method provided in this application embodiment includes:

[0051] An array of capacitive level sensors with equal spacing is arranged along the vertical direction of the stratified pool. The capacitance value of each sensor is read at the set time interval. The interface height is determined based on the abrupt change position of the capacitance value along the vertical direction, and a time series sequence of interface height is generated. A near-infrared spectroscopy fiber optic probe is installed on the lower sidewall of the stratified pool. The absorption spectrum of the lower lithium-rich heavy phase at the characteristic wavenumber is collected at the set time interval. The peak height value of the absorption spectrum at the characteristic wavenumber is extracted as the lithium salt characteristic peak intensity, and a time series sequence of lithium salt characteristic peak intensity is generated.

[0052] For each sampling time, the N consecutive peak height values ​​preceding the current time in the lithium salt characteristic peak intensity time series are obtained. It is determined whether the coefficient of variation of these N consecutive peak height values ​​is less than 10%, where N is an integer greater than or equal to 3. When the coefficient of variation is less than 10%, the arithmetic mean of the N consecutive peak height values ​​is used as the stable intensity of the lithium salt characteristic peak at the current time, and the lithium ion concentration at the current time is obtained and added to the lithium ion concentration time series. When the coefficient of variation is greater than or equal to 10%, the peak height value at the current time is removed, and the lithium ion concentration from the previous sampling time is used as the lithium ion concentration at the current time and added to the lithium ion concentration time series. The interface height value and lithium ion concentration value at the same sampling time are aligned according to the sampling time to form a time-synchronized interface height time series and lithium ion concentration time series. A detailed explanation follows:

[0053] In this embodiment, the interface height refers to the vertical position of the interface between the upper light phase and the lower lithium-rich heavy phase in the stratified pool. The distance measured upwards from the bottom of the stratified pool is taken as the zero point, and its change over time reflects the macroscopic process of the light phase gradually increasing in volume and the heavy phase gradually decreasing in volume. The capacitive level sensor array refers to a row of capacitive probes installed at equal intervals along the vertical outer wall of the stratified pool. Utilizing the difference in dielectric constant between the light and heavy phases, the change in capacitance formed by each probe and the pool wall detects the location range of the gas-liquid or liquid-liquid interface. The lithium salt characteristic peak intensity refers to the peak height of the absorption peak produced by lithium hexafluorophosphate at a specific wavenumber position in the near-infrared spectrum. The peak height is positively correlated with the concentration of lithium ions in the heavy phase at that moment.

[0054] A near-infrared spectroscopy fiber optic probe is an online detection element installed on the lower sidewall of the stratified pool and connected to an external spectrometer via optical fiber. Its probe end face is in direct contact with the recurrent phase, emitting near-infrared light into the recurrent phase and receiving reflected or transmitted signals. The coefficient of variation (COP) is a statistical indicator that measures the relative fluctuation of N consecutive peak heights. It is used to determine the stability of the spectral signal at the current moment; a higher COP indicates a more unstable signal. The N consecutive peak heights are taken from the peak heights corresponding to the N sampling times prior to the current moment. When the COP is less than 10%, the signal is smoothed; when the COP is greater than or equal to 10%, the signal is considered to have abnormal fluctuations and is filled using the effective concentration value from the previous moment.

[0055] The threshold value for the coefficient of variation was determined as follows: During the process development phase, spectral stability tests were conducted on the stratification process of five different batches of waste electrolyte. After each batch reached a stable stratification stage, peak height values ​​were continuously collected at 30 sampling points, and the coefficient of variation for each batch was calculated. The statistical results showed that the coefficient of variation for the five batches ranged from 3.2% to 6.8%. Considering that the fiber optic probe end face may experience occasional signal jumps due to the adhesion of micro-bubbles or temperature fluctuations during online detection, the coefficient of variation corresponding to such jumps is usually between 10% and 25%. To effectively distinguish such abnormal fluctuations from normal signals, the threshold was set to 10%, which can effectively identify and eliminate abnormal fluctuation signals while covering the upper limit of the normal signal fluctuation range.

[0056] The value of N is determined based on the sampling frequency and the typical response time of the stratification process, and is typically between 5 and 10. This range is based on the fact that when the sampling interval is 1 to 5 minutes, the time window corresponding to 5 to 10 sampling points is 5 to 50 minutes, which matches the typical timescale of observable changes in interface height and lithium-ion concentration during the stratification process, achieving a balance between signal smoothing and rate calculation sensitivity. If the value of N is too small, the amount of data within the window will be insufficient to stabilize the regression slope; if the value of N is too large, the window spans too long, causing the rate calculation to lag behind the actual stratification process.

[0057] In this step, firstly, to acquire real-time data on interface height changes online, an array of equally spaced capacitive level sensors needs to be arranged vertically along the stratified pool. Each sensor reads its capacitance value at set time intervals. Due to the significant difference in dielectric constant between the light-phase carbonate solvent and the heavy-phase lithium-rich antisolvent, the capacitance value changes abruptly near the interface. The position corresponding to the point where the capacitance value of adjacent sensors along the vertical direction is the largest is taken as the interface height. The interface height values ​​at each sampling time are arranged in chronological order to generate an interface height time series.

[0058] Furthermore, to obtain real-time data on the changes in lithium salt concentration in the lower lithium-rich heavy phase, a near-infrared spectroscopy fiber optic probe needs to be installed on the lower sidewall of the stratified pool. Absorption spectra of the lower lithium-rich heavy phase at characteristic wavenumbers are acquired at set time intervals, and the peak heights at these wavenumbers are extracted as the characteristic peak intensities of the lithium salt, generating a time series sequence of lithium salt characteristic peak intensities. All spectral acquisition operations are performed under an inert atmosphere.

[0059] The time interval is typically set between 1 and 5 minutes. This range is determined based on the typical rate of change of interface height and lithium-ion concentration during the stratification process: in the early stages of stratification, the mass transfer rate is relatively fast, so a smaller time interval is preferable to capture the details of rapid interface movement and rapid concentration increases; in the later stages of stratification, the changes slow down, and the time interval can be appropriately widened to reduce data redundancy. The 1- to 5-minute interval range strikes a balance between data acquisition density and system storage overhead.

[0060] Furthermore, to stabilize the spectral signal and handle outliers, the coefficient of variation (COP) needs to be calculated for the N consecutive peak heights preceding each sampling time. When the COP is less than 10%, the signal is considered stable, and the arithmetic mean of these N peak heights is used as the stable intensity of the lithium salt characteristic peak. This COP is then substituted into the lithium salt concentration calibration curve to obtain the lithium ion concentration at the current time. When the COP is greater than or equal to 10%, the signal exhibits abnormal fluctuations. The peak height at that time is discarded, and the lithium ion concentration from the previous sampling time is used as the lithium ion concentration at the current time. The current lithium ion concentration is then added to the lithium ion concentration time series.

[0061] Finally, the interface height value and lithium ion concentration value at the same sampling time are aligned according to the sampling time to form a time-synchronized interface height time series and lithium ion concentration time series, which are used for the calculation of feature values ​​in the subsequent stratification process.

[0062] Step S300 in the method provided in this application embodiment further includes:

[0063] Read the peak height values ​​of the lithium salt characteristic peak intensity time series before the current sampling time, and calculate the arithmetic mean of the N consecutive peak height values; calculate the standard deviation of the N consecutive peak height values ​​based on the degree of difference between each peak height value and the arithmetic mean; divide the standard deviation by the arithmetic mean and multiply by 100% to obtain the coefficient of variation; when the coefficient of variation is less than 10%, use the arithmetic mean of the N consecutive peak height values ​​as the stable intensity of the lithium salt characteristic peak at the current sampling time, and substitute the stable intensity of the lithium salt characteristic peak into the lithium salt concentration calibration curve to obtain the lithium ion concentration at the current sampling time;

[0064] The steps for constructing the lithium salt concentration calibration curve include: preparing at least three standard solutions with different known lithium ion concentrations, wherein the matrix of the standard solutions is prepared by mixing trifluorotoluene or methyl pentafluoropropionate with carbonate solvents in a certain ratio of carbonate solvents in the organic solvent composition; acquiring the absorption spectra of each standard solution at the characteristic wavenumber and extracting the peak height values ​​of each standard solution at the characteristic wavenumber; performing linear regression fitting with the peak height values ​​of each standard solution on the x-axis and the corresponding known lithium ion concentration on the y-axis to obtain the lithium salt concentration calibration curve; when the coefficient of variation is greater than or equal to 10%, removing the peak height value at the current sampling time and using the lithium ion concentration at the previous sampling time as the lithium ion concentration at the current sampling time; adding the lithium ion concentration at the current sampling time to the lithium ion concentration time series to complete the update of the lithium ion concentration at the current sampling time. Detailed explanation follows:

[0065] In this embodiment, the standard deviation is a commonly used indicator in statistics to quantify the degree to which a set of values ​​deviates from its average level. Its calculation formula is the square root of the sum of the squares of the differences between each data point and the arithmetic mean, divided by the number of data points minus one. The coefficient of variation is a relative volatility indicator obtained by dividing the standard deviation by the arithmetic mean and then multiplying by 100%. It is dimensionless and used to compare the dispersion of datasets at different mean levels.

[0066] The stable intensity of a lithium salt characteristic peak refers to the arithmetic mean of N consecutive peak heights, used as the denoised spectral intensity at the current moment after confirming the signal stability. This improves the accuracy of peak height values ​​in characterizing lithium ion concentration. A lithium salt concentration calibration curve is a linear regression curve obtained by preparing a series of standard solutions of known concentrations, measuring their peak heights at characteristic wavenumbers, and fitting the peak height value to the known lithium ion concentration on the vertical axis. This curve is used to quantitatively convert spectral peak height values ​​into lithium ion concentration values.

[0067] In this step, firstly, to determine the reliability of the spectral signal at the current moment, the stability of the peak height values ​​needs to be evaluated based on the coefficient of variation. The peak height values ​​of the N consecutive sampling moments preceding the current sampling moment are read from the lithium salt characteristic peak intensity time series. The arithmetic mean of the N consecutive peak height values ​​is calculated, and then the standard deviation is calculated. The complete formula for calculating the standard deviation is: Standard Deviation The coefficient of variation is obtained by dividing the standard deviation by the arithmetic mean and then multiplying by 100%.

[0068] Furthermore, when the coefficient of variation is less than 10%, it indicates that the peak height fluctuation within the current time window is small and the signal is stable. This arithmetic mean is used as the stable intensity of the lithium salt characteristic peak at the current sampling time. The stable intensity of the lithium salt characteristic peak is substituted into the lithium salt concentration calibration curve, and the lithium ion concentration at the current sampling time is calculated through curve interpolation or regression equation.

[0069] Furthermore, to ensure the consistency between the calibration curve and the actual detection conditions, a standard solution matching the lower lithium-rich heavy phase matrix needs to be prepared. The matrix of the standard solution is prepared by mixing trifluorotoluene or methyl pentafluoropropionate with carbonate solvents according to the relative proportions of the components such as ethylene carbonate, dimethyl carbonate, and ethyl methyl carbonate in the organic solvent composition, and then adding a known amount of lithium hexafluorophosphate to prepare a concentration gradient solution. The absorption spectra of each standard solution are collected at the characteristic wavenumber, and the peak height values ​​of each standard solution at the characteristic wavenumber are extracted.

[0070] A linear regression was performed with the peak height of each standard solution on the x-axis and the corresponding known lithium-ion concentration on the y-axis to obtain the lithium salt concentration calibration curve. The linear correlation coefficient of the calibration curve should not be less than 0.999, and the linear range should cover 0.1 mol / L to 4.0 mol / L. The lower limit of this linear range was determined based on statistical analysis of lithium salt concentrations in 102 batches of waste electrolytes provided by multiple retired battery dismantling companies. The lowest lithium salt concentration was 0.18 mol / L. Considering that the lithium salt concentration of electrolytes in some deeply discharged batteries may be even lower, the lower limit of the linear range was set at 0.1 mol / L to cover extremely low concentration cases.

[0071] The upper limit of the linear range is determined as follows: In the layered process, the lithium-ion concentration in the lithium-rich heavy phase after anti-solvent extraction can reach up to 3.8 mol / L. Setting the upper limit of the linear range to 4.0 mol / L ensures that the lithium-ion concentration in the heavy phase is directly quantified within the linear range of the calibration curve, eliminating the need for dilution and reducing online detection errors. The limit of detection (LOD) is 0.02 mol / L, and the limit of quantitation (LOQ) is 0.05 mol / L. These are determined by serially diluting the lowest concentration standard solution and collecting spectra at characteristic wavenumbers. The concentration corresponding to a signal-to-noise ratio (SNR) of 3 is used as the LOD, and the concentration corresponding to a SNR of 10 is used as the LOQ. The preparation temperature of the standard solution and the spectral acquisition temperature must be consistent with the initial settling temperature or the first pre-cooling temperature in the layered process to ensure that the calibration curve is consistent with the thermodynamic state of the actual detection environment.

[0072] When the coefficient of variation is greater than or equal to 10%, it indicates that the peak height fluctuates significantly and the spectral signal is unstable within the current time window. This may be due to instantaneous noise in the spectrometer or obstruction of the fiber optic probe end face by tiny air bubbles. In this case, the peak height value at the current sampling time is discarded, and the lithium ion concentration at the previous sampling time is used as the effective concentration value at the current sampling time. The lithium ion concentration at the current sampling time is then added to the lithium ion concentration time series to complete the concentration update for the current sampling time.

[0073] In summary, this step simultaneously acquires interface height and lithium ion concentration data using an online capacitive liquid level sensor array and an online near-infrared spectral fiber optic probe. It also introduces a coefficient of variation judgment and signal anomaly rejection mechanism to process the spectral signal, enabling real-time, continuous, and stable time-series data on interface height and lithium ion concentration without opening the lid for sampling. This solves the problems of traditional manual sampling disrupting the stability of the stratified environment and offline detection data lag.

[0074] S400: Calculate the interface rise rate and concentration growth rate, determine the stratification process characteristic value and construct the stratification process characteristic sequence, input the stratification process characteristic sequence into the endpoint prediction model, output the stratification process completion degree, and determine that the separation endpoint has been reached when the stratification process completion degree reaches the preset threshold.

[0075] In this embodiment, after obtaining real-time time-series data on interface height and lithium-ion concentration, it is necessary to quantify the dynamic changes of these two dimensions into unified stratification process feature values, and use a model trained on historical data to predict the degree of completion of the stratification process to automatically determine the separation endpoint. In traditional methods, operators usually judge whether stratification is complete based on experience, or use a fixed settling time as the stopping standard. This can easily lead to problems such as stopping too early, resulting in incomplete separation and low lithium salt recovery rate, or stopping too late, resulting in increased time costs and excessive residual lithium concentration in the light phase. This step integrates information from two dimensions—interface rise and lithium salt enrichment—and uses a machine learning model to learn from historical stratification patterns, transforming manual experience-based judgment into data-driven automated endpoint determination, thereby achieving continuous quantitative evaluation of the degree of completion of the stratification process.

[0076] Step S400 in the method provided in this application embodiment includes:

[0077] The interface height values ​​at the current sampling time and the N consecutive sampling times preceding the current sampling time are read from the interface height time series. The lithium ion concentration values ​​corresponding to the N sampling times are read from the lithium ion concentration time series. A linear regression is performed on the N interface height values ​​with the sampling time as the independent variable and the interface height value as the dependent variable. The slope of the straight line obtained by the linear regression is taken as the interface rise rate. A linear regression is also performed on the N lithium ion concentration values ​​with the sampling time as the independent variable and the lithium ion concentration as the dependent variable. The slope of the straight line obtained by the linear regression is taken as the concentration increase rate.

[0078] The interface rise rate is divided by the interface height value at the initial sampling time in the interface height time series to obtain the normalized interface rise rate. The concentration increase rate is then divided by the maximum lithium ion concentration within the current window in the lithium ion concentration time series to obtain the normalized concentration increase rate. The normalized interface rise rate and the normalized concentration increase rate are multiplied to obtain the stratification process feature value at the current sampling time. This stratification process feature value is then added to the stratification process feature sequence to complete the construction of the stratification process feature sequence. A detailed explanation follows:

[0079] In this embodiment, the interface rise rate refers to the average rate of change of interface height with sampling time over the most recent N consecutive sampling times, extracted using a linear regression method. The concentration growth rate refers to the average rate of change of lithium ion concentration in the lower lithium-rich heavy phase with sampling time over the same time window. The stratification process characteristic value is a dimensionless comprehensive index obtained by multiplying the normalized interface rise rate by the normalized concentration growth rate, used to characterize the overall progress of phase separation and lithium salt enrichment during the stratification process.

[0080] The normalized interface rise rate is obtained by dividing the interface rise rate by the interface height value at the initial sampling time in the interface height time series. The interface height value at the initial sampling time is used as the normalization benchmark because it reflects the initial interface position of the batch of materials after mixing. Differences exist between batches due to the influence of antisolvent ratio and feed amount. Using this as a normalization factor eliminates the influence of batch initial interface height differences on rate comparison, making the normalized interface rise rate a dimensionless indicator reflecting only the speed of stratification. The normalized concentration growth rate is obtained by dividing the concentration growth rate by the maximum lithium ion concentration within the current window in the lithium ion concentration time series. The maximum lithium ion concentration within the current window is used as the normalization benchmark because the maximum concentration value within the window represents the upper limit of the heavy phase lithium salt enrichment degree in the current stratification stage. Using this as a normalization factor eliminates the influence of differences in lithium salt concentration between different batches of raw materials and differences in stratification stages on the absolute value of the concentration growth rate, making the normalized concentration growth rate reflect only the relative speed of lithium salt enrichment.

[0081] Both are dimensionless indicators after normalization, allowing for unified comparison and synthesis across different batches. This design is based on practical experience from online monitoring: in the early stages of stratification, the interface rapidly moves upward while the heavy phase lithium concentration increases rapidly; both rates are relatively large, and their product forms a significant peak.

[0082] As stratification nears completion, the interface stabilizes and the concentration approaches saturation, with both rates of increase decreasing synchronously, and their product gradually approaching zero. Therefore, this characteristic value can sensitively reflect the overall trend of stratification transitioning from an active phase to an equilibrium phase; the closer to the later stage, the smaller the characteristic value. The stratification process characteristic sequence refers to the numerical sequence formed by arranging the stratification process characteristic values ​​at each sampling time in chronological order. When the interface rate of increase is negative, i.e., when the interface experiences a temporary decrease due to temperature fluctuations, the concentration growth rate at that moment is usually also negative or close to zero. In this case, the product of the two is still positive or zero, and the stratification process characteristic value naturally decreases or remains unchanged, without affecting the overall trend judgment.

[0083] In this step, firstly, in order to quantify the rate of change of interface rise and lithium salt enrichment, data from the current and previous N consecutive sampling times need to be read from the interface height time series and lithium ion concentration time series. Linear regression is performed with the sampling time as the independent variable, and the slope of the regression line is extracted as the interface rise rate and concentration increase rate, so as to grasp the dynamic changes of the two independent dimensions of the stratification process in real time.

[0084] Furthermore, due to differences in the initial interface height and absolute lithium salt concentration among different batches of raw materials, directly using the original rate values ​​cannot provide a unified comparison and judgment between different batches. To eliminate the influence of batch differences on the rate and construct a unified stratification process index, the interface rise rate and concentration increase rate need to be normalized. The interface rise rate is obtained by dividing the interface rise rate by the interface height value at the initial sampling time in the interface height time series; the concentration increase rate is obtained by dividing the concentration increase rate by the maximum lithium ion concentration within the current window in the lithium ion concentration time series. Multiplying the two yields the stratification process characteristic value at the current sampling time.

[0085] Furthermore, the stratification process feature values ​​at each sampling time are arranged in chronological order to form a stratification process feature sequence, which serves as the feature input for the endpoint prediction model. When the stratification process feature values ​​continuously decrease and approach zero, it indicates that the net mass transfer of lithium salt between the two phases has been basically completed.

[0086] Step S400 in the method provided in this application embodiment further includes:

[0087] Obtain stratified process feature sequence samples from historical stratified experiments and stratified process completion labels for each sample at the separation endpoint time. The stratified process completion labels are set to 1.0 when the separation endpoint time is reached. Using the stratified process feature sequence samples as input and the stratified process completion labels as supervised output, construct a machine learning-based endpoint prediction model. Perform supervised training with the goal of minimizing the error between the predicted and labeled values. After training until the validation set error converges, use the trained endpoint prediction model as the endpoint prediction model for outputting the stratified process completion degree. Detailed explanation follows:

[0088] In this embodiment, the stratification process completion degree refers to a value normalized to a maximum of 1.0, representing the completion ratio of the stratification process at the current sampling time relative to the final separation endpoint. The separation endpoint refers to the moment when the interface height no longer changes, the heavy phase lithium ion concentration tends to stabilize, and further extending the settling time has no significant improvement on the enrichment effect. The machine learning is selected from at least one of neural network regression models, gradient boosting tree regression models, or support vector regression models, and can be used for online prediction after training. The model takes the stratification process feature sequence before the current time as input and outputs a predicted stratification process completion degree between 0 and 1, with the value closer to 1 indicating closer proximity to the separation endpoint. The model converges when the validation set loss does not decrease for several consecutive rounds or reaches the preset upper limit of the training rounds.

[0089] In this step, firstly, in order to establish the labeled data required for the endpoint prediction model, it is necessary to collect the stratification process feature sequence samples of each batch from the historical stratification experiment, and label the stratification process completion degree of each sample at the separation endpoint as 1.0.

[0090] Furthermore, to train the endpoint prediction model, a machine learning regression algorithm is used, employing historical stratified process feature sequence samples as input and stratified process completion labels as output for supervised training. After training, the endpoint prediction model is obtained and used for online inference of the completion rate of the current sample.

[0091] Furthermore, after obtaining the feature sequence of the current stratification process each time, it is input into the endpoint prediction model, and the model outputs the current stratification process completion degree. When the stratification process completion degree reaches a preset threshold, it is determined that the separation endpoint has been reached. The preset threshold is typically set between 0.95 and 0.98.

[0092] At this point, the rate of increase in interface concentration and the rate of increase in concentration have usually decreased to less than 5% to 1% of their initial peak values, indicating that the net mass transfer of lithium salt between the two phases has been basically completed. Continuing to extend the time will have limited effect on improving the enrichment amount and residual concentration, while significantly increasing the time cost. Therefore, setting the preset threshold to this range can achieve a balance between separation efficiency and process time.

[0093] The threshold was set based on statistical analysis of stratification experimental data from different batches of waste electrolyte. When the interface rise rate and concentration increase rate decayed to 5% to 1% of their initial peak values, the stratification completion rate corresponded to 0.95 to 0.98. At this point, further extending the settling time would not increase lithium enrichment by more than 2%, and the effect on reducing the residual lithium concentration in the light phase would no longer be significant, indicating that the net mass transfer of lithium salt between the two phases was basically completed. Setting the threshold within this range achieves a balance between separation efficiency and process time, avoiding the increased time cost caused by excessively extending the settling time.

[0094] For example, the pre-training process of the endpoint prediction model is as follows: Multiple sets of stratified process feature sequence samples are collected from historical stratified experiments. Each set of samples contains a stratified process feature sequence representing the complete process from the start of stratification to the endpoint separation. The stratified process completion degree at the endpoint separation time is labeled as 1.0 for each sample. The completion degree of each sampling time before the endpoint is linearly interpolated according to the time ratio between each time and the endpoint time. The samples are divided into training and test sets in an 8:2 ratio.

[0095] The endpoint prediction model employs a gradient boosting tree regression model with 100 trees, a maximum tree depth of 3, a learning rate of 0.1, and a subsampling ratio of 0.8. Supervised training is performed using hierarchical process feature sequence samples as input and hierarchical process completion labels as output. After training, the model is evaluated on the test set, and the correlation coefficient between the predicted completion rate and the labeled completion rate reaches over 0.96, indicating good endpoint prediction accuracy.

[0096] For example, consider the stratified lithium extraction process of a batch of waste lithium battery electrolyte. Taking the data from the 30th sampling time as an example, the interface height and lithium ion concentration values ​​of the current time and the previous 9 consecutive sampling times constitute a time window. Linear regression was performed on the interface height value, and the resulting slope is the interface rise rate of 0.08 mm / min; linear regression was performed on the lithium ion concentration value, and the resulting slope is the concentration increase rate of 0.005 mol / (L·min).

[0097] The peak interface rise rate at the initial stage of stratification in this batch was 0.25 mm / min, and the current sampling time interface rise rate was 0.08 mm / min. The normalized interface rise rate = 0.08 ÷ 0.25 = 0.32. The peak concentration increase rate at the initial stage of stratification was 0.012 mol / (L·min), and the current sampling time concentration increase rate was 0.005 mol / (L·min). The normalized concentration increase rate = 0.005 ÷ 0.012 ≈ 0.42. The stratification process feature value = 0.32 × 0.42 ≈ 0.134. Adding the stratification process feature value at the current sampling time to the stratification process feature sequence and inputting it into the endpoint prediction model, the model outputs a stratification process completion rate of 0.96, reaching the preset threshold of 0.95, thus determining that the separation endpoint has been reached.

[0098] In summary, this step integrates information from two dimensions—interface rise and lithium salt enrichment—into a unified stratification process feature value. It then uses a machine learning model to predict the completion of the stratification process and automatically determine the separation endpoint. This solves the problems of traditional methods that rely solely on experience to determine the endpoint timing or use fixed settling times, leading to incomplete separation or excessively long processing times.

[0099] S500: With the optimization objectives of maximizing the lithium enrichment in the heavy phase and minimizing the residual lithium concentration in the light phase per unit time, a multi-objective fitness function is constructed to iteratively optimize the antisolvent ratio and settling temperature until convergence, thereby obtaining the optimal antisolvent ratio and optimal settling temperature for the current batch and executing the algorithm.

[0100] In this embodiment, after determining the separation endpoint and obtaining stratification data through the endpoint prediction model, significant differences exist in the lithium salt concentration and organic solvent composition of different batches of waste electrolyte. Fixed process parameters cannot guarantee optimal separation for each batch. There is a mutually restrictive relationship between the two objectives of lithium recovery rate and separation efficiency. Increasing the amount of antisolvent helps improve the enrichment of lithium salts, but it also increases the volume of the light phase and solvent recovery costs. Increasing the settling temperature can accelerate the stratification process, but excessively high temperatures may cause the solubility of lithium salts in the carbonate phase to rebound, leading to an increase in the residual lithium concentration in the light phase. Therefore, it is necessary to specifically optimize the antisolvent ratio and settling temperature based on the actual characteristics of each batch of material, automatically finding the optimal balance between lithium enrichment and separation efficiency.

[0101] Step S500 in the method provided in this application embodiment includes:

[0102] The search range for the antisolvent ratio is determined based on the lithium salt concentration, and the search range for the settling temperature is determined based on the initial settling temperature. Multiple candidate combinations of antisolvent ratios and settling temperatures are generated within the search range. For each candidate combination, mixing and settling operations are performed according to the candidate antisolvent ratio and candidate settling temperature. Interface height and lithium ion concentration are continuously collected at the set time intervals until the stratification process completion rate output by the endpoint prediction model reaches the preset threshold. The lithium ion concentration of the lithium-rich heavy phase and the residual lithium concentration of the light phase are recorded at this point, and the lithium enrichment per unit time is determined based on the lithium ion concentration and the time taken to reach the separation endpoint.

[0103] Based on the lithium enrichment amount and light phase residual lithium concentration per unit time, a multi-objective fitness value is calculated, and candidate combinations are non-dominated. Combinations in the non-dominated frontier are selected as parent combinations. The antisolvent ratio and settling temperature in the parent combinations are weighted and randomly perturbed to generate offspring candidate combinations. These offspring candidate combinations replace the parent combinations, and the data acquisition, sorting, and generation operations are repeated until the change in multi-objective fitness value between two adjacent iterations is less than the convergence threshold. The combination currently in the least crowded position in the non-dominated frontier is output as the optimal antisolvent ratio and optimal settling temperature. Detailed explanation follows:

[0104] In this embodiment, the multi-objective fitness function refers to an evaluation criterion that uses lithium enrichment and residual lithium concentration in the light phase per unit time as two independent evaluation indicators to assess the comprehensive performance of each candidate combination of antisolvent ratio and settling temperature. The two objectives correspond to separation efficiency and separation quality, respectively. The antisolvent ratio search range is an adjustable range of volume ratios determined based on lithium salt concentration, with the lower search limit set to the lower limit of the volume ratio corresponding to the lithium salt concentration and the upper search limit set to the upper limit of the volume ratio corresponding to the lithium salt concentration. The settling temperature search range is an adjustable range determined based on the initial settling temperature, with the lower search limit set to the initial settling temperature minus 5°C and the upper search limit set to the initial settling temperature plus 10°C.

[0105] Candidate combinations refer to multiple pairs of antisolvent ratios and settling temperatures generated within the search range. Non-dominated ranking, based on the Pareto optimality principle, categorizes candidate combinations hierarchically according to their dominance; combinations at lower non-dominated fronts are superior to, or at least not inferior to, higher-level fronts in both optimization objectives. The convergence threshold is defined as the change in multi-objective fitness values ​​between two consecutive iterations being less than 0.5% to 1% of the maximum fitness value. This value is chosen because when the change falls to this range, it indicates that the optimization has plateaued, and further iterations offer limited improvement in solution quality, thus controlling computational costs while maintaining optimization accuracy.

[0106] In this step, it is important to note that the iterative optimization experiment is not carried out on the current batch of electrolyte in the main reactor, but rather a small portion is diverted online from the dehydrated electrolyte mainstream as an experimental sample and distributed to each independent channel of the micro multichannel experimental device connected in parallel with the main stratified cell for parallel testing.

[0107] The specific parallel experimental procedure of the micro multichannel experimental device is as follows: Before the start of each generation of iterative evaluation, the control system automatically completes the quantitative injection of antisolvent and temperature control setting of each channel based on the parameter values ​​of each candidate combination in the current population. Specifically, each channel is pre-injected with an equal amount of dehydrated electrolyte diversion sample, and then, according to the corresponding candidate antisolvent ratio, the corresponding volume of pre-cooled antisolvent is synchronously injected into each channel by a multichannel injection pump; after injection, each channel is independently temperature-controlled according to its corresponding candidate settling temperature, and the stirring program is consistent with that of the main layered cell.

[0108] After each channel begins mixing and settling simultaneously, miniature near-infrared spectral probes located at the bottom of each channel and miniature capacitive liquid level sensors on the sidewalls are used to collect real-time time-series data on the interface height and lithium-ion concentration of each channel. When a channel is determined by the endpoint prediction model to have reached the separation endpoint, the lithium-rich heavy phase lithium-ion concentration and the residual lithium concentration in the light phase of that channel are automatically recorded, thus completing the evaluation of the candidate combination.

[0109] After all channels have been evaluated, the data from each group are aggregated into the main control system for non-dominated ranking and generation of next-generation candidate combinations. Throughout the parallel experiment, physical isolation and independent temperature control units ensure that experimental conditions do not interfere with each other. The sample volume for each channel is only a few milliliters, far smaller than the batch volume of the main stratification cell. Therefore, even with multiple iterations, the total sample volume consumed is only a small fraction of the total shunt volume and will not significantly affect the mainstream batch yield.

[0110] After the optimal antisolvent ratio and optimal settling temperature are obtained through optimization convergence, the optimal parameters are fed back in real time and applied to the mixing and settling of the remaining electrolyte in the main reactor. This ensures that the iterative optimization does not interfere with the main batch processing process, achieving both online closed-loop intelligent optimization and ensuring industrial feasibility.

[0111] Furthermore, to find the optimal balance between lithium enrichment and separation efficiency, a multi-objective fitness function needs to be constructed. The search range for the antisolvent ratio is defined by the volume ratio corresponding to the lithium salt concentration, and the search range for the settling temperature is defined by the initial settling temperature plus or minus a certain margin. The initial number of candidate combinations is typically 20 to 50, a value determined empirically based on the population size of the genetic algorithm.

[0112] Furthermore, in order to conduct a practical evaluation of each candidate combination, the dehydrated electrolyte needs to be diverted to each independent channel of the micro multichannel experimental device. The mixing and isothermal standing operations are performed synchronously according to the candidate antisolvent ratio and candidate standing temperature set for each channel. The stratification process of each channel is continuously monitored until the endpoint prediction model determines that the separation endpoint has been reached. At this time, the lithium ion concentration of the lithium-rich heavy phase and the residual lithium concentration of the light phase in each channel are recorded.

[0113] Furthermore, to select parameter combinations that demonstrate excellent separation efficiency and quality from the candidate combinations, a multi-objective fitness value needs to be calculated based on the lithium enrichment amount and the residual lithium concentration in the light phase per unit time. The candidate combinations are then ranked in a non-dominant order, with the combination at the first non-dominant front serving as the parent. Then, a weighted combination with random perturbation is used to generate offspring candidate combinations. The offspring replace the parent in this repeated evaluation and ranking process, approximating the optimal solution generation by generation. Optimization convergence is determined when the change in the multi-objective fitness value between two consecutive iterations is less than the convergence threshold.

[0114] Finally, the combination at the least crowded position in the current non-dominant front is output as the optimal antisolvent ratio and optimal settling temperature. This optimal parameter is fed back and applied to the mixing and settling of the remaining electrolyte in the main reactor.

[0115] Step S500 in the method provided in this application embodiment further includes:

[0116] For any two candidate combinations, compare the lithium enrichment amount and the residual lithium concentration in the light phase per unit time. If the lithium enrichment amount in the first candidate combination per unit time is greater than or equal to the lithium enrichment amount in the second candidate combination per unit time, and the residual lithium concentration in the light phase of the first candidate combination is less than or equal to the residual lithium concentration in the light phase of the second candidate combination, and the value of at least one indicator is better than the corresponding indicator value of the second candidate combination, then it is determined that the first candidate combination dominates the second candidate combination. From all candidate combinations, candidate combinations that are not dominated by any other candidate combinations are selected to form the first non-dominant frontier.

[0117] The combinations already assigned to the first non-dominated frontier are removed from the candidate combinations. From the remaining candidate combinations, a second non-dominated frontier is formed by repeatedly screening for candidate combinations not dominated by any other remaining candidate combinations. This process is repeated layer by layer until all candidate combinations are assigned to their corresponding non-dominated frontiers. For candidate combinations within the same non-dominated frontier, the crowding distance between each candidate combination and its adjacent candidate combinations in the two-dimensional target space composed of lithium enrichment and light-phase residual lithium concentration per unit time is calculated. The crowding distance is measured as the sum of the normalized difference in lithium enrichment and the normalized difference in light-phase residual lithium concentration per unit time. When selecting parent generations, starting from the level with the smallest non-dominated frontier index, candidate combinations with the largest crowding distance are selected first, until the required number of parent generations are selected. A detailed explanation follows:

[0118] In this embodiment, dominance relationship refers to the criterion used to compare the merits of two candidate combinations in multi-objective optimization. When the first candidate combination is not inferior to the second candidate combination on all objectives, and is strictly superior to the second candidate combination on at least one objective, the first candidate combination is determined to dominate the second candidate combination. The non-dominated front refers to the set of all candidate combinations not dominated by any other candidate combination; the smaller the index, the better the front.

[0119] The specific steps for calculating the crowding distance are as follows: First, for all candidate combinations within the same non-dominant front, sort them in descending order of lithium enrichment per unit time, and assign a maximum value to the crowding distance of the first and last candidate combinations to ensure that the diversity of boundary solutions is preserved; Second, for each of the remaining candidate combinations, calculate the absolute difference between its lithium enrichment per unit time and that of its adjacent candidate combinations, as well as the absolute difference in the concentration of residual lithium in the light phase.

[0120] The third step involves dividing the two sets of differences by the difference between the maximum and minimum lithium enrichment amounts per unit time within the non-dominated front, and the difference between the maximum and minimum residual lithium concentrations in the light phase, respectively, to obtain normalized differences. The fourth step involves adding the normalized differences in lithium enrichment amounts per unit time to the normalized differences in residual lithium concentrations in the light phase; the sum is the crowding distance of the candidate combination. A larger crowding distance indicates a sparser distribution of solutions around the candidate combination, and selecting this combination as the parent generation is beneficial for maintaining population diversity and avoiding premature convergence.

[0121] The number of parents is 30% to 50% of the initial number of candidate combinations, or it can be adaptively adjusted according to the number of iterations. In the initial stage, the number of parents is larger to maintain population diversity, and in the later stage of iteration, the number of parents is gradually reduced to accelerate convergence.

[0122] In this step, firstly, to compare the merits of multiple candidate combinations, the dominance relationship needs to be determined pairwise based on two indicators: lithium enrichment per unit time and residual lithium concentration in the light phase. Then, undominated combinations are selected from all candidate combinations to form the first non-dominated frontier, followed by the selection of the second non-dominated frontier from the remaining combinations, and so on, layer by layer. For candidate combinations within the same non-dominated frontier, the crowding distance between each candidate combination and its adjacent combinations in the two-dimensional target space is calculated. When selecting parent generations, the selection starts from the level with the smallest frontier index, prioritizing the candidate combination with the largest crowding distance, until a sufficient number of parent generations are selected, ensuring that the selected parent generations possess both optimality and diversity.

[0123] For example, taking the layered lithium extraction process of a batch of waste lithium battery electrolyte as an example, continuing the previous example. The lithium salt concentration of the current batch of dehydrated electrolyte is 1.2 mol / L, corresponding to an antisolvent ratio search range of 1.2:1 to 2.0:1 (volume ratio); the initial settling temperature is 25℃, and the settling temperature search range is 20℃ to 35℃. Thirty initial candidate combinations are randomly generated within the search range. Each candidate combination is simultaneously mixed and settling at a constant temperature in an independent channel of a micro multi-channel experimental device, with continuous monitoring of the layering process until the endpoint prediction model of each channel determines that the separation endpoint has been reached. The lithium-rich heavy phase lithium ion concentration and the residual lithium concentration of the light phase at the endpoint of each channel are extracted, and the lithium enrichment per unit time is calculated.

[0124] Thirty candidate combinations were ranked non-dominated. Two candidate combinations had the following data: Combination A had a lithium enrichment rate of 0.36 mol / (L·h) and a residual lithium concentration in the light phase of 0.12 mol / L; Combination B had a lithium enrichment rate of 0.30 mol / (L·h) and a residual lithium concentration in the light phase of 0.15 mol / L. Combination A was superior to Combination B in both lithium enrichment and residual lithium concentration, thus A was determined to dominate Combination B. All combinations were then divided strata according to their dominance relationship. The top 15 combinations from the first non-dominated frontier were selected as the parent generation based on their crowding distance.

[0125] The parent generation was weighted and randomly perturbed to generate 30 new candidate combinations, and the evaluation and ranking process was repeated. After several iterations, the maximum fitness change between adjacent iterations decreased to 0.6%, which is below the convergence threshold of 1%, indicating that the optimization had converged. The combination with the largest crowding distance in the current first non-dominated front was output as the optimal solution. The optimal antisolvent ratio was 1.65:1 (volume ratio), and the optimal settling temperature was 28.5℃. These optimal parameters were then fed back to the mixing and settling operations of the remaining electrolyte in the main reactor.

[0126] In summary, this step constructs a multi-objective fitness function with lithium enrichment and light phase residual lithium concentration as dual objectives. It iterative optimization of the antisolvent ratio and settling temperature in a micro multi-channel experimental device achieves adaptive dynamic matching of process parameters to the characteristics of each batch of raw materials. It automatically finds the optimal balance between lithium recovery rate and separation efficiency, solving the problem that fixed process parameters cannot adapt to raw material fluctuations, leading to batch-to-batch instability in recovery rate and product quality.

[0127] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0128] This application proposes a purification process for waste lithium battery electrolyte, achieving liquid-liquid stratification purification and efficient lithium salt recovery of waste lithium battery electrolyte under non-high temperature conditions. By analyzing the key components of the incoming electrolyte and using a combination of 3A molecular sieve adsorption and vacuum evaporation for dehydration, the dehydrated electrolyte and pre-cooled antisolvent are mixed under inert atmosphere and subjected to temperature-controlled stirring and constant-temperature settling to construct a light phase and a lithium-rich heavy phase. The interface height and lithium ion concentration are simultaneously acquired using online capacitive level sensing and near-infrared spectroscopy, and the separation endpoint is determined online. A micro multi-channel experimental device is used to iteratively optimize the antisolvent ratio and settling temperature, and the optimal parameters are fed back to the mainstream batch processing. This effectively improves the technical problems of traditional direct heating distillation, which easily leads to lithium salt decomposition and the generation of toxic gases, and the inability of process parameters to adapt to raw material fluctuations, resulting in low and unstable recovery efficiency.

Claims

1. A purification process for waste lithium battery electrolyte, characterized in that, The process includes: Obtain the initial physical properties of the electrolyte to be treated, and dehydrate the electrolyte to a moisture content of ≤20ppm; Under an inert atmosphere, the dehydrated electrolyte is mixed with a non-aqueous antisolvent according to the initial ratio and kept at the initial settling temperature to form a light phase rich in carbonate solvent on the upper layer and a lithium-rich heavy phase rich in lithium salt on the lower layer. The interface height of the stratified pool is collected at set time intervals. At the same time, the intensity of the characteristic peak of lithium salt in the lower lithium-rich heavy phase is obtained by online spectroscopy and converted into lithium ion concentration. The interface height time series and lithium ion concentration time series are constructed. The interface rise rate and concentration growth rate are calculated to determine the stratification process characteristic value and construct the stratification process characteristic sequence. The stratification process characteristic sequence is input into the endpoint prediction model and the stratification process completion degree is output. When the stratification process completion degree reaches the preset threshold, the separation endpoint is determined to be reached. With the optimization objectives of maximizing the lithium enrichment in the lithium-rich heavy phase and minimizing the residual lithium concentration in the light phase, a multi-objective fitness function is constructed to iteratively optimize the antisolvent ratio and settling temperature until convergence, thereby obtaining the optimal antisolvent ratio and optimal settling temperature for the current batch and executing them.

2. The purification process for waste lithium battery electrolyte according to claim 1, characterized in that, The initial physical properties include at least lithium salt concentration, organic solvent composition, and moisture content. The lithium salt concentration is obtained from the electrolyte to be treated by inductively coupled plasma atomic emission spectrometry (ICP-AES). The organic solvent composition is obtained from the electrolyte to be treated by gas chromatography-mass spectrometry (GC-MS). The moisture content is obtained from the electrolyte to be treated by Karl Fischer method. Dehydration is performed by a combination of 3A molecular sieve adsorption and vacuum evaporation. The amount of 3A molecular sieve used is 2% to 10% of the electrolyte mass. Vacuum evaporation is carried out at a temperature of 30°C to 60°C and an absolute pressure of 0.1 kPa to 5 kPa. Dehydration is performed until the moisture content is ≤20 ppm as determined by Karl Fischer method.

3. The purification process for waste lithium battery electrolyte according to claim 1, characterized in that, Under an inert atmosphere, the dehydrated electrolyte and a non-aqueous antisolvent are mixed according to the initial ratio and allowed to stand at the initial settling temperature to form an upper light phase rich in carbonate solvent and a lower lithium-rich heavy phase enriched with lithium salts, including: Under stirring conditions, the entire amount of the non-aqueous antisolvent is added to the dehydrated electrolyte within 10 to 60 minutes, wherein the system temperature is controlled not to exceed 25°C during the addition process; After the addition is complete, the system temperature is raised to the second temperature value at a heating rate of 0.5℃ / min to 2.0℃ / min, and stirred at the second temperature value for 10min to 30min at a stirring speed of 100rpm to 300rpm, wherein the second temperature value is 5℃ to 20℃ higher than the initial settling temperature. The system temperature is reduced from the second temperature value to the initial settling temperature at a cooling rate of 0.2℃ / min to 1.0℃ / min, and the system is stirred at a stirring speed of 30 rpm to 80 rpm during the cooling process. After the temperature drops to the initial settling temperature, stirring is stopped, and the mixture is kept at the initial settling temperature until a clear two-phase interface appears, thus completing the construction of the light phase and the lithium-rich heavy phase.

4. The purification process for waste lithium battery electrolyte according to claim 3, characterized in that, The non-aqueous antisolvent is selected from at least one of trifluorotoluene or methyl pentafluoropropionate, and the water content of the non-aqueous antisolvent is less than 50 ppm; the volume ratio of the non-aqueous antisolvent to the dehydrated electrolyte is determined by the lithium salt concentration, and when the lithium salt concentration is ≤1.0 mol / L, the volume ratio is 0.8:1 to 1.2:1, and when the lithium salt concentration is >1.0 mol / L, the volume ratio is 1.2:1 to 2.0:1; before adding the non-aqueous antisolvent to the dehydrated electrolyte, it is first diluted with 0. The lithium salt is pre-cooled at a rate from 0.2℃ / min to 0.5℃ / min to a first pre-cooling temperature, which is determined by the lithium salt concentration and the mass fraction of ethylene carbonate in the organic solvent composition. When the lithium salt concentration is ≤1.0 mol / L and the mass fraction of ethylene carbonate is ≤15%, the first pre-cooling temperature is 15℃ to 20℃; when the lithium salt concentration is >1.0 mol / L or the mass fraction of ethylene carbonate is >15%, the first pre-cooling temperature is 5℃ to 15℃.

5. The purification process for waste lithium battery electrolyte according to claim 1, characterized in that, The interface height of the stratified pool was collected at set time intervals. Simultaneously, the characteristic peak intensity of the lithium salt in the lower lithium-rich heavy phase was obtained through online spectroscopy and converted into lithium-ion concentration. Time series sequences of interface height and lithium-ion concentration were constructed, including: An array of capacitive level sensors with equal spacing is arranged along the vertical direction of the stratified pool. The capacitance value of each sensor is read at the set time interval. The interface height is determined based on the abrupt change position of the capacitance value along the vertical direction, and an interface height time sequence is generated. Near-infrared spectroscopy fiber optic probes are installed on the lower sidewall of the layered pool. The absorption spectra of the lower lithium-rich heavy phase at the characteristic wavenumber are collected at the set time intervals. The peak height values ​​of the absorption spectra at the characteristic wavenumber are extracted as the characteristic peak intensities of lithium salts, and a time series sequence of lithium salt characteristic peak intensities is generated. For each sampling time, obtain the N consecutive peak height values ​​before the current time in the time series of lithium salt characteristic peak intensity, and determine whether the coefficient of variation of the N consecutive peak height values ​​is less than 10%, where N is an integer greater than or equal to 3; When the coefficient of variation is less than 10%, the arithmetic mean of the N consecutive peak heights is used as the stability intensity of the lithium salt characteristic peak at the current moment, the lithium ion concentration at the current moment is obtained, and added to the lithium ion concentration time series. When the coefficient of variation is greater than or equal to 10%, the peak value at the current time is removed, the lithium ion concentration at the previous sampling time is used as the lithium ion concentration at the current time, and it is added to the lithium ion concentration time series. The interface height value and lithium ion concentration value at the same sampling time are aligned according to the sampling time to form a time-synchronized interface height time sequence and lithium ion concentration time sequence.

6. The purification process for waste lithium battery electrolyte according to claim 5, characterized in that, The steps for obtaining the coefficient of variation and the lithium-ion concentration at the current moment include: Read the peak height values ​​of the lithium salt characteristic peak intensity time series before the current sampling time, and calculate the arithmetic mean of the N consecutive peak height values; The standard deviation of the N consecutive peak heights is calculated based on the degree of difference between each peak height value and the arithmetic mean. The coefficient of variation is obtained by dividing the standard deviation by the arithmetic mean and then multiplying by 100%. When the coefficient of variation is less than 10%, the arithmetic mean of the N consecutive peak heights is taken as the stable intensity of the lithium salt characteristic peak at the current sampling time. The stable intensity of the lithium salt characteristic peak is substituted into the lithium salt concentration calibration curve to obtain the lithium ion concentration at the current sampling time. The steps for constructing the lithium salt concentration calibration curve include: Prepare at least three standard solutions with different known lithium ion concentrations, wherein the matrix of the standard solutions is prepared by mixing trifluorotoluene or methyl pentafluoropropionate with carbonate solvents in the relative proportion of carbonate solvents in the organic solvent composition; Absorption spectra of each standard solution were collected at the characteristic wavenumber, and the peak height values ​​of each standard solution at the characteristic wavenumber were extracted. A linear regression was performed with the peak height of each standard solution on the x-axis and the corresponding known lithium ion concentration on the y-axis to obtain the lithium salt concentration calibration curve. When the coefficient of variation is greater than or equal to 10%, the peak value at the current sampling time is removed, and the lithium ion concentration at the previous sampling time is used as the lithium ion concentration at the current sampling time. The lithium ion concentration at the current sampling time is added to the lithium ion concentration time sequence to complete the lithium ion concentration update at the current sampling time.

7. The purification process for waste lithium battery electrolyte according to claim 1, characterized in that, Calculate the interface rise rate and concentration growth rate, determine the stratification process characteristic values, and construct the stratification process characteristic sequence, including: Read the interface height values ​​at the current sampling time and the N consecutive sampling times preceding the current sampling time from the interface height time sequence; and read the lithium ion concentration values ​​corresponding to the N sampling times from the lithium ion concentration time sequence. Using the sampling time as the independent variable and the interface height as the dependent variable, a linear regression is performed on N interface height values, and the slope of the straight line obtained from the linear regression is taken as the interface rise rate. Using the sampling time as the independent variable and the lithium ion concentration as the dependent variable, a linear regression was performed on N lithium ion concentration values, and the slope of the straight line obtained from the linear regression was taken as the concentration growth rate. The interface rise rate is divided by the interface height value at the initial sampling time in the interface height time series to obtain the normalized interface rise rate, and the concentration increase rate is divided by the maximum lithium ion concentration in the current window in the lithium ion concentration time series to obtain the normalized concentration increase rate. Multiply the normalized interface rise rate by the normalized concentration increase rate to obtain the stratification process characteristic value at the current sampling time; The hierarchical process feature value at the current sampling time is added to the hierarchical process feature sequence to complete the construction of the hierarchical process feature sequence.

8. The purification process for waste lithium battery electrolyte according to claim 1, characterized in that, The steps for constructing the endpoint prediction model include: Obtain the stratification process feature sequence samples of historical stratification experiments and the stratification process completion label value of the separation endpoint time corresponding to each sample, wherein the stratification process completion label value is labeled with 1.0 when the separation endpoint time is reached; Using the hierarchical process feature sequence samples as input and the hierarchical process completion label value as supervision output, a machine learning-based endpoint prediction model is constructed. Supervised training is conducted with the goal of minimizing the error between the predicted and labeled values. After training until the error on the validation set converges, the trained endpoint prediction model is used as the endpoint prediction model for outputting the hierarchical process completion degree.

9. The purification process for waste lithium battery electrolyte according to claim 1, characterized in that, With the optimization objectives of maximizing the lithium enrichment in the lithium-rich heavy phase and minimizing the residual lithium concentration in the light phase, a multi-objective fitness function is constructed. The antisolvent ratio and settling temperature are iteratively optimized until convergence, yielding the optimal antisolvent ratio and optimal settling temperature for the current batch, which are then executed. This includes: The search range for determining the antisolvent ratio is based on the lithium salt concentration, and the search range for the settling temperature is based on the initial settling temperature. Multiple candidate combinations of antisolvent ratios and settling temperatures are generated within the search range; For each candidate combination, mixing and standing operations are performed according to the candidate antisolvent ratio and candidate standing temperature. The interface height and lithium ion concentration are continuously collected at the set time interval until the stratification process completion rate output by the endpoint prediction model reaches the preset threshold. Record the lithium ion concentration in the lithium-rich heavy phase and the residual lithium concentration in the light phase at this time, and determine the amount of lithium enrichment per unit time based on the lithium ion concentration and the time taken to reach the separation endpoint. Based on the lithium enrichment amount and light phase residual lithium concentration per unit time, the multi-objective fitness value is calculated, and the candidate combinations are ranked in a non-dominant order. The combination in the non-dominant front is selected as the parent generation. The antisolvent ratio and settling temperature in the parent combination are weighted and random perturbed to generate candidate offspring combinations. Replace the parent combination with the offspring candidate combination, and repeat the collection, recording, sorting and selection and generation operations until the change in the multi-objective fitness value between two adjacent iterations is less than the convergence threshold. Output the combination in the least crowded position in the current non-dominated front as the optimal antisolvent ratio and the optimal standing temperature.

10. The purification process for waste lithium battery electrolyte according to claim 9, characterized in that, Based on the lithium enrichment amount and light phase residual lithium concentration per unit time, a multi-objective fitness value is calculated, and candidate combinations are ranked in a non-dominated order. Combinations in the non-dominated frontier are selected as the parent generation, including: For any two candidate combinations, compare the lithium enrichment amount and the residual lithium concentration in the light phase per unit time. If the lithium enrichment amount in the first candidate combination per unit time is greater than or equal to the lithium enrichment amount in the second candidate combination per unit time, and the residual lithium concentration in the light phase of the first candidate combination is less than or equal to the residual lithium concentration in the light phase of the second candidate combination, and the value of at least one indicator is better than the corresponding indicator value of the second candidate combination, then it is determined that the first candidate combination dominates the second candidate combination. The candidate combinations that are not dominated by any other candidate combinations are selected from all candidate combinations and constitute the first non-dominated frontier; Remove the combinations that have been assigned to the first non-dominated front from the candidate combinations. Repeat the screening of candidate combinations that are not dominated by any other remaining candidate combinations to form the second non-dominated front. Continue this process layer by layer until all candidate combinations are assigned to the corresponding non-dominated front. For candidate combinations within the same non-dominant frontier, calculate the crowding distance between each candidate combination and its neighboring candidate combinations in a two-dimensional target space composed of lithium enrichment and light phase residual lithium concentration per unit time. The crowding distance is measured by the sum of the normalized difference in lithium enrichment and the normalized difference in light phase residual lithium concentration per unit time. When selecting the parent generation, start from the level with the smallest non-dominant frontier index and prioritize the candidate combination with the largest crowding distance until the required number of parent generations are selected.

Images