Battery recycling material separation and purification process optimization method based on swarm intelligence algorithm

Abstract

The application provides a battery recycling material separation and purification process optimization method based on a swarm intelligence algorithm, relates to the technical field of battery recycling, and comprises the following steps: constructing an optimization target based on recycling material composition data, determining a multi-level constraint boundary of process parameters according to separation equipment operation state data; segmentally analyzing historical process parameter data according to a process flow time sequence, extracting parameter-target correlation degrees of each process segment through statistical analysis, decomposing the constraint boundary into a plurality of sub-constraint boundaries according to the parameter-target correlation degrees, and obtaining a simplified parameter search domain; performing segmental and parallel optimization in the simplified parameter search domain by using the swarm intelligence algorithm, iteratively updating various swarm individuals in the sub-constraint boundaries, combining the various swarm individuals according to a process segment order to form a candidate scheme, and calculating a separation purity value; establishing a transfer constraint relationship of material states among the process segments, performing process continuity determination, calculating an adaptability value based on the separation purity value, and obtaining an optimal process.

Description

Technical Field

[0001] This invention relates to the field of battery recycling technology, and in particular to an optimization method for the separation and purification process of battery recycled materials based on swarm intelligence algorithms. Background Technology

[0002] With the rapid development of new energy vehicles and portable electronic devices, the recycling of used batteries has become an important issue in resource recycling. The separation and purification process of battery recycled materials is a key step in achieving efficient recovery of valuable metals, and the optimization of its process parameters directly affects the purity, recovery rate, and production cost of the recycled products. Traditional battery recycled material separation and purification processes mainly include multiple stages such as physical separation, chemical leaching, solvent extraction, and precipitation separation. These stages involve complex material transfer and state transition relationships.

[0003] Existing process parameter optimization methods typically treat the entire separation and purification process as a whole, failing to fully consider the differences in characteristics and parameter influence patterns among different process stages. Battery recyclable material separation and purification involves multiple process stages connected in series. The process mechanisms, key influencing factors, and parameter sensitivities of each stage differ significantly. Simple global optimization methods result in an excessively large parameter search space, low optimization efficiency, and difficulty in converging to a satisfactory solution within a reasonable timeframe. Summary of the Invention

[0004] This invention provides an optimization method for the separation and purification process of battery recycled materials based on swarm intelligence algorithms, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for optimizing the separation and purification process of battery recycled materials based on swarm intelligence algorithms, comprising:

[0006] The optimization objective is constructed based on the composition data of the battery recycled material, and the multi-level constraint boundary of the process parameters is determined based on the operating status data of the separation equipment.

[0007] The historical process parameter data of the separation and purification process are segmented and analyzed according to the process flow sequence. By statistically analyzing the impact of process parameter changes in each process segment on the optimization objective, the parameter-objective correlation of each process segment is extracted. Based on the parameter-objective correlation, the constraint boundary is decomposed into multiple sub-constraint boundaries to obtain the simplified parameter search domain corresponding to each process segment.

[0008] A swarm intelligence algorithm is used to perform segmented parallel optimization within the simplified parameter search domain. Population individuals are initialized for each process segment, and each population individual is iteratively updated within the sub-constraint boundary. In each iteration, the population individuals are combined across segments according to the process segment order to form candidate process parameter schemes for the entire process and the process effect is calculated to obtain the separation purity value corresponding to each candidate process parameter scheme.

[0009] Establish the transfer constraint relationship of material state between process sections, determine the process continuity of each candidate process parameter scheme based on the transfer constraint relationship, calculate the fitness value of each candidate process parameter scheme based on the separation purity value, and obtain the battery recycled material separation and purification process with the optimal fitness value.

[0010] The optimization objective is constructed based on the composition data of the battery recycled material, and the multi-level constraint boundaries of the process parameters are determined based on the operating status data of the separation equipment, including:

[0011] The impurity element content ratio and particle size distribution characteristics are extracted from the battery recycled material composition data; a composite operation of reciprocal transformation and logarithmic transformation is performed on the impurity element content ratio to construct a material loss rate sub-objective; kernel density is calculated on the particle size distribution characteristics, and kurtosis coefficient and skewness coefficient are extracted from the calculation results to construct an energy efficiency sub-objective; the material loss rate sub-objective and the energy efficiency sub-objective are sampled from the target space using Pareto front sampling, and the target space is divided into multiple sub-regions using the reference point decomposition method to obtain a multi-level decomposition optimization objective;

[0012] The equipment energy consumption curve, equipment fault records, and equipment maintenance logs are extracted from the operating status data. A Fourier transform is performed on the equipment energy consumption curves to extract the dominant frequency component and harmonic components from the energy consumption frequency domain features. Based on the dominant frequency component and the harmonic components, the safe operating ranges for temperature control parameters and flow rate control parameters are determined. Fault precursor parameter combination patterns are extracted from the equipment fault records and converted into prohibited ranges for timing control parameters. Text semantic analysis is performed on the equipment maintenance logs to extract the applicable ranges of the parameters. The intersection of the applicable ranges of the parameters with the safe operating ranges and the prohibited ranges is calculated to obtain the multi-level constraint boundaries.

[0013] Historical process parameter data for the separation and purification process are segmented and analyzed according to the process flow sequence. The impact of process parameter changes within each process segment on the optimization objective is statistically analyzed, and the parameter-objective correlation for each process segment is extracted, including:

[0014] Based on the time nodes of material state transformation in the battery recycling material separation and purification process, the historical process parameter data is divided into multiple process time segments, and the time series of temperature parameters, flow rate parameters, and time series of time control parameters are extracted from each process time segment.

[0015] Within each process time segment, a sliding window difference operation is performed on the temperature parameter time series to calculate the local rate of change sequence of the temperature parameter; wavelet decomposition operation is performed on the flow velocity parameter time series to extract the multi-scale fluctuation feature sequence of the flow velocity parameter; a state transition matrix is ​​constructed on the time series of the time series control parameters; and the mode switching frequency sequence of the time series parameters is extracted from the state transition matrix.

[0016] The local rate of change sequence, the multi-scale fluctuation characteristic sequence, and the mode switching frequency sequence are respectively subjected to mutual information calculation with the material loss rate sub-objective and the energy consumption efficiency sub-objective to construct a mutual information matrix. Principal component analysis is performed on the mutual information matrix to reduce dimensionality and extract the principal component vectors whose cumulative contribution rate reaches a preset contribution rate threshold. The values ​​of each element of the principal component vectors are used as quantitative indicators of the correlation strength between the corresponding process parameters and the optimization target to obtain the parameter-target correlation degree for each process time segment.

[0017] Based on the parameter-target correlation, the constraint boundary is decomposed into multiple sub-constraint boundaries, resulting in a simplified parameter search domain for each process segment, including:

[0018] The correlation values ​​of temperature parameter, flow rate parameter, and time sequence control parameter in the parameter-target correlation are sorted in descending order, and the optimization priority sequence of each process parameter in the corresponding process time segment is determined according to the sorting order.

[0019] The global constraint ranges of temperature control parameters, flow rate control parameters, and timing control parameters in the multi-level constraint boundary are obtained and mapped to each process timing segment to obtain the parameter correlation vector of each process timing segment; the parameter correlation vector is then used to perform interval shrinkage operation with a preset constraint shrinkage coefficient to obtain the parameter sub-constraint range of each process timing segment.

[0020] The parameter sub-constraint ranges are reorganized according to the optimization priority sequence. The parameter sub-constraint range with the highest optimization priority is used as the main search dimension, and the remaining parameter sub-constraint ranges are used as auxiliary search dimensions. The simplified parameter search domain corresponding to each process time segment is constructed by performing a Cartesian product operation between the main search dimension and the auxiliary search dimensions.

[0021] A swarm intelligence algorithm is used to perform segmented parallel optimization within the simplified parameter search domain. The population individuals are initialized separately for each process segment, including:

[0022] The material loss rate sub-objective and the energy efficiency sub-objective are subjected to Chebyshev decomposition to generate uniformly distributed reference vectors and calculate the Euclidean distance between the reference vectors and the origin of the target space. The reference vectors are divided into multiple reference vector subsets according to the Euclidean distance values ​​and sorted in ascending order. The reference vector subsets are mapped to the corresponding process time segments according to the time position of each process time segment.

[0023] Within the process time segment, density peak clustering is performed on the reference vector subset to obtain cluster center scores. Based on the cluster center scores, the population size of each process time segment is weighted and allocated to obtain the allocated population size for each process time segment. Within the parameter sub-constraints of the process parameters, random numbers following a uniform distribution are generated as positional components of temperature, flow rate, and time sequence. The positional components are concatenated to form a random position vector, and the inverse positional components of the positional components are concatenated to form an inverse position vector.

[0024] The random position vector and the reverse position vector are concatenated and non-dominated sorting is performed. Pareto front individuals are selected according to the non-dominated level. From the Pareto front individuals, a corresponding number of individual position vectors are selected according to the allocated population size to complete the population individual initialization for each process time segment.

[0025] The individuals in the population are combined across different process segments according to the sequence of the process segments to form candidate process parameter schemes for the entire process. The process effect is then calculated to obtain the separation purity value corresponding to each candidate process parameter scheme, including:

[0026] Obtain the set of population individual position vectors within the current iteration period, perform velocity update and position update operations on the set of population individual position vectors, and calculate the Euclidean distance between the vectors to obtain the individual distance matrix. Based on the individual distance matrix, construct a two-dimensional projection space, perform density clustering on the projected population individual position points in the two-dimensional projection space, and identify the center point of the density region as the representative population individual position vector.

[0027] Representative population individual position vectors of adjacent process time segments are selected for cross-segment combination, and the material state transmission continuity score value after cross-segment combination is calculated. Cross-segment combination schemes with material state transmission continuity score values ​​higher than a preset continuity threshold are screened to obtain a set of candidate process parameter schemes.

[0028] The temperature parameter setpoint, flow rate parameter setpoint, and time sequence parameter setpoint from the candidate process parameter scheme set are used as process parameter input tensors. Multi-layer convolution feature extraction and fully connected layer mapping are performed on the process parameter input tensors to obtain the predicted mass concentration of the target substance and the predicted mass concentration of the impurity residue, and the ratio is calculated to obtain the separation purity value of the candidate process parameter scheme.

[0029] Establish transitive constraints on material states between process sections, determine the process continuity of each candidate process parameter scheme based on the transitive constraints, and calculate the fitness value of each candidate process parameter scheme based on the separation purity value, including:

[0030] The target space is divided into multiple discrete intervals. The distribution frequencies of the output material state parameters of the preceding process time segment and the input material state parameters of the subsequent process time segment are statistically analyzed in each discrete interval. The information entropy value is calculated based on the distribution frequency, and the material state mutual information value is obtained based on the information entropy value. The KL divergence value is obtained by performing a logarithmic ratio operation on the distribution frequency interval.

[0031] Based on the mutual information value of the material state, a mutual information weight coefficient is constructed, and the KL divergence value is weighted using the mutual information weight coefficient to establish the transmission constraint relationship of material state between process sections.

[0032] Extract parameter setting values ​​from the candidate process parameter scheme set, predict the output material state parameter distribution for each process time segment based on the parameter setting values, input the transitive constraint relationship, and calculate the overall KL divergence value corresponding to the candidate process parameter scheme;

[0033] The process continuity penalty factor is obtained by calculating the ratio of the overall KL divergence value to the preset KL divergence threshold. The fitness value of the candidate process parameter scheme is obtained by multiplying the separation purity value by the exponential function value of the process continuity penalty factor.

[0034] A second aspect of the present invention provides a battery recyclable material separation and purification process optimization system based on swarm intelligence algorithms, comprising:

[0035] The first unit is used to construct optimization objectives based on battery recycled material composition data and determine multi-level constraint boundaries of process parameters based on the operating status data of the separation equipment.

[0036] The second unit is used to segment and analyze the historical process parameter data of the separation and purification process according to the process flow sequence. By statistically analyzing the impact of the changes in process parameters in each process segment on the optimization target, the parameter-target correlation of each process segment is extracted. Based on the parameter-target correlation, the constraint boundary is decomposed into multiple sub-constraint boundaries to obtain the simplified parameter search domain corresponding to each process segment.

[0037] The third unit is used to perform segmented parallel optimization in the simplified parameter search domain using a swarm intelligence algorithm. It initializes population individuals for each process segment, and iteratively updates each population individual within the sub-constraint boundary. In each iteration, the population individuals are combined across segments according to the process segment order to form candidate process parameter schemes for the entire process and calculate the process effect to obtain the separation purity value corresponding to each candidate process parameter scheme.

[0038] The fourth unit is used to establish the transfer constraint relationship of material state between process sections, determine the process continuity of each candidate process parameter scheme based on the transfer constraint relationship, calculate the fitness value of each candidate process parameter scheme based on the separation purity value, and obtain the battery recycled material separation and purification process with the optimal fitness value.

[0039] A third aspect of the embodiments of the present invention,

[0040] An electronic device is provided, comprising:

[0041] processor;

[0042] Memory used to store processor-executable instructions;

[0043] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0044] Fourth aspect of the present invention,

[0045] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0046] The beneficial effects of this application are as follows:

[0047] This invention constructs optimization objectives based on battery recycled material composition data and determines multi-level constraint boundaries of process parameters based on the operating status data of separation equipment. It can establish an optimization model that meets actual production needs based on the actual composition characteristics of battery recycled materials and equipment operating conditions, making the optimization results more in line with actual production, improving the pertinence and feasibility of process optimization, and avoiding the problem in traditional methods where optimization results are difficult to implement due to ignoring material characteristics and equipment status.

[0048] This invention analyzes historical process parameter data of the separation and purification process in segments according to the process flow sequence, statistically analyzes the impact of process parameter changes on the optimization objective within each process segment, extracts the parameter-objective correlation of each process segment, and decomposes the constraint boundary into multiple sub-constraint boundaries based on the parameter-objective correlation, thereby obtaining the simplified parameter search domain corresponding to each process segment. This effectively reduces the dimensionality and complexity of the parameter search space, enabling swarm intelligence algorithms to perform more efficient optimization within the simplified search domain, significantly improving the convergence speed and optimization efficiency of the algorithm, and shortening the computation time required for process optimization.

[0049] This invention establishes a constraint relationship for the transfer of material states between process segments and determines the process continuity of each candidate process parameter scheme. This ensures the rationality of parameter matching and material state connection between process segments, avoids the process discontinuity problem caused by segmented optimization, and guarantees the operability and stability of the optimized whole-process process parameter scheme in actual production. This improves the overall effect and product purity of the battery recycled material separation and purification process. Attached Figure Description

[0050] Figure 1 This is a schematic flowchart of the battery recycling material separation and purification process optimization method based on swarm intelligence algorithm according to an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of the process for analyzing the composition of battery recycled materials and determining equipment constraint boundaries. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0054] Figure 1 This is a schematic flowchart of the battery recycling material separation and purification process optimization method based on swarm intelligence algorithm according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0055] The optimization objective is constructed based on the composition data of the battery recycled material, and the multi-level constraint boundary of the process parameters is determined based on the operating status data of the separation equipment.

[0056] The historical process parameter data of the separation and purification process are segmented and analyzed according to the process flow sequence. By statistically analyzing the impact of process parameter changes in each process segment on the optimization objective, the parameter-objective correlation of each process segment is extracted. Based on the parameter-objective correlation, the constraint boundary is decomposed into multiple sub-constraint boundaries to obtain the simplified parameter search domain corresponding to each process segment.

[0057] A swarm intelligence algorithm is used to perform segmented parallel optimization within the simplified parameter search domain. Population individuals are initialized for each process segment, and each population individual is iteratively updated within the sub-constraint boundary. In each iteration, the population individuals are combined across segments according to the process segment order to form candidate process parameter schemes for the entire process and the process effect is calculated to obtain the separation purity value corresponding to each candidate process parameter scheme.

[0058] Establish the transfer constraint relationship of material state between process sections, determine the process continuity of each candidate process parameter scheme based on the transfer constraint relationship, calculate the fitness value of each candidate process parameter scheme based on the separation purity value, and obtain the battery recycled material separation and purification process with the optimal fitness value.

[0059] In one optional implementation, the optimization objective is constructed based on the composition data of the recycled battery material, and the multi-level constraint boundaries of the process parameters are determined based on the operating status data of the separation equipment, including:

[0060] The impurity element content ratio and particle size distribution characteristics are extracted from the battery recycled material composition data; a composite operation of reciprocal transformation and logarithmic transformation is performed on the impurity element content ratio to construct a material loss rate sub-objective; kernel density is calculated on the particle size distribution characteristics, and kurtosis coefficient and skewness coefficient are extracted from the calculation results to construct an energy efficiency sub-objective; the material loss rate sub-objective and the energy efficiency sub-objective are sampled from the target space using Pareto front sampling, and the target space is divided into multiple sub-regions using the reference point decomposition method to obtain a multi-level decomposition optimization objective;

[0061] The equipment energy consumption curve, equipment fault records, and equipment maintenance logs are extracted from the operating status data. A Fourier transform is performed on the equipment energy consumption curves to extract the dominant frequency component and harmonic components from the energy consumption frequency domain features. Based on the dominant frequency component and the harmonic components, the safe operating ranges for temperature control parameters and flow rate control parameters are determined. Fault precursor parameter combination patterns are extracted from the equipment fault records and converted into prohibited ranges for timing control parameters. Text semantic analysis is performed on the equipment maintenance logs to extract the applicable ranges of the parameters. The intersection of the applicable ranges of the parameters with the safe operating ranges and the prohibited ranges is calculated to obtain the multi-level constraint boundaries.

[0062] like Figure 2 As shown, the method includes:

[0063] The composition data of the battery recycled materials was obtained by analyzing the mixed recycled materials using X-ray fluorescence spectrometry. This analysis was conducted in a vacuum environment for 300 seconds to obtain the mass percentage data of each element. In a specific example, a batch of recycled materials contained 23.5% aluminum, 18.7% copper, 12.3% iron, 8.9% carbon, and 6.2% other impurity elements such as silicon, magnesium, and zinc. After normalizing these impurity element content values ​​according to their respective proportions, a standardized impurity element content ratio vector was obtained, which contained data in seven dimensions.

[0064] The particle size distribution characteristics of the material were collected using a laser particle size analyzer. The equipment scanned from 0.1 micrometers to 3000 micrometers, with a sampling frequency of 50 times per second and a duration of 120 seconds. The particle size distribution of a batch of recycled material showed that particles smaller than 50 micrometers accounted for 15.3%, particles between 50 and 200 micrometers accounted for 42.7%, particles between 200 and 800 micrometers accounted for 31.5%, and particles larger than 800 micrometers accounted for 10.5%. This distribution data formed a discrete distribution function containing 128 particle size intervals.

[0065] The reciprocal of the content ratio of each impurity element was used to convert the value from the content domain to the purity domain. Specifically, the reciprocal of aluminum was 4.255, copper was 5.348, and iron was 8.130. A logarithmic transformation was then performed on these reciprocals, using the natural logarithm as the transformation benchmark. The logarithmic transformation results were 1.448 for aluminum, 1.676 for copper, and 2.095 for iron. These transformed values ​​were then used to construct a sub-target for material loss rate using a weighted summation method. The weighting coefficients were determined based on the recovery value of each element: aluminum had a weight of 0.35, copper 0.45, and iron 0.20, resulting in a weighted sum of 1.623.

[0066] A Gaussian kernel function was used as the smoothing kernel to calculate the kernel density of the material particle size distribution characteristics. The bandwidth parameter was set to 1.06 times the standard deviation of the particle size range. The calculation process mapped the discrete particle size distribution data to a continuous probability density space, generating a smooth density curve. This curve reached a peak of 0.0087 at a particle size of 150 micrometers on the horizontal axis and a secondary peak of 0.0063 at 250 micrometers. The kurtosis coefficient was extracted from this density curve and calculated as the ratio of the fourth central moment of the curve to the fourth power of the standard deviation. Subtracting 3 yielded a super-valued kurtosis coefficient of 0.87, indicating a slight peaked distribution. The skewness coefficient was calculated as the ratio of the third central moment to the cube of the standard deviation, resulting in -0.23, indicating a slightly left-skewed distribution. These two coefficients were linearly combined to construct an energy efficiency sub-objective, with combination coefficients of 0.6 and 0.4, respectively, yielding a calculated value of 0.43.

[0067] The material loss rate sub-objective and the energy efficiency sub-objective form a distribution point set in the two-dimensional target space. Fifty initial sampling points are selected and evenly distributed in the target space. Each sampling point corresponds to a set of process parameter configurations. The Pareto front is identified using a fast non-dominated sorting algorithm, which contains 17 non-dominated solutions. A reference point decomposition method is used, setting three reference points: the first reference point has coordinates of material loss rate 1.2 and energy efficiency 0.3; the second reference point has coordinates of material loss rate 1.8 and energy efficiency 0.5; and the third reference point has coordinates of material loss rate 2.1 and energy efficiency 0.7. Based on these reference points, the target space is divided into three sub-regions, each associated with a different optimization focus strategy, forming a multi-level decomposed optimization objective.

[0068] Equipment operating status data is collected in real time from the sensor network of the isolated equipment, including temperature sensors, pressure sensors, flow sensors, and power sensors. The energy consumption curve data within the equipment operating status data records the total power consumption of the equipment at a frequency of 10 sampling points per second, continuously recording for 7200 seconds to form 72,000 data points. The energy consumption curve for a typical operating cycle shows that during the equipment startup phase, the power rises from zero to 45 kW in 180 seconds; during the stable operation phase, the power fluctuates between 42 and 48 kW; and during the shutdown phase, the power drops to zero in 90 seconds.

[0069] The Fourier transform of the energy consumption curve converts the time-domain signal to the frequency-domain space, yielding the amplitude and phase spectra. The amplitude spectrum shows that the dominant frequency component is located at 0.0167 Hz, corresponding to a periodic fluctuation of 60 seconds, with an amplitude of 3.2 kW. The first harmonic component is located at 0.0333 Hz with an amplitude of 1.5 kW, and the second harmonic component is located at 0.05 Hz with an amplitude of 0.8 kW. The dominant frequency component reflects the basic operating rhythm of the equipment, corresponding to a safe operating range of 285°C to 315°C for the temperature control parameters. The harmonic components reveal the flow rate fluctuation characteristics, determining a safe operating range of 1.2 cubic meters per minute to 1.8 cubic meters per minute for the flow rate control parameters.

[0070] The equipment failure record database stores 274 failure events that occurred over the past three years. Each event record includes the time of failure, failure type, failure duration, and a parameter snapshot taken 30 minutes prior to the failure. Analysis of these records using association rule mining algorithms identified high-frequency failure precursor parameter combinations. The first pattern is a temperature exceeding 320 degrees Celsius and a pressure below 0.8 MPa, which occurs on average 12 minutes before a failure. The second pattern is a flow rate fluctuation exceeding 15% of the nominal value for a duration exceeding 180 seconds. These patterns were converted into prohibited ranges for timing control parameters: a prohibited range for temperature above 320 degrees Celsius, a prohibited range for pressure below 0.8 MPa, and a prohibited range for flow rate fluctuation exceeding 15% of the nominal value.

[0071] The equipment maintenance log, in free text format, records engineers' operation records and experience summaries, totaling 1580 log entries. Key terms were extracted using the term frequency-inverse document frequency method for text semantic analysis, identifying 32 high-frequency terms such as "temperature range," "flow rate setting," and "pressure threshold." Dependency parsing was used to extract parameter-value pairing relationships, resulting in 127 sets of valid parameter application ranges. Specific examples include "maintaining the temperature between 290 and 310 degrees Celsius yields the best results," "the flow rate should not exceed 1.7 cubic meters per minute," and "maintaining the pressure between 1.0 and 1.3 MPa ensures stable equipment operation." These ranges are summarized as follows: the applicable temperature parameter range is 290 to 310 degrees Celsius, the applicable flow rate parameter range is 1.2 to 1.7 cubic meters per minute, and the applicable pressure parameter range is 1.0 to 1.3 MPa.

[0072] The intersection calculation of the applicable range, safe operating range, and prohibited range of parameters is performed. An interval intersection algorithm is used to process each control parameter one by one. The intersection calculation process for temperature parameters is as follows: the applicable range of 290 to 310 degrees Celsius is intersected with the safe operating range of 285 to 315 degrees Celsius to obtain 290 to 310 degrees Celsius. Excluding the prohibited range above 320 degrees Celsius, the multi-level temperature constraint boundary is determined to be 290 to 310 degrees Celsius. The intersection calculation for flow velocity parameters is as follows: the applicable range of 1.2 to 1.7 cubic meters per minute is intersected with the safe operating range of 1.2 to 1.8 cubic meters per minute to obtain 1.2 to 1.7 cubic meters per minute. After excluding dynamic constraints with fluctuations exceeding 15%, the multi-level flow velocity constraint boundary is determined to be 1.2 to 1.7 cubic meters per minute with a fluctuation range not exceeding 15% of the nominal value. The multi-level constraint boundary for pressure parameters is determined through similar calculations to be a safety baseline of 1.0 to 1.3 MPa, not lower than 0.8 MPa.

[0073] In one optional implementation, historical process parameter data of the separation and purification process are segmented and analyzed according to the process flow sequence. By statistically analyzing the impact of process parameter changes within each process segment on the optimization objective, the parameter-objective correlation of each process segment is extracted, including:

[0074] Based on the time nodes of material state transformation in the battery recycling material separation and purification process, the historical process parameter data is divided into multiple process time segments, and the time series of temperature parameters, flow rate parameters, and time series of time control parameters are extracted from each process time segment.

[0075] Within each process time segment, a sliding window difference operation is performed on the temperature parameter time series to calculate the local rate of change sequence of the temperature parameter; wavelet decomposition operation is performed on the flow velocity parameter time series to extract the multi-scale fluctuation feature sequence of the flow velocity parameter; a state transition matrix is ​​constructed on the time series of the time series control parameters; and the mode switching frequency sequence of the time series parameters is extracted from the state transition matrix.

[0076] The local rate of change sequence, the multi-scale fluctuation characteristic sequence, and the mode switching frequency sequence are respectively subjected to mutual information calculation with the material loss rate sub-objective and the energy consumption efficiency sub-objective to construct a mutual information matrix. Principal component analysis is performed on the mutual information matrix to reduce dimensionality and extract the principal component vectors whose cumulative contribution rate reaches a preset contribution rate threshold. The values ​​of each element of the principal component vectors are used as quantitative indicators of the correlation strength between the corresponding process parameters and the optimization target to obtain the parameter-target correlation degree for each process time segment.

[0077] Based on the actual operation of the battery recycling material separation and purification process, the division of process time segments is determined by monitoring changes in the physical state and chemical properties of the materials. In actual operation, the time point when the material changes from a solid slurry to a dissolved liquid is recorded as the first segment point; when the dissolved liquid enters the precipitation reaction stage, it is recorded as the second segment point; when the precipitate enters the filtration separation stage, it is recorded as the third segment point; and when the filtrate enters the concentration and crystallization stage, it is recorded as the fourth segment point. In this way, the historical data of the entire process flow is divided into four process time segments: dissolution, precipitation, filtration, and crystallization. Each time segment contains all process parameter data recorded for that stage. For example, if the dissolution stage lasts 120 minutes and the sampling frequency is once per minute, then this segment contains 120 data sampling points for temperature, flow rate, and time-series control parameter values.

[0078] For processing time series data of temperature parameters, a sliding window with a length of 5 sampling points is set, and the difference operation is performed by moving one sampling point at a time. Specifically, the temperature value of the last sampling point within the window is subtracted from the temperature value of the first sampling point, and then divided by the time interval to obtain the temperature change rate for that window. For example, if the temperatures of the 5 sampling points within a window during the melting period are 75°C, 76°C, 78°C, 79°C, and 81°C respectively, and the time span is 4 minutes, then the local change rate for that window is 81 minus 75, divided by 4, yielding a change rate of 1.5°C per minute. The window is then slid forward sequentially to calculate the local change rate sequence for the entire time period. This sequence reflects the dynamic changes of the temperature parameter at different times.

[0079] For processing the time series of flow velocity parameters, multi-level wavelet decomposition was used to decompose the original flow velocity signal into sub-signals of different frequency scales. A three-level decomposition structure was selected. The first level decomposed the original signal into low-frequency approximate components and high-frequency detail components. The low-frequency components reflected the overall trend change of the flow velocity, while the high-frequency components reflected the rapid fluctuations in the flow velocity. The second level further decomposed the low-frequency components of the first level, and the third level further decomposed the low-frequency components of the second level. Taking the flow velocity data of the filtered section as an example, the original flow velocity sequence was 20 liters, 18 liters, 22 liters, 19 liters, 21 liters, and 17 liters per minute. After the first level decomposition, the low-frequency components were 19.5, 20.5, and 19.0, and the high-frequency components were 0.5, -1.5, and 1.0. Further decomposition yielded coarser-scale trend features and finer-scale fluctuation features. The component sequences obtained from each level of decomposition were used as multi-scale fluctuation feature sequences of the flow velocity parameters for subsequent correlation analysis.

[0080] The time series of timing control parameters includes discrete state parameters such as stirring speed level, feed valve opening level, and heating power level. A state transition matrix is ​​constructed, and the number of transitions between each state is counted. Taking the stirring speed level as an example, this parameter has three states: low speed, medium speed, and high speed. In the sedimentation section, the following were recorded: 5 transitions from low speed to medium speed, 3 transitions from medium speed to high speed, 4 transitions from high speed to medium speed, 6 transitions from medium speed to low speed, 12 times maintaining the low speed state, 8 times maintaining the medium speed state, and 7 times maintaining the high speed state. Based on these statistical data, a 3x3 state transition matrix is ​​constructed, where each element represents the frequency of transitioning from one state to another. The mode switching frequency sequence is extracted from this matrix, i.e., the total number of transitions for each state is calculated: 11 transitions for the low speed state, 13 transitions for the medium speed state, and 4 transitions for the high speed state.

[0081] After obtaining the characteristic sequences of each process parameter, the mutual information value between these characteristic sequences and the optimization objective is calculated. The material loss rate sub-objective has a value of 3.2% in the dissolution stage, 1.8% in the precipitation stage, 2.5% in the filtration stage, and 1.1% in the crystallization stage. The energy efficiency sub-objective has values ​​of 3.5 kWh per kilogram of product, 2.8 kWh per kilogram, 4.2 kWh per kilogram, and 5.1 kWh per kilogram, respectively. The mutual information calculation quantifies the dependency between the characteristic sequences and the target values ​​by statistically analyzing the joint probability distribution and marginal probability distribution. For example, when calculating the mutual information between the local temperature change rate sequence in the dissolution stage and the material loss rate, the change rate sequence is discretized into three intervals: low, medium, and high. The joint probability of the material loss rate being below the mean in the low change rate interval is 0.25, and the joint probability of it being above the mean is 0.05. In the high change rate interval, the joint probability of the material loss rate being below the mean is 0.08, and the joint probability of it being above the mean is 0.35. Based on these probability distributions, the mutual information value is calculated to be 0.42 bits.

[0082] For the multi-scale fluctuation characteristic sequence of flow velocity in the sedimentation section, the mutual information values ​​of the first-layer low-frequency component, the first-layer high-frequency component, the second-layer low-frequency component, and the energy efficiency sub-objective were calculated, yielding three values ​​of 0.38 bits, 0.56 bits, and 0.29 bits, respectively. For the time-series parameter mode switching frequency sequence of the filtration section, the mutual information value with the material loss rate was calculated to be 0.47 bits, and the mutual information value with the energy efficiency was 0.52 bits. The mutual information values ​​of all process parameter characteristic sequences and the two sub-objectives were summarized to construct a mutual information matrix. The number of rows in this matrix represents the number of process parameter characteristic sequences, and the number of columns represents the number of optimization sub-objectives.

[0083] Principal component analysis (PCA) was performed on the constructed mutual information matrix for dimensionality reduction. This was achieved by calculating the covariance matrix, eigenvalues, and eigenvectors of the matrix. Each row of the mutual information matrix was considered as the coordinates of a sample in the feature space, and the covariance between each feature dimension was calculated to obtain the covariance matrix. The eigenvalues ​​of the covariance matrix were calculated and sorted from largest to smallest as follows: 2.85, 1.62, 0.73, 0.41, and 0.22. The contribution rate of each eigenvalue was calculated. The contribution rate of the first principal component (2.85) divided by the sum of all eigenvalues ​​yielded 48%, the contribution rate of the second principal component (27%) was 27%, and the cumulative contribution rate was 75%. A preset contribution rate threshold of 70% was set, and the first two principal component vectors were extracted. The element values ​​of the first principal component vector were 0.62, 0.58, -0.45, and 0.31, and the element values ​​of the second principal component vector were 0.35, -0.51, 0.67, and 0.42.

[0084] The absolute values ​​of each element in the principal component vectors are used as quantitative indicators of the correlation strength between the corresponding process parameter features and the optimization objective. The first principal component vector shows that the correlation strength of the local temperature change rate feature is 0.62, the correlation strength of the low-frequency component of flow velocity is 0.58, the correlation strength of the high-frequency component of flow velocity is 0.45, and the correlation strength of the timing switching frequency feature is 0.31. The second principal component vector shows that the strength of the other correlation dimension of each feature is 0.35, 0.51, 0.67, and 0.42, respectively. Combining the information from the two principal component vectors, the parameter-objective correlation degree between the crystallization section process parameters and the optimization objective is obtained. Among them, the correlation degree between the flow velocity fluctuation feature and energy consumption efficiency is the highest, followed by the correlation degree between the temperature change feature and material loss rate.

[0085] In one optional implementation, the constraint boundary is decomposed into multiple sub-constraint boundaries based on the parameter-target correlation, resulting in a simplified parameter search domain for each process segment, including:

[0086] The correlation values ​​of temperature parameter, flow rate parameter, and time sequence control parameter in the parameter-target correlation are sorted in descending order, and the optimization priority sequence of each process parameter in the corresponding process time segment is determined according to the sorting order.

[0087] The global constraint ranges of temperature control parameters, flow rate control parameters, and timing control parameters in the multi-level constraint boundary are obtained and mapped to each process timing segment to obtain the parameter correlation vector of each process timing segment; the parameter correlation vector is then used to perform interval shrinkage operation with a preset constraint shrinkage coefficient to obtain the parameter sub-constraint range of each process timing segment.

[0088] The parameter sub-constraint ranges are reorganized according to the optimization priority sequence. The parameter sub-constraint range with the highest optimization priority is used as the main search dimension, and the remaining parameter sub-constraint ranges are used as auxiliary search dimensions. The simplified parameter search domain corresponding to each process time segment is constructed by performing a Cartesian product operation between the main search dimension and the auxiliary search dimensions.

[0089] In the actual process data acquisition for the preheating section, the correlation coefficients for temperature, flow rate, and timing control parameters were calculated to be 0.87, 0.62, and 0.45, respectively. After sorting these three correlation coefficients in descending order, temperature was ranked first, flow rate second, and timing control third. This order determines the optimization priority sequence for each process parameter within the preheating section. For the main processing section, the correlation coefficients for temperature, flow rate, and timing control parameters were 0.92, 0.78, and 0.69, respectively, and were also sorted in descending order. For the cooling section, the correlation coefficients for temperature, flow rate, and timing control parameters were 0.73, 0.54, and 0.38, respectively, and their order determined the optimization priority sequence for this section.

[0090] The global constraint ranges defined in the multi-level constraint boundaries are as follows: the global range for temperature control parameters is between 150°C and 850°C; the global range for flow rate control parameters is between 2 L / min and 20 L / min; and the global range for timing control parameters is between 30 seconds and 300 seconds. These global constraint ranges need to be mapped to each process timing segment. In the preheating segment, the mapped constraint range for temperature control parameters is 150°C to 400°C, the mapped range for flow rate control parameters is 2 L / min to 8 L / min, and the mapped range for timing control parameters is 30 seconds to 90 seconds. In the main processing segment, the mapped range for temperature control parameters is 400°C to 850°C, the mapped range for flow rate control parameters is 8 L / min to 18 L / min, and the mapped range for timing control parameters is 90 seconds to 200 seconds. The temperature control parameters for the cooling section are mapped from 850 degrees Celsius to 200 degrees Celsius, the flow rate control parameters are mapped from 5 liters per minute to 15 liters per minute, and the timing control parameters are mapped from 60 seconds to 180 seconds.

[0091] The parameter mapping ranges of each process segment are combined with their corresponding correlation values ​​to form a parameter correlation vector. The correlation vector for the preheating segment includes a combination of a temperature interval of 250 degrees Celsius and a correlation of 0.87, a flow rate interval of 6 liters per minute and a correlation of 0.62, and a time sequence interval of 60 seconds and a correlation of 0.45. The preset constraint contraction coefficient is set to 0.35 in this embodiment; this coefficient reflects the degree of contraction of the constraint range based on the correlation.

[0092] When performing interval contraction calculations for the preheating section, the product of the original interval length of the temperature parameter (250 degrees Celsius) and its correlation coefficient (0.87) is 217.5. Multiplying this by the contraction coefficient (0.35) yields a contraction of 76.125 degrees Celsius. The midpoint of the original temperature range (150 degrees Celsius to 400 degrees Celsius) is 275 degrees Celsius. Extending from this midpoint by 38.0625 degrees Celsius on both sides, the sub-constraint range of the preheating section temperature parameter is obtained as 236.9375 degrees Celsius to 313.0625 degrees Celsius. The product of the original interval length of the flow rate parameter (6 liters per minute) and its correlation coefficient (0.62) is 3.72. Multiplying this by the contraction coefficient (0.35) yields a contraction of 1.302 liters per minute. The midpoint of the original flow rate range is 5 liters per minute. After extending this range by 0.651 liters per minute on both sides, the sub-constraint range is 4.349 liters per minute to 5.651 liters per minute. The product of the original interval length of the timing control parameter (60 seconds) and the correlation coefficient (0.45) is 27. Multiplying this by the shrinkage coefficient (0.35) yields a shrinkage of 9.45 seconds. The midpoint of the original timing range is 60 seconds. After expanding to both sides by 4.725 seconds, the sub-constraint range is from 55.275 seconds to 64.725 seconds.

[0093] Based on the optimization priority sequence of the preheating section, temperature parameter has the highest priority and is determined as the primary search dimension. Flow rate parameter and timing control parameter are used as auxiliary search dimensions. The sub-constraint range of the primary search dimension is uniformly discretized into 21 sampling points, starting from 236.9375 degrees Celsius and increasing to 313.0625 degrees Celsius at intervals of 3.8 degrees Celsius. The sub-constraint range of the flow rate parameter is discretized into 9 sampling points, starting from 4.349 liters per minute and increasing to 5.651 liters per minute at intervals of 0.1626 liters per minute. The sub-constraint range of the timing control parameter is discretized into 7 sampling points, starting from 55.275 seconds and increasing to 64.725 seconds at intervals of 1.575 seconds.

[0094] A simplified parameter search domain for the preheating section is constructed using the Cartesian product of the main and auxiliary search dimensions. This domain includes all combinations of 21 sampling points in the temperature dimension, 9 sampling points in the flow rate dimension, and 7 sampling points in the timing dimension, totaling 1323 parameter combination points. Each parameter combination point represents a specific process parameter configuration scheme; for example, a temperature of 240.7375 degrees Celsius, a flow rate of 4.5116 liters per minute, and a timing of 56.85 seconds constitute a valid search point. The main processing section and the cooling section each construct their corresponding simplified parameter search domains using the same method. The search domain for the main processing section contains 1785 parameter combination points, and the search domain for the cooling section contains 945 parameter combination points. Compared to the complete search space under global constraints, the simplified parameter search domains for each process section reduce the number of parameter combinations by approximately 68%, significantly improving search efficiency.

[0095] In one optional implementation, a swarm intelligence algorithm is used to perform segmented parallel optimization within the simplified parameter search domain, and the population individuals are initialized separately for each process segment, including:

[0096] The material loss rate sub-objective and the energy efficiency sub-objective are subjected to Chebyshev decomposition to generate uniformly distributed reference vectors and calculate the Euclidean distance between the reference vectors and the origin of the target space. The reference vectors are divided into multiple reference vector subsets according to the Euclidean distance values ​​and sorted in ascending order. The reference vector subsets are mapped to the corresponding process time segments according to the time position of each process time segment.

[0097] Within the process time segment, density peak clustering is performed on the reference vector subset to obtain cluster center scores. Based on the cluster center scores, the population size of each process time segment is weighted and allocated to obtain the allocated population size for each process time segment. Within the parameter sub-constraints of the process parameters, random numbers following a uniform distribution are generated as positional components of temperature, flow rate, and time sequence. The positional components are concatenated to form a random position vector, and the inverse positional components of the positional components are concatenated to form an inverse position vector.

[0098] The random position vector and the reverse position vector are concatenated and non-dominated sorting is performed. Pareto front individuals are selected according to the non-dominated level. From the Pareto front individuals, a corresponding number of individual position vectors are selected according to the allocated population size to complete the population individual initialization for each process time segment.

[0099] The material loss rate sub-objective and the energy efficiency sub-objective are decomposed using Chebyshev decomposition. This decomposition is achieved by calculating the weighted distance between the objective function value and the reference point. Specifically, for the material loss rate sub-objective and the energy efficiency sub-objective, corresponding weight coefficients are set. The difference between each objective function value and its ideal point is calculated, multiplied by the corresponding weight coefficient, and the maximum value among all weighted differences is taken as the objective function value of the decomposed subproblem. This decomposition method can transform a multi-objective optimization problem into a series of single-objective subproblems, each subproblem corresponding to a specific weight vector.

[0100] The Chebyshev decomposition operation generates a uniformly distributed set of reference vectors. Assuming the reference vectors need to be generated in a 2D target space, the number of segments can be set to 12. Weighted components are generated at equal intervals within the interval from 0 to 1, such that the sum of all weighted components equals 1. For example, the generated reference vectors might include weight combinations of 1.0 and 0.0, 0.917 and 0.083, 0.833 and 0.167, 0.75 and 0.25, 0.667 and 0.333, etc. For each generated reference vector, the Euclidean distance between it and the origin of the target space is calculated. This distance is obtained by taking the square root of the sum of the squares of the components of the reference vector. For example, the Euclidean distance between reference vectors 0.6 and 0.8 is 1.0, and the Euclidean distance between reference vectors 0.3 and 0.4 is approximately 0.5.

[0101] Based on the calculated Euclidean distance values, all reference vectors are sorted in ascending order of distance, with those closer to the origin ranked first and those farther away ranked last. The sorted set of reference vectors is then divided into multiple subsets, the number of which corresponds to the number of process time segments. Assuming there are 5 process time segments, the sorted reference vectors are evenly divided into 5 subsets, each containing approximately the same number of reference vectors. According to the temporal position of each process time segment, the reference vector subsets are mapped sequentially to their corresponding process time segments. Reference vector subsets ranked higher are mapped to process segments with earlier temporal positions, and reference vector subsets ranked lower are mapped to process segments with later temporal positions, ensuring differentiated search preferences across different process segments.

[0102] Within each process time segment, density peak clustering is performed on the corresponding subset of reference vectors. This clustering operation identifies cluster centers by calculating the local density and relative distance of each reference vector. Local density is calculated by counting the number of other reference vectors within a given cutoff distance whose distance to the current reference vector is less than the cutoff distance. Relative distance is calculated by finding the closest distance among all reference vectors with a local density greater than the current reference vector. Multiplying the local density by the relative distance yields the cluster center score; reference vectors with higher scores are identified as cluster centers. For example, in a process segment containing 20 reference vectors, density peak clustering identifies 5 cluster centers with scores of 15.3, 12.8, 10.2, 8.9, and 7.1, respectively.

[0103] Based on the cluster center scores calculated for each process time segment, the population size for each process time segment is weighted and allocated. The cluster center scores for all process segments are then normalized so that the sum of all scores equals 1. The normalized scores are used as weight coefficients, multiplied by the total population size, to obtain the allocated population size for each process time segment. Assuming a total population size of 100 individuals, and normalized scores for the five process segments of 0.28, 0.24, 0.19, 0.16, and 0.13, the corresponding allocated population sizes are 28, 24, 19, 16, and 13 individuals, respectively.

[0104] An initial population of individuals is generated within the process parameter constraints of each process time segment. These process parameters include temperature, flow rate, and time parameters, each with its corresponding sub-constraint range. For temperature, assuming a constraint range of 200°C to 600°C, a uniformly distributed random number is generated within this range as the temperature's positional component. For flow rate, assuming a constraint range of 0.5 L / s to 3.0 L / s, a uniformly distributed random number is also generated within this range as the flow rate's positional component. For time parameters, assuming a constraint range of 10 seconds to 60 seconds, random numbers are generated within this range as the time's positional component. The positional components of temperature, flow rate, and time are then concatenated sequentially to form a random position vector; for example, a random position vector might be 350°C, 1.8 L / s, and 35 seconds.

[0105] Simultaneously, reverse position vectors are generated. The reverse position component is calculated by subtracting the corresponding random position component from the sum of the upper and lower bounds of the parameter constraint range. In the example above, the reverse position component of temperature is 200 + 600 - 350 = 450 degrees Celsius; the reverse position component of flow velocity is 0.5 + 3.0 - 1.8 = 1.7 liters per second; and the reverse position component of time is 10 + 60 - 35 = 35 seconds. The reverse position components of each parameter are concatenated to form the reverse position vector, for example, 450 degrees Celsius, 1.7 liters per second, and 35 seconds. According to the required population size after allocation, a corresponding number of random position vectors and reverse position vectors are generated.

[0106] All generated random position vectors are concatenated with their reverse position vectors to form a candidate individual set containing twice the size of the assigned population. A non-dominated ranking operation is performed on this candidate individual set. This ranking operation determines dominance by comparing the objective function values ​​between individuals. If individual A is not inferior to individual B on all objectives and is superior to individual B on at least one objective, then individual A dominates individual B. Individuals not dominated by any other individual are assigned to the first non-dominated level. From the remaining individuals, non-dominated individuals are further searched and assigned to the second non-dominated level, and so on, until all individuals are assigned to a non-dominated level. Pareto front individuals are selected based on the non-dominated level, prioritizing individuals in the first non-dominated level. If the number of individuals in the first level exceeds the size of the assigned population, crowding distance is calculated within that level, and individuals with larger crowding distances are selected to maintain population diversity. Crowding distance is obtained by calculating the sum of the normalized distances of neighboring individuals across all objective dimensions. From the Pareto front individuals, select the corresponding number of individual position vectors according to the population size after allocation to complete the population individual initialization operation for this process time segment, providing a high-quality initial solution set for subsequent swarm intelligence algorithm optimization.

[0107] In one optional implementation, the individuals in the population are combined across process segments according to the sequence of the process segments to form candidate process parameter schemes for the entire process, and the process effect is calculated to obtain the separation purity value corresponding to each candidate process parameter scheme, including:

[0108] Obtain the set of population individual position vectors within the current iteration period, perform velocity update and position update operations on the set of population individual position vectors, and calculate the Euclidean distance between the vectors to obtain the individual distance matrix. Based on the individual distance matrix, construct a two-dimensional projection space, perform density clustering on the projected population individual position points in the two-dimensional projection space, and identify the center point of the density region as the representative population individual position vector.

[0109] Representative population individual position vectors of adjacent process time segments are selected for cross-segment combination, and the material state transmission continuity score value after cross-segment combination is calculated. Cross-segment combination schemes with material state transmission continuity score values ​​higher than a preset continuity threshold are screened to obtain a set of candidate process parameter schemes.

[0110] The temperature parameter setpoint, flow rate parameter setpoint, and time sequence parameter setpoint from the candidate process parameter scheme set are used as process parameter input tensors. Multi-layer convolution feature extraction and fully connected layer mapping are performed on the process parameter input tensors to obtain the predicted mass concentration of the target substance and the predicted mass concentration of the impurity residue, and the ratio is calculated to obtain the separation purity value of the candidate process parameter scheme.

[0111] At the start of the current iteration cycle, the set of individual position vectors of the population is read from the storage module. This set contains parameter vectors that are independently optimized for each process segment. Taking the extraction process as an example, a certain individual position vector contains specific values ​​such as temperature parameter 82 degrees Celsius, flow rate parameter 150 ml / min, and residence time parameter 45 minutes. These position vectors are grouped and stored according to the process segment, forming multiple subsets, each subset corresponding to a specific process stage.

[0112] A velocity update operation is performed on the set of individual position vectors in the population. This operation adjusts the movement direction based on the difference between the individual's historical best position and the global best position. Specifically, the current position vector of the individual is calculated dimension-wise from its historical best position vector, and the same dimension-wise difference is calculated between the current position vector of the individual and the global best position vector of the population. These two difference vectors are multiplied by random weight coefficients of 0.4 and 0.6, respectively, and then added to the inertia term of the original velocity vector, which uses a weight coefficient of 0.7. After the velocity update, the updated velocity vector is added to the current position vector to obtain the new position vector. In a specific example, an individual's original position in the temperature dimension of the extraction section is 80 degrees Celsius, the historical best is 85 degrees Celsius, and the global best is 87 degrees Celsius. After the velocity and position updates, the new position is adjusted to 83.5 degrees Celsius.

[0113] For all updated individual position vectors in the population, Euclidean distances are calculated between each pair of individuals, constructing an individual distance matrix. This matrix is ​​symmetric, where each element represents the distance between two individual position vectors in a multidimensional parameter space. To calculate the distance, the squared differences between the corresponding dimensions of the two position vectors are summed, and the square root of the sum is taken. Assuming the population contains 120 individuals, and each individual position vector contains 8 parameter dimensions, the resulting individual distance matrix is ​​120 x 120. In the specific data, the temperature difference between individuals A and B is 3 degrees Celsius, the flow rate difference is 20 ml / min, and the time difference is 5 minutes. The combined distance value obtained after taking the square root of the sum of the squared differences in each dimension is 12.4.

[0114] A two-dimensional projection space is constructed based on the individual distance matrix using a multidimensional scaling analysis method. This method iteratively optimizes the distances between points in the two-dimensional space to maintain the distance relationships in the original high-dimensional space. Random coordinate points in the two-dimensional space are initialized, and then the positions of each point are iteratively adjusted to minimize the sum of squared differences between the distances between any two points in the two-dimensional space and the corresponding elements in the original distance matrix. A convergence threshold of 0.001 is set during the iteration process; iteration stops when the improvement margin of five consecutive iterations is less than this threshold. After 18 iterations, the projected coordinates of 120 individuals on the two-dimensional plane are determined, with one individual having projected coordinates of x-coordinate 5.2 and y-coordinate 3.8.

[0115] Density clustering is performed in a two-dimensional projected space. This operation uses a density-based spatial clustering algorithm to identify high-density regions. The neighborhood radius parameter is set to 0.8, and the minimum number of contained points parameter is set to 5. Each projected point in the two-dimensional space is scanned, and the number of points within its neighborhood radius is counted. When the number of contained points in the neighborhood of a point reaches or exceeds the minimum number of contained points, the point is marked as a core point. Starting from any core point, all points in its neighborhood are recursively visited, and points that are density-reachable are grouped into the same cluster. In one clustering operation, 7 valid clusters were identified, with the largest cluster containing 26 individuals and the smallest cluster containing 8 individuals. An additional 13 individuals were marked as noise points.

[0116] For each identified density cluster, the sum of distances from all points within the cluster to the cluster center is calculated, and the point with the smallest sum of distances is selected as the representative population individual location vector. In the largest cluster containing 26 individuals, the cumulative distance from each point to all other points within the cluster is calculated. The cumulative distance of the point located at coordinates 5.7, 4.1 is 78.3, which is the smallest within the cluster. Therefore, the original high-dimensional location vector corresponding to this point is selected as the representative location vector. This vector includes parameters such as temperature 85 degrees Celsius, flow rate 155 ml / min, and residence time 48 minutes in the extraction section. After performing this operation on all 7 clusters, 7 representative population individual location vectors are obtained.

[0117] Representative position vectors from adjacent process segments are selected based on their temporal relationships for cross-segment combinations. In a process flow comprising four segments—pretreatment, extraction, washing, and concentration—the representative vector from the pretreatment segment is combined with the representative vector from the extraction segment, the representative vector from the extraction segment is combined with the representative vector from the washing segment, and so on. Assuming the pretreatment segment identifies 3 representative vectors, the extraction segment identifies 5, the washing segment identifies 4, and the concentration segment identifies 6, the total number of combinations for the entire process is 360.

[0118] For each cross-segment combination, a material state transfer continuity score is calculated. This score measures the degree of matching between the output state of the previous process segment and the input state of the next process segment. State parameters such as material temperature, material concentration, and solvent ratio from the previous segment are extracted and compared with the required input state parameters for the next segment. When the output temperature of the previous segment is 75 degrees Celsius and the optimal input temperature for the next segment is 70 degrees Celsius, the deviation in the temperature dimension is 5 degrees Celsius. The deviations in each dimension are normalized: a temperature deviation of 5 degrees Celsius divided by the temperature adjustment range of 40 degrees Celsius yields a normalized deviation of 0.125. The average of the normalized deviations for all state dimensions is then subtracted from 1 to obtain the continuity score. In a certain combination, the temperature dimension score is 0.875, the concentration dimension score is 0.92, the solvent ratio dimension score is 0.88, and the overall continuity score is 0.892.

[0119] A preset continuity threshold of 0.75 was set, and cross-segment combination schemes with continuity scores higher than this threshold were selected as candidate process parameter schemes. Among 360 combinations, 218 combinations had continuity scores exceeding the threshold, and these combinations constituted the candidate process parameter scheme set. Each candidate scheme contained a complete full-process parameter configuration, and all operating parameters from the pretreatment stage to the final concentration stage were determined.

[0120] Temperature, flow rate, and time parameters are extracted from the set of candidate process parameter schemes to construct a process parameter input tensor. This tensor has a three-dimensional structure: the first dimension represents the process segment sequence with a length of 4; the second dimension represents the parameter type, including temperature, flow rate, and time; and the third dimension is the parameter value itself. For a specific candidate scheme, the input tensor includes a temperature of 65 degrees Celsius, a flow rate of 100 ml / min, and a time of 30 minutes in the pretreatment segment, and a temperature of 85 degrees Celsius, a flow rate of 155 ml / min, and a time of 48 minutes in the extraction segment.

[0121] The process parameters are input as a tensor to a trained deep neural network model. This model consists of three convolutional layers and two fully connected layers. The first convolutional layer uses 16 kernels (3x3) with a stride of 1 to extract features from the input tensor, outputting 16 feature maps. The second convolutional layer uses 32 kernels (3x3) to further extract higher-level features based on the output of the first layer. The third convolutional layer uses 64 kernels (2x2) to extract the most abstract representation of the process features. Modified linear units (MLUs) are used as activation functions between all convolutional layers, and max-pooling layers are added after the second and third layers to reduce the feature dimensionality.

[0122] The feature map output from the third convolutional layer is flattened into a one-dimensional vector and input into the first fully connected layer. This layer contains 256 neurons, mapping the convolutional features to a high-dimensional representation space. The second fully connected layer contains two output neurons, corresponding to the predicted mass concentration of the target substance and the predicted mass concentration of residual impurities, respectively. In the prediction of a certain candidate solution, the activation value of the first output neuron is 0.876, corresponding to a target substance mass concentration of 87.6 g / L; the activation value of the second output neuron is 0.034, corresponding to a residual impurity mass concentration of 3.4 g / L.

[0123] The ratio of the predicted mass concentration of the target substance to the predicted mass concentration of the impurity residue is calculated; this ratio is the separation purity value. Dividing 87.6 g / L by 3.4 g / L yields a separation purity value of 25.76, indicating that the concentration of the target substance is 25.76 times the concentration of the impurity. The above prediction process is executed on all 218 candidate process parameter schemes in the set, yielding the corresponding separation purity values ​​for each candidate scheme. The highest purity value is 28.43, the lowest is 18.92, and the average purity value is 23.65. These separation purity values ​​will serve as fitness evaluation indicators for subsequent optimization iterations.

[0124] In one optional implementation, a transitive constraint relationship for material states between process sections is established. Based on this transitive constraint relationship, the process continuity of each candidate process parameter scheme is determined. The fitness value of each candidate process parameter scheme is calculated based on the separation purity value, including:

[0125] The target space is divided into multiple discrete intervals. The distribution frequencies of the output material state parameters of the preceding process time segment and the input material state parameters of the subsequent process time segment are statistically analyzed in each discrete interval. The information entropy value is calculated based on the distribution frequency, and the material state mutual information value is obtained based on the information entropy value. The KL divergence value is obtained by performing a logarithmic ratio operation on the distribution frequency interval.

[0126] Based on the mutual information value of the material state, a mutual information weight coefficient is constructed, and the KL divergence value is weighted using the mutual information weight coefficient to establish the transmission constraint relationship of material state between process sections.

[0127] Extract parameter setting values ​​from the candidate process parameter scheme set, predict the output material state parameter distribution for each process time segment based on the parameter setting values, input the transitive constraint relationship, and calculate the overall KL divergence value corresponding to the candidate process parameter scheme;

[0128] The process continuity penalty factor is obtained by calculating the ratio of the overall KL divergence value to the preset KL divergence threshold. The fitness value of the candidate process parameter scheme is obtained by multiplying the separation purity value by the exponential function value of the process continuity penalty factor.

[0129] In a certain chemical separation process, the material temperature is used as a state parameter, ranging from 50°C to 150°C. This range is evenly divided into 20 discrete intervals, each spanning 5 degrees. For the output material temperature of the preceding distillation process, temperature data was collected every 10 minutes over a 24-hour operating cycle, totaling 144 data points. The number of these data points falling into each discrete interval was counted; for example, there were 23 data points in the 75°C to 80°C interval and 31 data points in the 80°C to 85°C interval. Dividing the number of data points in each interval by the total number of data points (144) yielded the frequency distribution for that interval. The frequencies for the two intervals were 0.1597 and 0.2153, respectively.

[0130] The same statistical processing is applied to the input material temperatures of subsequent distillation process stages to obtain their distribution frequencies in each discrete interval. When calculating the information entropy value of the preceding process output, the distribution frequency of each interval is multiplied by its logarithm, and the result is negative. This result is then summed over all 20 intervals. If the distribution frequency of an interval is 0.1597, this value is multiplied by its natural logarithm (-2.8349) to obtain -0.4527. This operation is performed on all intervals, and the summation yields the information entropy value of the preceding output, which is 2.7834. The information entropy value of the subsequent input is calculated using the same method, resulting in 2.6912.

[0131] Calculating the joint information entropy of material states requires constructing a two-dimensional distribution table. In this table, the horizontal axis represents 20 discrete intervals of the preceding output temperature, and the vertical axis represents 20 discrete intervals of the subsequent input temperature, forming 400 joint interval grid points. The joint distribution of the preceding output temperature and the corresponding subsequent input temperature at 144 time points is statistically analyzed. For example, there are 18 data points where the preceding output temperature falls between 75 and 80 degrees Celsius and the subsequent input temperature falls between 72 and 77 degrees Celsius; the distribution frequency of this joint interval is 0.1250. Multiplying the distribution frequency of each joint interval by its logarithm, taking the negative sign, and summing the results yields a joint information entropy value of 4.8126.

[0132] The mutual information value of material states is obtained by calculating three information entropy values. The information entropy value of the preceding output (2.7834) is added to the information entropy value of the subsequent input (2.6912) to obtain 5.4746. Then, the combined information entropy value (4.8126) is subtracted to obtain the mutual information value of 0.6620. This mutual information value reflects the correlation between the material states of the preceding and following process stages.

[0133] The KL divergence value is calculated by taking the logarithmic ratio of the distribution frequencies of the preceding output and subsequent input over each interval. In the interval from 75°C to 80°C, the distribution frequency of the preceding output is 0.1597, and the distribution frequency of the subsequent input is 0.1389. Dividing the preceding output frequency by the subsequent input frequency yields a ratio of 1.1498, and taking the natural logarithm of this ratio gives 0.1397. Multiplying this logarithmic value by the preceding output distribution frequency of 0.1597 gives 0.0223. This operation is performed on all 20 discrete intervals, and the results from each interval are summed to obtain a total KL divergence value of 0.1834.

[0134] A normalized mutual information value is selected to construct the mutual information weight coefficient. The calculated mutual information value of 0.6620 is divided by the smaller value of 2.6912 between the previous output information entropy value of 2.7834 and the subsequent input information entropy value of 2.6912, resulting in a normalized mutual information coefficient of 0.2460. This coefficient is used to weight the KL divergence value. Multiplying the KL divergence value of 0.1834 by the mutual information weight coefficient of 0.2460 yields a weighted transitive constraint value of 0.0451. The established transitive constraint relationship represents that the deviation between the state distribution of the subsequent process input material and the state distribution of the previous process output material should be controlled within this constraint value.

[0135] The parameter settings for a specific process parameter scheme were extracted from the candidate scheme set. This scheme sets the reflux ratio of the preceding distillation section to 3.2, the operating pressure to 120 kPa, and the feed location to the 8th tray. Based on these parameter settings, the output material temperature distribution of this section was predicted using a process mechanism model. The prediction results show that the output temperature is mainly concentrated in the range of 82°C to 92°C, with a predicted distribution frequency of 0.3125 in the 85°C to 90°C range. The predicted output temperature distribution was used as the input for the subsequent rectification section, with parameters set at a reflux ratio of 2.8 and an operating pressure of 80 kPa. The expected distribution of the input temperature for the subsequent section was predicted, with an expected frequency of 0.2847 in the 85°C to 90°C range.

[0136] The overall KL divergence value corresponding to the candidate scheme is calculated by performing an interval-by-interval logarithmic ratio operation on the predicted output distribution and the expected input distribution. In the interval of 85 degrees to 90 degrees Celsius, the predicted frequency 0.3125 is divided by the expected frequency 0.2847 to obtain a ratio of 1.0976, the logarithm of which is 0.0932. Multiplying this ratio by the predicted frequency gives 0.0291. Summing up the results for all intervals yields the overall KL divergence value of the scheme as 0.2156.

[0137] A preset KL divergence threshold is set to 0.1500, which is determined based on statistical data from historical stable operating conditions. The ratio of the overall KL divergence value of the candidate schemes to the preset threshold is calculated. Dividing 0.2156 by 0.1500 yields 1.4373, which serves as the basis for the process continuity penalty factor.

[0138] The separation purity value obtained by the separation purity calculation module for this candidate scheme is 0.9520. Multiplying this value by the exponential function value of the process continuity penalty factor yields the fitness value. Taking the negative value of the penalty factor 1.4373 and then taking the exponential function (calculating the natural constant to the power of -1.4373), we get an exponential function value of 0.2373. Multiplying the separation purity value of 0.9520 by the exponential function value of 0.2373 gives a fitness value of 0.2259 for this candidate process parameter scheme. This fitness value comprehensively reflects the separation effect and process continuity of the scheme; a higher fitness value indicates a better scheme. When the overall KL divergence value of the candidate scheme is lower than a preset threshold, the penalty factor is less than 1, and the exponential function value is greater than 0.3679, resulting in a smaller reduction in separation purity, indicating that the scheme can maintain good process continuity.

[0139] This invention relates to a battery recycling material separation and purification process optimization system based on swarm intelligence algorithms. The system includes:

[0140] The first unit is used to construct optimization objectives based on battery recycled material composition data and determine multi-level constraint boundaries of process parameters based on the operating status data of the separation equipment.

[0141] The second unit is used to segment and analyze the historical process parameter data of the separation and purification process according to the process flow sequence. By statistically analyzing the impact of the changes in process parameters in each process segment on the optimization target, the parameter-target correlation of each process segment is extracted. Based on the parameter-target correlation, the constraint boundary is decomposed into multiple sub-constraint boundaries to obtain the simplified parameter search domain corresponding to each process segment.

[0142] The third unit is used to perform segmented parallel optimization in the simplified parameter search domain using a swarm intelligence algorithm. It initializes population individuals for each process segment, and iteratively updates each population individual within the sub-constraint boundary. In each iteration, the population individuals are combined across segments according to the process segment order to form candidate process parameter schemes for the entire process and calculate the process effect to obtain the separation purity value corresponding to each candidate process parameter scheme.

[0143] The fourth unit is used to establish the transfer constraint relationship of material state between process sections, determine the process continuity of each candidate process parameter scheme based on the transfer constraint relationship, calculate the fitness value of each candidate process parameter scheme based on the separation purity value, and obtain the battery recycled material separation and purification process with the optimal fitness value.

[0144] A third aspect of the present invention provides an electronic device, comprising:

[0145] processor;

[0146] Memory used to store processor-executable instructions;

[0147] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0148] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0149] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the separation and purification process of battery recycled materials based on swarm intelligence algorithms, characterized in that, include: The optimization objective is constructed based on the composition data of the battery recycled material, and the multi-level constraint boundary of the process parameters is determined based on the operating status data of the separation equipment. Historical process parameter data for the separation and purification process are segmented and analyzed according to the process flow sequence. The impact of process parameter changes within each process segment on the optimization objective is statistically analyzed. The parameter-objective correlation degree for each process segment is extracted. Based on this correlation degree, the constraint boundary is decomposed into multiple sub-constraint boundaries, resulting in a simplified parameter search domain for each process segment, including: The correlation values ​​of temperature parameter, flow rate parameter, and time sequence control parameter in the parameter-target correlation are sorted in descending order, and the optimization priority sequence of each process parameter in the corresponding process time segment is determined according to the sorting order. The global constraint ranges of temperature control parameters, flow rate control parameters, and timing control parameters in the multi-level constraint boundary are obtained and mapped to each process timing segment to obtain the parameter correlation vector of each process timing segment; the parameter correlation vector is then used to perform interval shrinkage operation with a preset constraint shrinkage coefficient to obtain the parameter sub-constraint range of each process timing segment. The parameter sub-constraint range is reorganized according to the optimization priority sequence. The parameter sub-constraint range with the highest optimization priority is used as the main search dimension, and the remaining parameter sub-constraint ranges are used as auxiliary search dimensions. The simplified parameter search domain corresponding to each process time segment is constructed by performing a Cartesian product operation between the main search dimension and the auxiliary search dimensions. A swarm intelligence algorithm is used to perform segmented parallel optimization within the simplified parameter search domain. Population individuals are initialized for each process segment, and each population individual is iteratively updated within the sub-constraint boundary. In each iteration, the population individuals are combined across segments according to the process segment order to form candidate process parameter schemes for the entire process and the process effect is calculated to obtain the separation purity value corresponding to each candidate process parameter scheme. Establish the transfer constraint relationship of material state between process sections, determine the process continuity of each candidate process parameter scheme based on the transfer constraint relationship, calculate the fitness value of each candidate process parameter scheme based on the separation purity value, and obtain the battery recycled material separation and purification process with the optimal fitness value.

2. The method according to claim 1, characterized in that, The optimization objective is constructed based on the composition data of the battery recycled material, and the multi-level constraint boundaries of the process parameters are determined based on the operating status data of the separation equipment, including: The impurity element content ratio and particle size distribution characteristics are extracted from the battery recycled material composition data; a composite operation of reciprocal transformation and logarithmic transformation is performed on the impurity element content ratio to construct a material loss rate sub-objective; kernel density is calculated on the particle size distribution characteristics, and kurtosis coefficient and skewness coefficient are extracted from the calculation results to construct an energy efficiency sub-objective; the material loss rate sub-objective and the energy efficiency sub-objective are sampled from the target space using Pareto front sampling, and the target space is divided into multiple sub-regions using the reference point decomposition method to obtain a multi-level decomposition optimization objective; The equipment energy consumption curve, equipment fault records, and equipment maintenance logs are extracted from the operating status data. A Fourier transform is performed on the equipment energy consumption curves to extract the dominant frequency component and harmonic components from the energy consumption frequency domain features. Based on the dominant frequency component and the harmonic components, the safe operating ranges for temperature control parameters and flow rate control parameters are determined. Fault precursor parameter combination patterns are extracted from the equipment fault records and converted into prohibited ranges for timing control parameters. Text semantic analysis is performed on the equipment maintenance logs to extract the applicable ranges of the parameters. The intersection of the applicable ranges of the parameters with the safe operating ranges and the prohibited ranges is calculated to obtain the multi-level constraint boundaries.

3. The method according to claim 2, characterized in that, Historical process parameter data for the separation and purification process are segmented and analyzed according to the process flow sequence. The impact of process parameter changes within each process segment on the optimization objective is statistically analyzed, and the parameter-objective correlation degree for each process segment is extracted, including: Based on the time nodes of material state transformation in the battery recycling material separation and purification process, the historical process parameter data is divided into multiple process time segments, and the time series of temperature parameters, flow rate parameters, and time series of time control parameters are extracted from each process time segment. Within each process time segment, a sliding window difference operation is performed on the temperature parameter time series to calculate the local rate of change sequence of the temperature parameter; wavelet decomposition operation is performed on the flow velocity parameter time series to extract the multi-scale fluctuation feature sequence of the flow velocity parameter; a state transition matrix is ​​constructed on the time series of the time series control parameters; and the mode switching frequency sequence of the time series parameters is extracted from the state transition matrix. The local rate of change sequence, the multi-scale fluctuation characteristic sequence, and the mode switching frequency sequence are respectively subjected to mutual information calculation with the material loss rate sub-objective and the energy consumption efficiency sub-objective to construct a mutual information matrix. Principal component analysis is performed on the mutual information matrix to reduce dimensionality and extract the principal component vectors whose cumulative contribution rate reaches a preset contribution rate threshold. The values ​​of each element of the principal component vectors are used as quantitative indicators of the correlation strength between the corresponding process parameters and the optimization target to obtain the parameter-target correlation degree for each process time segment.

4. The method according to claim 2, characterized in that, A swarm intelligence algorithm is used to perform segmented parallel optimization within the simplified parameter search domain. The population individuals are initialized separately for each process segment, including: The material loss rate sub-objective and the energy efficiency sub-objective are subjected to Chebyshev decomposition to generate uniformly distributed reference vectors and calculate the Euclidean distance between the reference vectors and the origin of the target space. The reference vectors are divided into multiple reference vector subsets according to the Euclidean distance values ​​and sorted in ascending order. The reference vector subsets are mapped to the corresponding process time segments according to the time position of each process time segment. Within the process time segment, density peak clustering is performed on the reference vector subset to obtain cluster center scores. Based on the cluster center scores, the population size of each process time segment is weighted and allocated to obtain the allocated population size for each process time segment. Within the parameter sub-constraints of the process parameters, random numbers following a uniform distribution are generated as positional components of temperature, flow rate, and time sequence. The positional components are concatenated to form a random position vector, and the inverse positional components of the positional components are concatenated to form an inverse position vector. The random position vector and the reverse position vector are concatenated and non-dominated sorting is performed. Pareto front individuals are selected according to the non-dominated level. From the Pareto front individuals, a corresponding number of individual position vectors are selected according to the allocated population size to complete the population individual initialization for each process time segment.

5. The method according to claim 1, characterized in that, The individuals in the population are combined across different process segments according to the sequence of the process segments to form candidate process parameter schemes for the entire process. The process effect is then calculated to obtain the separation purity value corresponding to each candidate process parameter scheme, including: Obtain the set of population individual position vectors within the current iteration period, perform velocity update and position update operations on the set of population individual position vectors, and calculate the Euclidean distance between the vectors to obtain the individual distance matrix. Based on the individual distance matrix, construct a two-dimensional projection space, perform density clustering on the projected population individual position points in the two-dimensional projection space, and identify the center point of the density region as the representative population individual position vector. Representative population individual position vectors of adjacent process time segments are selected for cross-segment combination, and the material state transmission continuity score value after cross-segment combination is calculated. Cross-segment combination schemes with material state transmission continuity score values ​​higher than a preset continuity threshold are screened to obtain a set of candidate process parameter schemes. The temperature parameter setpoint, flow rate parameter setpoint, and time sequence parameter setpoint from the candidate process parameter scheme set are used as process parameter input tensors. Multi-layer convolution feature extraction and fully connected layer mapping are performed on the process parameter input tensors to obtain the predicted mass concentration of the target substance and the predicted mass concentration of the impurity residue, and the ratio is calculated to obtain the separation purity value of the candidate process parameter scheme.

6. The method according to claim 5, characterized in that, Establish transitive constraints on material states between process sections, determine the process continuity of each candidate process parameter scheme based on the transitive constraints, and calculate the fitness value of each candidate process parameter scheme based on the separation purity value, including: The target space is divided into multiple discrete intervals. The distribution frequencies of the output material state parameters of the preceding process time segment and the input material state parameters of the subsequent process time segment are statistically analyzed in each discrete interval. The information entropy value is calculated based on the distribution frequency, and the material state mutual information value is obtained based on the information entropy value. The KL divergence value is obtained by performing a logarithmic ratio operation on the distribution frequency interval. Based on the mutual information value of the material state, a mutual information weight coefficient is constructed, and the KL divergence value is weighted using the mutual information weight coefficient to establish the transmission constraint relationship of material state between process sections. Extract parameter setting values ​​from the candidate process parameter scheme set, predict the output material state parameter distribution for each process time segment based on the parameter setting values, input the transitive constraint relationship, and calculate the overall KL divergence value corresponding to the candidate process parameter scheme; The process continuity penalty factor is obtained by calculating the ratio of the overall KL divergence value to the preset KL divergence threshold. The fitness value of the candidate process parameter scheme is obtained by multiplying the separation purity value by the exponential function value of the process continuity penalty factor.

7. A battery recycling material separation and purification process optimization system based on swarm intelligence algorithm, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to construct optimization objectives based on battery recycled material composition data and determine multi-level constraint boundaries of process parameters based on the operating status data of the separation equipment. The second unit is used to segment and analyze the historical process parameter data of the separation and purification process according to the process flow sequence. By statistically analyzing the impact of the changes in process parameters in each process segment on the optimization target, the parameter-target correlation of each process segment is extracted. Based on the parameter-target correlation, the constraint boundary is decomposed into multiple sub-constraint boundaries to obtain the simplified parameter search domain corresponding to each process segment. The third unit is used to perform segmented parallel optimization in the simplified parameter search domain using a swarm intelligence algorithm. It initializes population individuals for each process segment and iteratively updates each population individual within the sub-constraint boundary. In each iteration, the individuals in the population are combined across segments according to the process segment order to form candidate process parameter schemes for the entire process and the process effect is calculated to obtain the separation purity value corresponding to each candidate process parameter scheme. The fourth unit is used to establish the transfer constraint relationship of material state between process sections, determine the process continuity of each candidate process parameter scheme based on the transfer constraint relationship, calculate the fitness value of each candidate process parameter scheme based on the separation purity value, and obtain the battery recycled material separation and purification process with the optimal fitness value.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images