Process parameter optimization method and device for large-scale production of ternary positive electrode material

Abstract

Embodiments of the present disclosure relate to the technical field of chemical engineering process scale-up, and specifically provide a process parameter optimization method and device for large-scale production of ternary positive electrode materials. The method comprises: obtaining reference data corresponding to a target process section of large-scale production of ternary positive electrode materials, the reference data including a set of pilot process parameter values, a set of pilot quality detection values, and a set of large-scale production quality target values; retrieving matching historical data from a pre-set historical database, the historical data including historical process parameters and quality detection data of pilot and large-scale production; using a process parameter optimization model corresponding to the target process section, obtaining a set of predicted values of large-scale production process parameters based on the above data; carrying out large-scale process verification according to the predicted values, if the deviation between the actual value of the verified product quality and the quality target value meets a pre-set condition, determining the set of optimized process parameters, otherwise adjusting the parameters and iteratively verifying until the pre-set condition is met.

Description

Technical Field

[0001] This disclosure relates to the field of chemical process scale-up technology, and more specifically, to a method and apparatus for optimizing process parameters for the large-scale production of ternary cathode materials. Background Technology

[0002] The continuous improvement in energy density, cycle life, and fast-charging performance of lithium-ion batteries, along with the rapid expansion of the power battery and energy storage market, has placed higher demands on the iteration speed and mass production stability of ternary cathode material production processes. For example, after completing laboratory research and pilot-scale verification, ternary cathode material manufacturers typically need to scale up their processes from pilot-scale to industrial-scale production within a short period to quickly determine the optimal combination of process parameters suitable for large-scale production.

[0003] The industrial production of ternary cathode materials typically employs a high-temperature solid-state synthesis process. Target electrochemical performance is achieved through the oxidation, sintering, and modification of metal hydroxide precursors. Typical process stages include mixing, sintering, pulverizing, washing, drying, sieving, demagnetizing, batch mixing, and packaging. During the scaling-up from pilot-scale production to full-scale production, multiple process stages, equipment type switching, and multi-physical field coupling are often involved, resulting in a significant nonlinear, strongly coupled, and non-monotonic correlation between process parameters and product quality.

[0004] Traditional process scale-up mainly relies on general scale-up criteria or manual trial and error debugging. However, since the matching degree between the scale-up process parameters determined by the universal scale-up criteria and the actual production conditions is low, a lot of experiments and long-term on-site debugging are often required to ensure that the key quality indicators of the scaled products meet the standards. As a result, the process scale-up cycle is long, the trial and error cost is high, and the parameter stability is poor, making it difficult to meet the market demand for rapid iteration and efficient mass production of ternary cathode materials. Summary of the Invention

[0005] To address the aforementioned issues, this disclosure proposes a method and apparatus for optimizing process parameters in the large-scale production of ternary cathode materials. This method provides a way to shorten the cycle from pilot-scale to large-scale production, thereby quickly determining the optimal combination of process parameters suitable for large-scale production.

[0006] According to one aspect of this disclosure, a method for optimizing process parameters for the large-scale production of ternary cathode materials is provided, comprising: acquiring pilot-scale reference data corresponding to a target process segment, wherein the target process segment is any process segment for the large-scale production of ternary cathode materials, and the pilot-scale reference data includes a set of pilot-scale process parameter values, a set of pilot-scale quality test values, and a set of large-scale production quality target values; retrieving historical process parameter reference data matching the pilot-scale reference data from a preset historical process parameter reference database, wherein the historical process parameter reference data includes a set of historical pilot-scale process parameter values, a set of historical pilot-scale quality test values, a set of historical large-scale production process parameter values, and a set of historical large-scale production quality test values; and using a pre-trained process parameter optimization model corresponding to the target process segment, at least based on the pilot-scale reference data and the matching historical process parameter reference data, obtaining the corresponding parameters for the target process segment. The set of predicted values ​​for large-scale production process parameters is used as the current set of predicted values ​​for large-scale production process parameters, and the following verification steps are performed: A set of actual product quality values ​​for process verification is obtained, where the actual product quality values ​​are the product quality test results obtained by using the current set of predicted values ​​for large-scale production process parameters; when the deviation between the actual product quality values ​​and the target value set for large-scale production quality meets a predetermined deviation condition, the current set of predicted values ​​for large-scale production process parameters is determined as the optimized set of large-scale production process parameters for the target process segment; and when the deviation does not meet the predetermined deviation condition, the current set of predicted values ​​for large-scale production process parameters is adjusted, and the adjusted set of predicted values ​​for large-scale production process parameters is used as the new set of predicted values ​​for large-scale production process parameters, and the verification steps continue.

[0007] According to one aspect of this disclosure, a process parameter optimization apparatus for the large-scale production of ternary cathode materials is provided, comprising: one or more processors; and one or more memories storing computer-executable instructions, which, when executed by the one or more processors, execute a process parameter optimization method for the large-scale production of ternary cathode materials according to an embodiment of this disclosure.

[0008] According to one aspect of this disclosure, a process parameter optimization device for the large-scale production of ternary cathode materials is provided, comprising: a data acquisition module configured to acquire pilot-scale reference data corresponding to a target process segment, wherein the target process segment is any process segment for the large-scale production of ternary cathode materials, and the pilot-scale reference data includes a set of pilot-scale process parameter values, a set of pilot-scale quality test values, and a set of large-scale production quality target values; a data retrieval module configured to retrieve historical process parameter reference data matching the pilot-scale reference data from a preset historical process parameter reference database, wherein the historical process parameter reference data includes a set of historical pilot-scale process parameter values, a set of historical pilot-scale quality test values, a set of historical large-scale production process parameter values, and a set of historical large-scale production quality test values; and a parameter optimization module configured to use a pre-trained process parameter optimization model corresponding to the target process segment to optimize the process parameters based at least on the pilot-scale reference data and the matching historical process parameter reference data. A set of predicted values ​​for large-scale production process parameters corresponding to the target process segment is obtained. The verification iteration module is configured to use this set of predicted values ​​as the current set of predicted values ​​for large-scale production process parameters and execute the following verification steps: obtaining a set of actual product quality values ​​for process verification, where the actual product quality values ​​are the product quality test results obtained from large-scale process verification using the current set of predicted values ​​for large-scale production process parameters; when the deviation between the actual product quality values ​​and the target value set for large-scale production quality meets a predetermined deviation condition, the current set of predicted values ​​for large-scale production process parameters is determined as the optimized set of large-scale production process parameters for the target process segment; and when the deviation does not meet the predetermined deviation condition, the current set of predicted values ​​for large-scale production process parameters is adjusted, and the adjusted set of predicted values ​​is used as the new set of predicted values ​​for large-scale production process parameters, and the verification steps continue.

[0009] According to one aspect of this disclosure, a machine-readable storage medium is provided that stores executable instructions, which, when executed, cause one or more processors to perform a process parameter optimization method for the large-scale production of ternary cathode materials according to embodiments of this disclosure.

[0010] According to one aspect of this disclosure, a computer program product is provided, which includes executable instructions that, when executed, cause one or more processors to perform a process parameter optimization method for the large-scale production of ternary cathode materials according to embodiments of this disclosure.

[0011] Embodiments of various aspects of this disclosure provide a method for optimizing process parameters that can shorten the cycle from pilot-scale to mass production, thereby enabling the rapid determination of the optimal combination of process parameters suitable for mass production. Other advantages of the embodiments of this disclosure will be described below. Attached Figure Description

[0012] A further understanding of the nature and advantages of this application can be achieved by referring to the accompanying drawings. In the drawings, similar components or features may have the same reference numerals.

[0013] Figure 1 An exemplary process is shown for a method of optimizing process parameters for the large-scale production of ternary cathode materials according to one embodiment.

[0014] Figure 2 An exemplary schematic diagram of the retrieval process in a method for optimizing process parameters for the large-scale production of ternary cathode materials, according to one embodiment, is shown.

[0015] Figure 3 An exemplary parameter optimization process is shown in a method for optimizing process parameters for the large-scale production of ternary cathode materials according to one embodiment.

[0016] Figure 4 This illustrates yet another exemplary parameter optimization process in a method for optimizing process parameters for the large-scale production of ternary cathode materials according to one embodiment.

[0017] Figure 5 This illustrates yet another exemplary parameter optimization process in a method for optimizing process parameters for the large-scale production of ternary cathode materials according to one embodiment.

[0018] Figure 6 Another exemplary parameter optimization process is shown in a method for optimizing process parameters for the large-scale production of ternary cathode materials according to one embodiment.

[0019] Figure 7 An exemplary process for determining the initial theoretical set of process parameters for large-scale production of ternary cathode materials is shown in an embodiment of a method for optimizing process parameters for large-scale production of ternary cathode materials.

[0020] Figure 8 This illustrates yet another exemplary process of a method for optimizing process parameters for the large-scale production of ternary cathode materials, according to one embodiment.

[0021] Figure 9 Another exemplary process is shown, according to one embodiment, for optimizing process parameters for the large-scale production of ternary cathode materials.

[0022] Figure 10 An exemplary flowchart of a method for optimizing process parameters for the large-scale production of ternary cathode materials, according to one embodiment, is shown.

[0023] Figure 11 A schematic block diagram of a process parameter optimization apparatus for the large-scale production of ternary cathode materials is shown according to one embodiment.

[0024] Figure 12 Another schematic block diagram of a process parameter optimization apparatus for the mass production of ternary cathode materials according to one embodiment is shown. Detailed Implementation

[0025] The subject matter described herein will be discussed below with reference to exemplary embodiments. It should be understood that these embodiments are discussed merely to enable those skilled in the art to better understand and implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. The function and arrangement of the elements discussed may be changed without departing from the scope of the embodiments disclosed herein. Various processes or components may be omitted, substituted, or added as needed in the various examples. Furthermore, features described in some examples may be combined in other examples.

[0026] As used herein, the term "comprising" and its variations are open terms meaning "including but not limited to". The term "based on" means "at least partially based on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". Other definitions, whether explicit or implicit, may be included below. Unless explicitly indicated by the context, the definition of a term shall remain consistent throughout the specification.

[0027] The flowcharts used in this disclosure illustrate operations implemented according to some embodiments of this disclosure. It should be clearly understood that the operations in the flowcharts may not be implemented sequentially. Instead, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.

[0028] The exemplary method and apparatus for optimizing process parameters for the large-scale production of ternary cathode materials according to embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings.

[0029] Figure 1 An exemplary process 100 for optimizing process parameters for the large-scale production of ternary cathode materials is shown according to one embodiment.

[0030] like Figure 1As shown, pilot-scale reference data 101 corresponding to the target process segment can be obtained. The target process segment can be any process segment for the large-scale production of ternary cathode materials. In some examples, the target process segment may include any of the following: mixing process segment, sintering process segment, pulverizing process segment, washing process segment, drying process segment, sieving process segment, demagnetizing process segment, and batch mixing process segment. The pilot-scale reference data 101 may include a set of pilot-scale process parameter values ​​1011, a set of pilot-scale quality test values ​​1012, and a set of large-scale production quality target values ​​1013. In some examples, the process parameters included in the above-mentioned set of pilot-scale process parameter values ​​1011 can specifically cover various process control parameters involved in the actual operation of the corresponding process segment, such as the stirring speed, stirring time, and material feeding ratio of the mixing process segment, and the sintering temperature, heating rate, holding time, and atmosphere content of the sintering process segment. It can be understood that different process segments often correspond to different types of core process parameters and key quality indicators, resulting in significant differences in parameter selection and optimization directions. In some examples, the set of pilot-scale quality test values ​​1012 can be a set of measured indicators obtained from quality testing of the target process segment's products or the final ternary cathode material samples produced in the pilot-scale test after the pilot-scale test is completed. These indicators may include, for example, particle size distribution, tap density, specific surface area, pH value, and initial discharge specific capacity of the pilot-scale products. In some examples, the set of mass production quality target values ​​1013 can be a set of expected quality indicators set for the mass production stage. Each indicator in this set can be a standard value that the mass-produced products or ternary cathode materials of the target process segment must meet. These indicators may include threshold values ​​for indicators such as cycle life, rate performance, moisture content, and particle uniformity required for mass-produced products.

[0031] In step S110, historical process parameter reference data 1021 matching the pilot-scale reference data 101 can be retrieved from the preset historical process parameter reference database 102. Each piece of historical process parameter reference data in the historical process parameter reference database 102 can be real test and production result data accumulated during the actual process of scaling up from pilot scale to large-scale production. Each piece of historical process parameter reference data can include a set of historical process parameter values ​​for pilot scale, a set of historical quality inspection values ​​for pilot scale, a set of historical process parameter values ​​for large-scale production, and a set of historical quality inspection values ​​for large-scale production. It can be understood that the set of historical process parameter values ​​for large-scale production included in the historical process parameter reference data is the large-scale production process parameter value that was actually adopted and put into use during the historical scale-up test, and is used as one of the optimization references for process parameters. In some examples, the historical process parameter reference database 102 can also be classified and stored according to different process segments to facilitate accurate retrieval of historical data for the same type of process segment. It is understandable that the historical process parameter reference data finally retrieved and matching the pilot-scale reference data 101 can be either a single historical data point (e.g., historical process parameter reference data 1021) or multiple historical data points that meet the matching criteria (e.g., ...). Figure 1 The historical process parameter reference data 1021 shown in the figure are as follows: Figure 1 Other historical process parameter reference data (not shown in the figure) can provide sufficient and reliable reference data for subsequent parameter prediction.

[0032] In step S120, a pre-trained process parameter optimization model 103 corresponding to the target process segment can be used to obtain a set of predicted large-scale production process parameters 104 corresponding to the target process segment, based at least on the pilot-scale reference data 101 and the matched historical process parameter reference data 1021. In some examples, the process parameter optimization model 103 can be a functional model with nonlinear fitting and process parameter correction capabilities, which can accurately predict large-scale process parameters based on pilot-scale data and historical data. For example, the process parameter optimization model 103 can adopt network or model structures with complex operating condition fitting capabilities, such as neural networks and gradient boosting regression models, and the network parameters and hyperparameter configurations of the model are all associated with the corresponding target process segment to adapt to the parameter change patterns of different process segments. In some examples, the process parameter optimization model 103 can be supervisedly trained using training samples constructed based on historical process parameter reference data in the historical process parameter reference database 102. By continuously iterating and optimizing the model parameters, the trained model can be equipped with the ability to map to the set of large-scale production process parameter values ​​based at least on the pilot-scale reference data and the matched historical process parameter reference data.

[0033] In one example, training samples can be constructed as follows: Two historical process parameter reference data sets are selected from the historical process parameter reference database 102, where the pairwise similarity meets a preset condition (e.g., exceeds a preset similarity threshold). These two historical process parameter reference data sets are compared according to their relative merits in the set of historical quality test values ​​from large-scale production. From the historical process parameter reference data sets with relatively better historical quality test values ​​from large-scale production, the set of historical process parameter values ​​from large-scale production is removed, and the remaining set of historical process parameter values ​​from pilot-scale production, the set of historical quality test values ​​from pilot-scale production, and the set of historical quality test values ​​from large-scale production are used to form the pilot-scale production sample reference data. The historical process parameter reference data sets with relatively poor historical quality test values ​​from large-scale production are used as the matching sample historical process parameter reference data corresponding to the aforementioned pilot-scale production sample reference data. The removed set of historical process parameter values ​​from large-scale production is used as the sample label corresponding to the aforementioned pilot-scale production sample reference data. Similarly, multiple training samples can be constructed to form a training sample set. During model training, the pilot-scale and large-scale sample reference data and the corresponding matching sample historical process parameter reference data in the training samples can be used as the input of the model, and the sample label corresponding to the training sample can be used as the expected output of the model. In this way, the process parameter optimization model 103 is finally obtained by iteratively optimizing the model parameters.

[0034] In some examples, the set of predicted large-scale production process parameters 104 can be a set of large-scale production process parameters to be verified that are adapted to the target process segment. It can usually cover all the key process parameters required for the production of the target process segment, such as the mixing and proportioning parameters of the mixing segment, the temperature and wind speed parameters of the drying segment, and the screening mesh and operating speed parameters of the screening segment, so as to provide a data basis for subsequent process verification.

[0035] In some examples, the set of predicted values ​​104 of the scaled production process parameters corresponding to the target process segment obtained through step S120 can be used as the set of predicted values ​​105 of the current scaled production process parameters.

[0036] In this embodiment, the predicted value set 105 of the current large-scale production process parameters can be used for large-scale process verification, and the products of the large-scale process verification can be quality tested to obtain the actual quality value set 106 of the process verification products. In some examples, the large-scale process verification can be a batch trial production verification based on the actual pilot production line of the factory, which is larger than the pilot scale but smaller than the actual production scale of the factory, so as to truly reflect the actual operating status of the target process segment in an industrial production environment. In some examples, the actual quality value set 106 of the process verification products can be a set of various measured quality indicators obtained by testing the trial production products. The specific indicators are directly related to the functional characteristics of the target process segment, such as the material uniformity corresponding to the mixing process segment, the crystal integrity of the product corresponding to the sintering process segment, the moisture content of the product corresponding to the drying process segment, and the particle size distribution corresponding to the screening process segment. In some examples, product quality data can be obtained either manually input by technicians based on on-site test results or directly collected from various dedicated quality testing equipment that communicates with the production system. For example, the aforementioned equipment may include a laser particle size analyzer, a tap density tester, a specific surface area analyzer, an electrochemical workstation, and a moisture analyzer. By directly obtaining the product quality test results corresponding to the large-scale process verification using the predicted value set 105 of the current large-scale production process parameters, the actual value set 106 of the process verification product quality can be accurately obtained.

[0037] In step S130, it can be determined whether the deviation between the actual quality value set 106 of the process verification product and the target quality value set 1013 of mass production meets a predetermined deviation condition. In some examples, the predetermined deviation condition can be a criterion used to measure the degree of fit between the quality of trial production products and the mass production quality target, which can intuitively reflect whether the product quality under the current process parameters meets the requirements of mass production. For example, the predetermined deviation condition can be that the difference between each corresponding quality indicator in the actual quality value set of the process verification product and the target quality value set of mass production is less than the corresponding preset threshold. Another example is that the predetermined deviation condition can also be that the comprehensive deviation value obtained after weighted calculation of the differences between each corresponding quality indicator is less than a preset comprehensive judgment threshold. In some examples, the above weighting coefficients can be adaptively set according to the importance of each quality indicator.

[0038] If step S130 determines "yes," the current set of predicted large-scale production process parameters 105 can be identified as the optimized set of large-scale production process parameters 107 for the target process segment. In some examples, the optimized set of large-scale production process parameters 107 for the target process segment can be the optimal combination of process parameters that is fully adapted to the factory's large-scale production conditions and can stably produce qualified products. Its specific parameters also match the target process segment. For example, for the mixing process segment, the optimized set of large-scale production process parameters 107 may include stirring speed and feed ratio values. For another example, for the sintering process segment, the optimized set of large-scale production process parameters 107 may include temperature curves and atmosphere control parameter values. For yet another example, for the drying process segment, the optimized set of large-scale production process parameters 107 may include temperature values ​​and air supply speed values.

[0039] If step S130 determines no, in step S140, the current set of predicted values ​​for large-scale production process parameters 105 can be adjusted, and the adjusted set of predicted values ​​for large-scale production process parameters can be used as the new set of predicted values ​​for the current large-scale production process parameters. Then, the process returns to execute the above-mentioned steps related to large-scale process verification and quality inspection, that is, to obtain the new set of actual values ​​for the quality of the process verification product 106 obtained by using the new set of predicted values ​​for the current large-scale production process parameters for large-scale process verification. Then, step S130 and subsequent determination and parameter adjustment steps are executed until the set of actual values ​​for the quality of the process verification product meets the predetermined deviation conditions.

[0040] It should be understood that all steps and their order in process 100 are exemplary, and embodiments of this disclosure will also cover any modifications to process 100. For example, in some implementations, process 100 may also include other steps as described in the following embodiments.

[0041] Figure 2 An exemplary schematic diagram of a retrieval process 200 in a method for optimizing process parameters for the mass production of ternary cathode materials according to one embodiment is shown. Figure 2 yes Figure 1 An exemplary implementation of step S110. To avoid repetition, only the differences will be described here.

[0042] like Figure 2As shown, the preset historical process parameter reference database can include multiple historical process parameter reference data. In some examples, the first historical process parameter reference data can be represented as [a11,a21,a31,q11,q21,A11,A21,A31,Q11,Q21], where a11, a21, a31, q11, q21, A11, A21, A31, Q11, and Q21 can respectively represent the raw material weight value, impeller speed value, and mixing time value used in historical pilot-scale tests, the corresponding lithium residue and pH value reflecting the product quality test results of historical pilot-scale tests, and the raw material weight value, impeller speed value, and mixing time value used in historical large-scale production, the corresponding lithium residue and pH value reflecting the product quality test results of historical large-scale production. In some examples, each historical process parameter reference data can first be normalized and then encoded using a locality-sensitive hash algorithm to obtain the corresponding hash code. In one example, the hash code corresponding to the first historical process parameter reference data mentioned above could be, for example, 110101. Similarly, each historical process parameter reference data in the historical process parameter reference database can have a corresponding hash code. It can be understood that each hash code can be viewed as a hash bucket, and historical process parameter reference data with the same hash code can be mapped to the same hash bucket. For example... Figure 2 As shown, in some examples, the 4th, 6th, and 7th historical process parameter reference data can be mapped to hash bucket 110011, the 2nd and 5th historical process parameter reference data can be mapped to hash bucket 110100, and the 1st and 3rd historical process parameter reference data can be mapped to hash bucket 110101.

[0043] like Figure 2 As shown, in some examples, the pilot-scale reference data 101 can be represented as [a1,a2,a3,q1,q2,Q1,Q2], where a1, a2, a3, q1, q2, Q1, and Q2 can respectively represent the raw material weight, impeller speed, and mixing time used in the pilot-scale test, the corresponding lithium residue and pH value reflecting the product quality test results of the pilot-scale test, and the corresponding lithium residue and pH value expected to be achieved in the large-scale production product quality test. Similarly, the pilot-scale reference data 101 [a1,a2,a3,q1,q2,Q1,Q2] can be normalized and then encoded using a locality-sensitive hash algorithm to obtain the corresponding hash code, such as 110011. Afterwards, historical process parameter reference data (e.g., ...) mapped to hash bucket 110011 can be... Figure 2The historical process parameter reference data (such as the 4th, 6th, and 7th historical process parameter reference data shown) are identified as candidate historical process parameter reference data. Next, the similarity between the pilot-scale reference data 101 [a1,a2,a3,q1,q2,Q1,Q2] and each candidate historical process parameter reference data can be calculated. In some examples, the similarity calculation can employ numerical methods to measure the degree of similarity between two sets of data vectors, including various methods that can intuitively reflect the feature matching degree between data. For example, numerical calculation methods such as cosine similarity calculation, Euclidean distance calculation, or Manhattan distance calculation can be used. Then, the candidate historical process parameter reference data whose similarity meets the predetermined similarity conditions can be identified as historical process parameter reference data 1021 that matches the pilot-scale reference data. In some examples, the predetermined similarity conditions can be a judgment rule used to filter out valid matching historical data, and this rule can be flexibly set according to actual process matching requirements. For example, the predetermined similarity condition could be that the similarity between candidate historical process parameter reference data and pilot-scale reference data is greater than a preset similarity threshold, or it could be selecting one or more candidate data points that are most similar to the pilot-scale reference data. For instance, the 6th historical process parameter reference data point with the highest similarity to pilot-scale reference data 101 could be identified as historical process parameter reference data 1021 that matches the pilot-scale reference data. It can be understood that, depending on the actual process parameter optimization needs, there can be more than one historical process parameter reference data point that matches the pilot-scale reference data, thus providing richer reference data for subsequent parameter prediction.

[0044] By employing the above method, locality-sensitive hashing (LSH) algorithms can be used to quickly screen historical process data. Combined with accurate similarity calculations, effective data matching can be achieved. This not only effectively reduces the computational load and time consumption of large-scale historical data retrieval, but also ensures that the matched historical data has a high degree of process adaptability with the current pilot-scale to large-scale reference data. This lays a reliable data foundation for the accurate prediction of subsequent large-scale production process parameters and significantly improves the efficiency of process parameter matching and optimization in the process of ternary cathode materials from pilot-scale to large-scale production.

[0045] Figure 3 An exemplary parameter optimization process 300 is shown in a method for optimizing process parameters for the mass production of ternary cathode materials according to one embodiment. Figure 3 yes Figure 1 An exemplary implementation of step S120. Figure 3 Zhongyu Figure 2 The same parts in the embodiments can be referred to Figure 2 The relevant descriptions in [the original text] will be presented here. To avoid repetition, only the differences will be described here.

[0046] like Figure 3 As shown, in step S310, a pilot-scale process parameter deviation feature 301 can be generated based on the corresponding difference between the set of pilot-scale process parameter values ​​1011 in the pilot-scale reference data 101 and the set of historical pilot-scale process parameter values ​​10211 in the matched historical process parameter reference data 1021. In some examples, the corresponding difference can be calculated by subtracting the corresponding process parameter values ​​in the set of pilot-scale process parameter values ​​1011 and the set of historical pilot-scale process parameter values ​​10211. For example, it can be referenced... Figure 2 Examples include the difference (a1-a16) between the raw material weight value a1 in the pilot-scale process parameter value set and the corresponding raw material weight value a16 in the historical pilot-scale process parameter value set, and the difference (a2-a26) between the pilot-scale blade speed value a2 and the historical blade speed value a26. In some examples, the pilot-scale process parameter deviation feature 301 can be a feature vector composed of the differences between the corresponding process parameters mentioned above. This vector is used to quantify the degree of difference between the current pilot-scale process parameters and the historical pilot-scale process parameters, and can intuitively reflect the parameter deviation between the current pilot-scale operating conditions and the historical pilot-scale operating conditions. For example, this feature vector can be represented as [a1-a16, a2-a26, a3-a36], where a3 can be the pilot-scale mixing time value, and a36 can be the historical pilot-scale mixing time value.

[0047] In step S320, a pilot-scale quality deviation feature 302 can be generated based on the corresponding differences between the pilot-scale quality test value set 1012 in the pilot-scale reference data 101 and the historical pilot-scale quality test value set 10212 in the matching historical process parameter reference data 1021. In some examples, the corresponding differences can be obtained by subtracting the measured values ​​of each corresponding quality indicator in the pilot-scale quality test value set 1012 from the measured values ​​of each corresponding quality indicator in the historical pilot-scale quality test value set 10212. For example, the difference between the lithium residue q1 of the pilot product and the lithium residue q16 of the historical pilot product (q1-q16), and the difference between the pH value q2 of the pilot product and the pH value q26 of the historical pilot product (q2-q26). In some examples, the pilot-scale quality deviation feature 302 can be a feature vector composed of the corresponding differences of the above quality indicators, used to characterize the difference between the current pilot-scale product quality and the historical pilot-scale product quality, so as to clearly reflect the superiority or inferiority of the current pilot-scale quality and the deviation range compared with the historical pilot-scale quality. For example, the feature vector can be represented as [q1-q16, q2-q26].

[0048] In step S330, a large-scale production quality deviation feature 303 can be generated based on the corresponding differences between the set of large-scale production quality target values ​​1013 in the pilot-scale reference data 101 and the set of historical large-scale production quality detection values ​​10213 in the matched historical process parameter reference data 1021. In some examples, the corresponding differences can be calculated by subtracting the corresponding measured values ​​in the set of historical large-scale production quality detection values ​​10213 from the quality target values ​​in the set of large-scale production quality target values ​​1013. For example, the difference between the target value of large-scale lithium residue Q1 and the historical measured value of lithium residue Q16 (Q1-Q16), and the difference between the target value of large-scale pH Q2 and the historical measured value of pH Q26 (Q2-Q26). In some examples, the large-scale production quality deviation feature 303 can be a feature vector composed of the corresponding differences between the above quality targets and historical measured values, used to quantify the gap between the large-scale quality targets and the actual quality of historical large-scale production, providing a clear directional basis for subsequent process parameter correction. For example, the feature vector can be represented as [Q1-Q16, Q2-Q26].

[0049] In this embodiment, at least the pilot-scale process parameter deviation features 301, pilot-scale quality deviation features 302, large-scale production quality deviation features 303, and the set of historical process parameter values ​​10214 for large-scale production can be input into the residual network 1031 to obtain the output set of residual correction values ​​304 for large-scale production process parameters. In this embodiment, the process parameter optimization model 103 may include the residual network 1031. In some examples, the residual network 1031 may be a deep learning network with nonlinear fitting capabilities and deviation correction functions, which can accurately output residual values ​​for correcting historical parameters based on various input deviation features and historical large-scale parameters. In some examples, the network structure of the residual network 1031 may include multiple residual blocks, fully connected layers, and activation layers, and the network parameters (such as the number of residual blocks, the number of neurons in the fully connected layers, etc.) are all related to the target process segment. For example, the number of residual blocks and the selection of activation functions for the residual network for the mixing process segment and the residual network for the sintering process segment will be adapted and adjusted according to the variation law of their respective process parameters.

[0050] Similarly, it can be based on Figure 1In step S120, the method for constructing training samples is used to construct training samples for training the residual network 1031. In one example, the corresponding difference between the set of remaining pilot-scale historical process parameter values ​​in the historical process parameter reference data after removing the set of historical process parameter values ​​for large-scale production and the set of pilot-scale historical process parameter values ​​in the historical process parameter reference data with relatively poor historical quality test values ​​for large-scale production can be calculated to generate sample pilot-scale process parameter deviation features. Similarly, the corresponding differences between the set of remaining pilot-scale historical quality test values ​​in the historical process parameter reference data after removing the set of historical process parameter values ​​for large-scale production and the set of pilot-scale historical quality test values ​​in the historical process parameter reference data with relatively poor historical quality test values ​​for large-scale production can be calculated to generate sample pilot-scale quality deviation features and sample large-scale production quality deviation features, respectively. Subsequently, the difference between the set of removed historical process parameter values ​​from large-scale production and the set of historical process parameter values ​​from the reference data of historical process parameters with relatively poor historical quality detection values ​​from large-scale production can be used as the sample label corresponding to the aforementioned sample pilot-scale process parameter deviation features, sample pilot-scale quality deviation features, and sample large-scale production quality deviation features. Similarly, multiple training samples can be constructed to form a training sample set. During model training, the sample pilot-scale process parameter deviation features, sample pilot-scale quality deviation features, sample large-scale production quality deviation features, and the set of historical process parameter values ​​from the reference data of historical process parameters with relatively poor historical quality detection values ​​from large-scale production can be used as the input to the model, and the sample label corresponding to the training sample can be used as the expected output of the model. Thus, by continuously iterating and optimizing the model parameters, the trained residual network 1031 is finally obtained.

[0051] In some examples, the set of residual correction values ​​304 for large-scale production process parameters can be a set of differences used to correct the set of historical process parameter values ​​10214 for large-scale production. In some examples, the set of residual correction values ​​304 for large-scale production process parameters can contain correction values ​​that correspond one-to-one with the set of historical process parameter values ​​10214 for large-scale production, used to adjust parameters matched based on historical experimental data to adapt to the current pilot-scale conditions and large-scale quality targets. For example, for the mixing process section, the set of residual correction values ​​304 for large-scale production process parameters can include correction values ​​for stirring speed, mixing time, etc. For example, for the sintering process section, the set of residual correction values ​​304 for large-scale production process parameters can include correction values ​​for sintering temperature, heating rate, etc.

[0052] In step S340, the historical process parameter value set 10214 of large-scale production and the residual correction value set 304 of large-scale production process parameters can be superimposed to obtain the predicted value set 104 of large-scale production process parameters for the target process segment. In some examples, the superposition operation can be to add the corresponding parameter values ​​in the two sets one by one. For example, if the stirring speed in the historical process parameter value set 10214 of large-scale production is 800 r / min, and the corresponding correction value for the stirring speed in the residual correction value set 304 of large-scale production process parameters is -50 r / min, then the predicted stirring speed obtained after superposition can be 750 r / min. As another example, if the sintering temperature in the historical process parameter value set 10214 of large-scale production is 850℃, and the corresponding correction value in the residual correction value set 304 of large-scale production process parameters is +20℃, then the predicted sintering temperature obtained after superposition can be 870℃.

[0053] By calculating various deviation characteristics, the differences between the current pilot-scale operating conditions, quality status, and historical data can be fully explored. With the help of a residual network adapted to the process segment, the accurate output of residual correction values ​​can be achieved. Then, through simple superposition calculation, the predicted values ​​of large-scale production process parameters adapted to the current needs can be obtained. This not only effectively avoids the problem of blind parameter adjustment in traditional process scale-up, but also makes full use of the experience advantage of historical process data, significantly improving the accuracy and efficiency of large-scale process parameter prediction. This provides reliable parameter support for the rapid and stable scale-up of ternary cathode materials from pilot-scale to large-scale production, while reducing the trial and error costs and cycle of process scale-up.

[0054] Figure 4 This illustrates yet another exemplary parameter optimization process 400 in a method for optimizing process parameters for the mass production of ternary cathode materials according to one embodiment. Figure 4 yes Figure 1 Another exemplary implementation of step S120, Figure 4 The exemplary parameter optimization process 400 can be based on Figure 3 The exemplary parameter optimization process 300 shown is used to achieve this. Figure 4 Zhongyu Figure 3 The same parts in the embodiments can be referred to the relevant description in section 3. To avoid repetition, only the differences will be described here.

[0055] like Figure 4As shown, a set of initial theoretical values ​​401 for large-scale production process parameters can be obtained based on the scale-up criterion. In some examples, the set of initial theoretical values ​​401 for large-scale production process parameters can be a set of initial large-scale parameters obtained by converting the values ​​of each parameter in the pilot-scale process parameter set using theoretical formulas according to the classical process scale-up criterion, serving as the initial reference parameters. For example, for the mixing section, theoretical formulas for calculating stirring speed, etc., can be obtained based on the same criterion for energy input per unit volume. As another example, for the drying section, theoretical formulas for calculating drying temperature, etc., can be obtained based on the constant Froude number criterion.

[0056] In step S410, a large-scale production process parameter deviation feature 402 can be generated based on the corresponding differences between the initial theoretical value set 401 of large-scale production process parameters and the historical process parameter value set 10214 in the matched historical process parameter reference data 1021. In some examples, the corresponding differences can be obtained by subtracting the values ​​of each corresponding parameter in the initial theoretical value set 401 of large-scale production process parameters and the historical process parameter value set 10214 of large-scale production. For example, the difference between the theoretical initial stirring speed and the historical large-scale stirring speed can be obtained. Another example is the difference between the theoretical initial sintering temperature and the historical large-scale sintering temperature. In some examples, the large-scale production process parameter deviation feature 402 can be used to characterize the difference between the theoretical scale-up results and the process parameters used in historical actual experiments. For example, the large-scale production process parameter deviation feature 402 can be a feature vector composed of the corresponding differences between the aforementioned theoretical initial parameters and historical large-scale parameters, obtained by referring to the various deviation features described above.

[0057] In this embodiment, the pilot-scale process parameter deviation features 301, pilot-scale quality deviation features 302, large-scale production quality deviation features 303, large-scale production process parameter deviation features 402, and the set of historical large-scale production process parameter values ​​10214 can be input into the residual network 1032 to obtain the output set of large-scale production process parameter residual correction values ​​403. In this embodiment, the process parameter optimization model 103 may include the residual network 1032. In some examples, the residual network 1032 may be a deep learning network with multi-feature fusion and parameter correction capabilities. In some examples, the network structure of the residual network 1032 may include residual blocks and fully connected layers, and its network layer number, number of neurons, and other parameters are adapted to the parameter type of the target process segment. It can be understood that the residual network 1032 and... Figure 3 The residual network 1031 shown in the figure may be similar in structure, but there are differences in input data dimension, network weight parameters, feature fusion method, etc.

[0058] Similarly, it can be based on Figure 3The method described in section 1031 for constructing training samples for residual network 1032 is used to construct training samples for residual network 1032. In one example, based on the amplification criterion, the initial theoretical set of sample-scale production process parameters can be obtained by converting the set of pilot-scale historical process parameter values ​​remaining in the historical process parameter reference data (after removing the set of historical process parameter values ​​from large-scale production) using theoretical formulas. Then, the corresponding difference between the initial theoretical set of sample-scale production process parameters and the set of historical process parameter values ​​in the historical process parameter reference data with relatively poor historical quality detection values ​​can be calculated to generate the sample-scale production process parameter deviation features. Similarly, multiple training samples can be constructed to form a training sample set. During model training, the sample input can be the sample pilot-scale process parameter deviation features, sample pilot-scale quality deviation features, sample large-scale production quality deviation features, sample large-scale production process parameter deviation features, and the set of large-scale production historical process parameter values ​​from the historical process parameter reference data with relatively poor historical quality detection values ​​in the training samples. The sample label corresponding to the training sample is used as the expected output of the model. Thus, by continuously iterating and optimizing the model parameters, the trained residual network 1032 is finally obtained.

[0059] In some examples, the set of residual correction values ​​403 for large-scale production process parameters can be a comprehensive set of correction parameters adapted to multiple types of deviation characteristics. It can be understood that the set of residual correction values ​​403 for large-scale production process parameters is related to... Figure 3 The set of residual correction values ​​for large-scale production process parameters 304 shown in the figure may be consistent in terms of parameter composition, but different in terms of specific parameter values.

[0060] By combining the prior knowledge of theoretical scale-up criteria with the empirical characteristics of historical process data, and further enriching the model's input information by adding deviation characteristics of large-scale production process parameters, the residual network can output correction values ​​that are more in line with actual operating conditions, based on both theoretical calculations and historical data. This can improve the rationality of large-scale process parameter predictions while further reducing the deviation between theoretical scale-up and actual production. The parameter optimization results not only conform to the basic principles of process scale-up but also adapt to the actual operating conditions of the production line, thereby significantly improving the parameter adaptability and implementation efficiency of ternary cathode materials from pilot-scale to large-scale production.

[0061] Figure 5 Another exemplary parameter optimization process 500 is shown in a method for optimizing process parameters for the mass production of ternary cathode materials according to one embodiment. Figure 5 yes Figure 1 Another exemplary implementation of step S120, Figure 5 The exemplary parameter optimization process 500 can be based on Figure 4 The exemplary parameter optimization process 400 shown is used to achieve this. Figure 5 Zhongyu Figure 4 The same parts in the embodiments can be referred to Figure 4 The relevant descriptions in [the original text] will be presented here. To avoid repetition, only the differences will be described here.

[0062] In this embodiment, the residual network 1032 may include a historical experience branch residual network 10321, an initial parameter branch residual network 10322, and a fusion network 10323. In some examples, the residual network 1032 may employ a method similar to that described above. Figure 4 The training method described in the embodiment is obtained by jointly training all network parameters of the historical experience branch residual network 10321, the initial parameter branch residual network 10322, and the fusion network 10323 through supervised learning, so that the three branch networks can work together to complete the bias feature learning and parameter correction. In some examples, the joint training of the residual network 1032 may specifically include: using the sample pilot-scale process parameter deviation features, sample pilot-scale quality deviation features, sample large-scale production quality deviation features, and the set of large-scale production historical process parameter values ​​from the historical process parameter reference data with relatively poor historical quality detection values ​​in the training samples as inputs to the historical experience branch residual network 10321; using the sample large-scale production process parameter deviation features and the set of large-scale production historical process parameter values ​​from the historical process parameter reference data with relatively poor historical quality detection values ​​in the training samples as inputs to the initial parameter branch residual network 10322; using the outputs of the historical experience branch residual network 10321 and the initial parameter branch residual network 10322 as inputs to the fusion network 10323, and using the sample label corresponding to the training sample as the expected output of the fusion network 10323, thereby continuously iterating and optimizing the model parameters to finally obtain the trained historical experience branch residual network 10321, initial parameter branch residual network 10322, and fusion network 10323.

[0063] like Figure 5As shown, the pilot-scale process parameter deviation features 301, pilot-scale quality deviation features 302, large-scale production quality deviation features 303, and the set of historical process parameter values ​​for large-scale production 10214 can be input into the historical experience branch residual network 10321 to obtain the output set of historical experience branch residual correction values ​​501. In some examples, the historical experience branch residual network 10321 can be a residual subnetwork specifically designed to learn the deviation features of historical process data, which can output correction values ​​by extracting the correlation features between pilot-scale operating condition deviations and quality deviations. In some examples, the number of residual blocks, activation function, and other parameters of the historical experience branch residual network 10321 are adapted to the target process segment. For example, the historical experience branch network adapted to the mixing process segment and the network adapted to the sintering process segment may differ in the number of layers. In some examples, the set of historical experience branch residual correction values ​​501 can be a set of correction parameters mainly obtained based on historical process experience, and its parameter composition is consistent with the set of large-scale production process parameter residual correction values ​​403.

[0064] The large-scale production process parameter deviation characteristics 402 and the large-scale production historical process parameter value set 10214 can be input into the initial parameter branch residual network 10322 to obtain the output initial parameter branch residual correction value set 502. In some examples, the initial parameter branch residual network 10322 can be a residual subnetwork specifically designed to learn the deviation characteristics between theoretically amplified initial parameters and historical parameters, which can output corresponding correction values ​​based on the theoretically amplified deviation information. Similarly, the network structure of the initial parameter branch residual network 10322 can also be associated with the parameter characteristics of the target process segment. In some examples, the initial parameter branch residual correction value set 502 can be a set of correction parameters mainly obtained based on theoretical amplification criteria, and its parameter composition is consistent with the large-scale production process parameter residual correction value set 403.

[0065] Alternatively, the set of historical experience branch residual correction values ​​501 and the set of initial parameter branch residual correction values ​​502 can be input into the fusion network 10323 to obtain the output set of scaled production process parameter residual correction values ​​403. In some examples, the fusion network 10323 can be a fully connected network or a convolutional network with feature weighting and fusion functions, which can adaptively fuse the correction values ​​output by the two branch networks, and then output the final residual correction value that integrates the two types of correction bases.

[0066] By using the above method, historical process data experience and theoretical a priori knowledge for scale-up can be separated into two independent branches for learning. Then, by using a fusion network, the advantages of the two types of correction information can be complemented. This not only preserves the reliability of historical data implementation but also takes into account the rationality of theoretical scale-up processes. Compared with a single residual network, it can further reduce the prediction bias of large-scale process parameters, making the parameter optimization results more in line with the actual operating conditions of ternary cathode materials mass production. This effectively improves the accuracy of parameter adjustment and the first-time implementation rate during the process from pilot-scale to large-scale scale-up.

[0067] Figure 6 Another exemplary parameter optimization process 600 is shown in a method for optimizing process parameters for the large-scale production of ternary cathode materials according to one embodiment. Figure 6 An exemplary parameter optimization process 600 may be Figure 5 This is another exemplary implementation of using the initial parameter branch residual network 10322 to output the initial parameter branch residual correction value set 502. The parameter optimization process 600 can be based on... Figure 5 The exemplary parameter optimization process 500 shown is used to achieve this. Figure 6 Zhongyu Figure 5 The same parts in the embodiments can be referred to Figure 5 The relevant descriptions in [the original text] will be presented here. To avoid repetition, only the differences will be described here.

[0068] like Figure 6 As shown, the large-scale production process parameter deviation features 402, large-scale production quality deviation features 303, and the set of historical large-scale production process parameter values ​​10214 can be input into the initial parameter branch residual network 10324 to obtain the output set of initial parameter branch residual correction values ​​601. In some examples, the initial parameter branch residual network 10324 can be a residual subnetwork specifically designed to learn the correlation features between large-scale production process parameter deviations, large-scale production quality deviations, and historical large-scale parameters. It can combine theoretical amplification of the deviation information of the initial parameters and the deviation information of the large-scale quality target to output adapted correction values. Similarly, the parameters of the initial parameter branch residual network 10324, such as the number of residual blocks and the configuration of fully connected layer neurons, can also be adaptively adjusted to the target process segment.

[0069] It is understandable that the initial parameter branch residual network 10324 and Figure 5 The initial parameter branch residual network 10322 shown can be structurally similar, but differs in the number of input data, network weight parameters, and feature extraction focus. Compared to Figure 5The initial parameter branch residual network 10322 in the network is used to add the large-scale production quality deviation feature 303 as input. Accordingly, the network parameters are also adapted and adjusted to better learn the mapping relationship between quality deviation and parameter correction.

[0070] In some examples, the joint training of the residual network 1032 can also be adjusted accordingly. Specifically, the original "using the sample scaled production process parameter deviation features in the training samples and the set of scaled production historical process parameter values ​​in the historical process parameter reference data with relatively poor scaled production historical quality detection values ​​as inputs to the initial parameter branch residual network 10322" can be replaced with "using the sample scaled production quality deviation features in the training samples, the sample scaled production process parameter deviation features, and the set of scaled production historical process parameter values ​​in the historical process parameter reference data with relatively poor scaled production historical quality detection values ​​as inputs to the initial parameter branch residual network 10322", while other processes can remain unchanged.

[0071] By incorporating the above methods into the learning process of the initial parameter branch, deviation information of the mass production quality target can be integrated. This ensures that the correction value output by the initial parameter branch not only adapts to the differences between theoretical scale-up and historical parameters but also meets the needs of the mass production quality target. This further improves the accuracy of the residual correction value of the initial parameter branch, providing a more comprehensive and mass production quality requirement-compliant correction basis for the feature fusion of the subsequent fusion network. Consequently, the final mass production process parameter prediction results are optimized, ensuring that parameter optimization conforms to the theoretical scale-up criteria and accurately matches the quality target of the mass production of ternary cathode materials. This further reduces trial-and-error costs, improves process scale-up efficiency, and enhances the stability of parameter implementation.

[0072] Figure 7 An exemplary schematic diagram is shown of a process 700 for determining the initial theoretical set of process parameters for large-scale production in a process parameter optimization method for large-scale production of ternary cathode materials according to one embodiment.

[0073] like Figure 7As shown, multiple parameter scaling-up theoretical formulas can be predetermined based on different scaling-up criteria. In some examples, the scaling-up criteria may include at least one of the following: fluid dynamics similarity criterion, power / volume similarity criterion, heat / mass transfer similarity criterion, and reaction kinetics similarity criterion. In some examples, the fluid dynamics similarity criterion may refer to ensuring consistent fluid flow states by keeping dimensionless numbers such as the Reynolds number constant; the power / volume similarity criterion may refer to ensuring consistent mixing intensity by keeping energy consumption per unit volume constant; the heat / mass transfer similarity criterion may refer to ensuring the efficiency of heat and mass transfer in the reaction process by focusing on the consistency of parameters such as heat transfer coefficient and mass transfer coefficient; and the reaction kinetics similarity criterion may refer to ensuring that the kinetic characteristics of the reaction process do not change significantly after scaling up by keeping parameters such as reaction rate constant and residence time consistent.

[0074] It is understandable that different scale-up criteria and the resulting theoretical formulas for parameter scale-up can be applied to different process stages. Target process stages may include any of the following: mixing process stage, sintering process stage, pulverizing process stage, washing process stage, drying process stage, sieving process stage, demagnetizing process stage, and batch mixing process stage. For example, the mixing process section can typically employ criteria such as power / volume similarity and fluid dynamics similarity to ensure uniform material mixing; the sintering process section can employ criteria such as heat transfer / mass transfer similarity and reaction kinetics similarity to ensure temperature uniformity and reaction degree in the solid-phase reaction; the pulverizing process section can employ criteria such as constant maximum shear force to control particle size distribution during particle crushing; the washing process section can employ criteria such as fluid dynamics similarity and power / volume similarity to ensure sufficient slurry washing; the drying process section can employ criteria such as constant Froude number to control the fluidization state and heat transfer efficiency during material drying; the screening, demagnetizing, and batch mixing sections can select appropriate scale-up criteria based on their respective material flow and processing characteristics. For ease of understanding, the following explanation uses the washing process section and the impeller speed of the stirred tank as a parameter as examples.

[0075] We can first select four parameters that match the water washing process section and amplify the theoretical formulas, which are as follows: , , , .in, , , , These formulas can be used to indicate the impeller speed and diameter of the stirred tank used in pilot-scale tests, as well as the impeller speed and diameter of the stirred tank used in large-scale production. The above formulas can be derived based on the Reynolds number invariance criterion, the Froude number invariance criterion, the energy consumption per unit volume invariance criterion, and the maximum shear force invariance criterion, respectively. For example, for a stirred tank, the impeller diameter D can be used as the characteristic length L, and the tip linear velocity πND as the characteristic velocity u, such that the Reynolds number... It can be equivalent to Thus, ND can be obtained. 2 It is a constant. Similarly, the Froude number... Unchanged means (ND) 2 / gD remains unchanged, thus N can be obtained 2 D is a constant. For example, in a stirred tank, the power P is usually related to the impeller speed N and the impeller diameter D, for example, P∝N. 3 D 5 Therefore, constant energy consumption per unit volume can be equivalent to P / V∝N. 3 D 5 / D 3 Thus, N can be obtained. 3 D 2 It is a constant. For example, the maximum shear force being constant usually means that the maximum tip velocity is constant, which is directly equivalent to ND being a constant.

[0076] In one example, given the known equipment parameters of the pilot-scale test equipment (e.g., the impeller diameter D1 of the stirred tank) and the equipment parameters of the large-scale production equipment (e.g., the impeller diameter D2 of the stirred tank), four candidate sets of large-scale production process parameter values ​​corresponding to the aforementioned four parameter amplification theoretical formulas can be obtained by performing amplification calculations on the pilot-scale process parameter value set 1011. For example, each candidate set of large-scale production process parameter values ​​may include a candidate value for the impeller speed of the large-scale production stirred tank. The four candidate values ​​for the impeller speed of the large-scale production stirred tank can be expressed, for example, as follows: , , , Then, the obtained sets of candidate large-scale production process parameter values ​​can be weighted and fused with their corresponding weights to obtain the initial theoretical set of large-scale production process parameters. In this embodiment, the corresponding weights can be preset values ​​or adaptive weight values ​​trained by machine learning based on a historical process parameter reference database. In some examples, the corresponding weights can be used to characterize the adaptability and reliability of the theoretical formulas for each parameter scaling up in the target process segment. For example, for the water washing process segment, if historical data shows that the scaling up results under the constant energy consumption per unit volume criterion are more in line with actual production, then the formula corresponding to this criterion can be given a higher weight. In one example, the preset values ​​can be determined based on the experience of process experts. In another example, the adaptive weight values ​​can be trained using historical data to dynamically adapt to the scaling up requirements under different operating conditions.

[0077] It should be noted that the above-described water washing process is merely illustrative and not restrictive. For the sintering process, other formulas based on heat / mass transfer similarity and reaction kinetics similarity can be applied, such as the formula relating heating rate to furnace volume and holding time to material thickness. It should also be noted that the above example uses impeller speed as the process parameter. It is understood that the pilot-scale process parameter set 1011 and the candidate large-scale production process parameter set can also include other process parameters, such as stirring speed, material flow rate, temperature, and pressure, and other parameter scaling-up theoretical formulas can be applied accordingly. This is not limited here.

[0078] The above approach allows for the construction of multiple sets of parameter scaling-up theoretical formulas based on various scaling-up criteria. By weighted fusion, the advantages of each criterion can be integrated, avoiding the limitations of a single scaling-up criterion in complex process scenarios. Furthermore, weights can be determined by combining historical data or expert experience, ensuring that the initial theoretical values ​​of process parameters for large-scale production conform to the basic principles of process scaling-up while also adapting to the actual production characteristics of the target process segment. This significantly improves the rationality and reliability of the theoretical scaling-up results, providing a more accurate initial reference for subsequent process parameter optimization and reducing trial-and-error costs and timelines in the process from pilot-scale to large-scale production scaling-up.

[0079] Figure 8 This illustrates yet another exemplary process of a method for optimizing process parameters for the large-scale production of ternary cathode materials, according to one embodiment. Figure 8 It can be based on Figure 1 This is achieved through the process parameter optimization method shown in the figure, and Figure 8 It shows Figure 1 An exemplary implementation of step S140. Figure 8 Zhongyu Figure 1 The same parts in the embodiments can be referred to Figure 1The relevant descriptions in [the original text] will be presented here. To avoid repetition, only the differences will be described here.

[0080] like Figure 8 As shown, if step S130 determines no, the process parameter optimization model 103 can be used to obtain an adjusted set of predicted values ​​for large-scale production process parameters based on the pilot-scale to large-scale reference data 101, the matched historical process parameter reference data 1021, and the current set of predicted values ​​for large-scale production process parameters 105. This adjusted set of predicted values ​​is then used as the new set of predicted values ​​for current large-scale production process parameters. In some examples, the process parameter optimization model 103 used in the verification step can be as follows: Figure 4 The residual network 1032 described in the embodiment can be referenced. Figure 4 The exemplary parameter optimization process 400 described in the embodiment only requires... Figure 4 The initial theoretical value set 401 of the large-scale production process parameters is replaced with the current predicted value set 105 of the large-scale production process parameters. Correspondingly, Figure 4 The output set of predicted values ​​for large-scale production process parameters 104 is replaced with the adjusted set of predicted values ​​for large-scale production process parameters, i.e., the new set of predicted values ​​for large-scale production process parameters 104.

[0081] In some examples, the process parameter optimization model 103 used before proceeding to the verification step can be as follows: Figure 1 The various specific models described in the embodiments.

[0082] In the above manner, the parameters can be iteratively optimized and adjusted based on the current parameter prediction results in the verification step. The predicted value of the previous round is used as the new initial input, and the parameters are iteratively corrected with the help of the process parameter optimization model 103 adapted to the process section. This makes the parameters after each round of adjustment more in line with the actual working conditions and quality objectives of large-scale production, and thus gradually approach the optimal combination of process parameters. This effectively improves the iterative efficiency of process parameter optimization and the accuracy of the final result, and reduces the trial and error costs and cycle in the process of scaling up large-scale production.

[0083] Figure 9 Another exemplary process is shown, according to one embodiment, for optimizing process parameters for the large-scale production of ternary cathode materials. Figure 9 It can be based on Figure 1 This is achieved through the process parameter optimization method shown in the figure, and Figure 9 It shows Figure 1 An exemplary implementation of step S140. Figure 9 Zhongyu Figure 1 The same parts in the embodiments can be referred to Figure 1The relevant descriptions in [the original text] will be presented here. To avoid repetition, only the differences will be described here.

[0084] like Figure 9 As shown, if step S130 determines no, in step S910, based on the deviation 901 between the actual product quality value set 106 and the mass production quality target value set 1013, gradient path reasoning can be performed along the associated edges in the pre-constructed ternary cathode material process knowledge graph 902, based on the positive / negative correlation between quality indicators and process parameters, to obtain the process parameter correction value set 903. The pre-constructed ternary cathode material process knowledge graph 902 may include process segment nodes, quality indicator nodes, process parameter nodes, and associated edges used to characterize the relationship between quality indicator nodes and the process segment nodes, and associated edges used to characterize the positive or negative correlation between process parameter nodes and quality indicator nodes. In some examples, process segment nodes may be entity nodes characterizing each production process segment of the ternary cathode material, such as mixing process segment nodes, sintering process segment nodes, washing process segment nodes, drying process segment nodes, screening process segment nodes, etc., which may cover the aforementioned target process segments. In some examples, quality indicator nodes can be entity nodes that characterize product performance quality indicators, such as lithium residue nodes, pH value nodes, particle size distribution nodes, tap density nodes, moisture content nodes, and cycle performance nodes. In other examples, process parameter nodes can be entity nodes that characterize control variables for each process stage, such as stirring speed nodes, sintering temperature nodes, holding time nodes, and water washing duration nodes.

[0085] In some examples, the association edges used to characterize the relationship between quality indicator nodes and process segment nodes can be connection edges that identify the affiliation. For example, the lithium residue node and pH value node are connected to the water washing section node through their respective association edges, indicating that this type of quality indicator is directly determined by the water washing section. In some examples, the association edges used to characterize the positive or negative correlation between process parameter nodes and quality indicator nodes can be connection edges that identify the direction and degree of influence. For example, water washing stirring speed and material uniformity are positively correlated, water washing temperature and lithium residue are negatively correlated, and sintering temperature and grain size are positively correlated. In some examples, negative and positive correlation edges can also correspond to correlation values ​​that characterize the degree of correlation (e.g., the value range can be 0-100%). In some examples, gradient path inference can be a reasoning process that, based on the magnitude and direction of quality deviation, locates key process parameters step by step along positive or negative correlation edges and calculates the corresponding correction direction and magnitude. In some examples, gradient path reasoning can be implemented using knowledge graph reasoning engines (such as Neo4j, Apache Jena, GraphDB, etc.) combined with rule-based reasoning, graph neural networks (GNNs), or path search algorithms. For instance, the built-in rule engine of the reasoning engine can be used to generate correction paths based on preset "quality deviation-process parameter" association rules; graph neural networks (such as GCN, GAT) can be used to learn the correlation strength between nodes to help determine the direction and magnitude of correction; or breadth-first / depth-first methods can be used to search and traverse the associated edges to locate key process parameters and calculate the correction amount. For example, when the lithium residue is too high, the correction path and value for increasing the washing temperature and extending the washing time can be obtained by reasoning along the negatively correlated edges.

[0086] In this embodiment, each process parameter correction value in the process parameter correction value set 903 is used to indicate the correction direction and magnitude of the corresponding process parameter. In some examples, the process parameter correction value set 903 may be a set of parameters containing positive and negative correction values. A positive value indicates that the corresponding process parameter needs to be adjusted upwards, and a negative value indicates that the corresponding process parameter needs to be adjusted downwards; the magnitude of the value indicates the specific adjustment magnitude. For example, the correction values ​​in the process parameter correction value set 903 may be used to indicate a water washing temperature of +5°C, a stirring speed of -30 r / min, a holding time of +15 min, etc.

[0087] In step S920, the current set of predicted large-scale production process parameters 105 can be superimposed with the set of corrected process parameters 903 to obtain an adjusted set of predicted large-scale production process parameters 905, which is then used as the new set of predicted large-scale production process parameters. In some examples, the superposition operation can be to add the corresponding process parameter values ​​from the two sets.

[0088] By using the above methods, based on the prior mechanism knowledge of the ternary cathode material process knowledge graph and combined with quality deviations, interpretable and highly targeted process parameter corrections can be completed. This ensures that the parameter adjustment direction aligns with the process physicochemical mechanism, allowing for rapid approximation of optimal parameters without extensive trial and error. This improves the efficiency of parameter iterative adjustment while ensuring the rationality and reliability of the adjustment logic. Consequently, the process debugging cycle of ternary cathode materials from pilot-scale to large-scale production can be further shortened, reducing the trial and error costs and time costs of mass production scale-up.

[0089] Figure 10 An exemplary flowchart of a method 1000 for optimizing process parameters for the mass production of ternary cathode materials according to one embodiment is shown.

[0090] like Figure 10 As shown, in step S1010, pilot-scale reference data corresponding to the target process segment is obtained. The target process segment is any process segment for the large-scale production of ternary cathode materials. The pilot-scale reference data includes a set of pilot process parameter values, a set of pilot quality test values, and a set of large-scale production quality target values.

[0091] In step S1020, historical process parameter reference data matching the pilot-scale reference data is retrieved from a preset historical process parameter reference database. The historical process parameter reference data includes a set of historical process parameter values ​​for pilot-scale production, a set of historical quality test values ​​for pilot-scale production, a set of historical process parameter values ​​for large-scale production, and a set of historical quality test values ​​for large-scale production.

[0092] In step S1030, using a pre-trained process parameter optimization model corresponding to the target process segment, at least based on the pilot-scale reference data and the matched historical process parameter reference data, a set of predicted values ​​for the large-scale production process parameters corresponding to the target process segment is obtained.

[0093] In step S1040, the set of predicted values ​​for large-scale production process parameters is used as the current set of predicted values ​​for large-scale production process parameters, and the following verification steps S1050-S1080 are performed.

[0094] In step S1050, a set of actual product quality values ​​for process verification is obtained. The set of actual product quality values ​​for process verification is the product quality test results of large-scale process verification using the predicted value set of current large-scale production process parameters.

[0095] In step S1060, it is determined whether the deviation between the actual set of product quality values ​​in process verification and the set of target quality values ​​in mass production meets a predetermined deviation condition.

[0096] If step S1060 determines that it is yes, in step S1070, the current set of predicted values ​​for large-scale production process parameters is determined as the optimized set of large-scale production process parameters for the target process segment.

[0097] If step S1060 determines no, in step S1080, the current set of predicted values ​​for large-scale production process parameters is adjusted, and the adjusted set of predicted values ​​for large-scale production process parameters is used as the new set of predicted values ​​for current large-scale production process parameters. Then, the process returns to step S1050.

[0098] This disclosure achieves rapid and accurate optimization of process parameters for ternary cathode materials from pilot-scale to large-scale production by acquiring pilot-scale to large-scale reference data of the target process segment, retrieving matching historical process data, using a trained model adapted to the process segment to predict large-scale process parameters, and iteratively adjusting parameters based on process verification results. This effectively shortens the process scale-up cycle, reduces trial-and-error costs, improves the adaptability of large-scale production process parameters and product quality stability, and ensures the interpretability and reliability of the parameter optimization process.

[0099] Figure 11 A schematic block diagram of a process parameter optimization apparatus 1100 for the mass production of ternary cathode materials according to one embodiment is shown.

[0100] like Figure 11 As shown, the process parameter optimization device 1100 for large-scale production of ternary cathode materials may include: a data acquisition module 1110, a data retrieval module 1120, a parameter optimization module 1130, and a verification iteration module 1140.

[0101] The data acquisition module 1110 is configured to acquire pilot-scale reference data corresponding to the target process segment. The target process segment is any process segment for the large-scale production of ternary cathode materials. The pilot-scale reference data includes a set of pilot process parameter values, a set of pilot quality test values, and a set of large-scale production quality target values.

[0102] The data retrieval module 1120 is configured to retrieve historical process parameter reference data that matches the pilot-scale reference data from a preset historical process parameter reference database. The historical process parameter reference data includes a set of historical process parameter values ​​for pilot-scale production, a set of historical quality test values ​​for pilot-scale production, a set of historical process parameter values ​​for large-scale production, and a set of historical quality test values ​​for large-scale production.

[0103] The parameter optimization module 1130 is configured to use a pre-trained process parameter optimization model corresponding to the target process segment to obtain a set of predicted values ​​for the large-scale production process parameters corresponding to the target process segment, based at least on the pilot-scale reference data and the matched historical process parameter reference data.

[0104] The verification iteration module 1140 is configured to use the set of predicted values ​​for large-scale production process parameters as the current set of predicted values ​​for large-scale production process parameters, and to perform the following verification steps: obtaining a set of actual values ​​for process verification product quality, wherein the set of actual values ​​for process verification product quality consists of product quality test results obtained by using the current set of predicted values ​​for large-scale production process parameters; when the deviation between the set of actual values ​​for process verification product quality and the set of target values ​​for large-scale production quality meets a predetermined deviation condition, determining the current set of predicted values ​​for large-scale production process parameters as the optimized set of large-scale production process parameters for the target process segment; and when the deviation does not meet the predetermined deviation condition, adjusting the current set of predicted values ​​for large-scale production process parameters, and using the adjusted set of predicted values ​​for large-scale production process parameters as the new set of predicted values ​​for current large-scale production process parameters, and continuing to perform the verification steps.

[0105] Figure 12 Another schematic block diagram of a process parameter optimization apparatus 1200 for the mass production of ternary cathode materials according to one embodiment is shown.

[0106] The process parameter optimization apparatus 1200 for the large-scale production of ternary cathode materials may include one or more processors 1210 that execute one or more machine-readable instructions stored in a machine-readable storage medium (i.e., memory 1220). In one embodiment, the processor 1210 is configured, when executing program instructions, to perform the above-described combination of... Figures 1 to 10 The various operations and functions described herein. Those skilled in the art will understand that the apparatus described in the embodiments of this disclosure may also include various other components, such as various communication modules, bus modules, and possibly user interface modules.

[0107] According to one embodiment, a machine-readable storage medium is provided. This readable medium may store executable instructions that, when executed by processor 1210, can perform the above-described combinations of various embodiments of this disclosure. Figures 1 to 10 The description includes various operations and functions.

[0108] According to one embodiment, a computer program product is provided. The computer program product includes machine-executable instructions that, when executed by a processor 1210, are capable of performing the above-described combinations in various embodiments of this disclosure. Figures 1 to 10The description includes various operations and functions.

[0109] According to one embodiment, the process parameter optimization method for large-scale production of ternary cathode materials disclosed herein can be derived from, as follows: Figure 11 or Figure 12 The described process parameter optimization device for the large-scale production of ternary cathode materials is implemented.

[0110] The specific embodiments described above with reference to the accompanying drawings are exemplary embodiments, but do not represent all embodiments that can be implemented or fall within the scope of the claims. The term "example" as used throughout this disclosure means "serving as an example, instance, or illustration" and does not imply that it is "preferred" or "advantageous" compared to other embodiments. Specific details are included to provide an understanding of the described techniques. However, these techniques can be practiced without these specific details. In some instances, well-known structures and apparatuses are shown in block diagram form to avoid obscuring the concepts of the described embodiments.

[0111] In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0112] Not all steps and units in the above process and system structure diagrams are mandatory; some steps or units can be omitted as needed. The execution order of each step is not fixed and can be determined as required. The device structure described in the above embodiments can be a physical structure or a logical structure. That is, some units may be implemented by the same physical entity, or some units may be implemented by multiple physical entities, or they may be jointly implemented by certain components in multiple independent devices.

[0113] The foregoing description of this application is provided to enable any person skilled in the art to implement or use the application. Various modifications to the application will be apparent to those skilled in the art, and the general principles defined herein can be applied to other variations without departing from the scope of protection of this application. Therefore, this application is not limited to the examples and designs described herein, but is consistent with the widest scope of the principles and novel features disclosed herein.

Claims

1. A method for optimizing process parameters for the large-scale production of ternary cathode materials, comprising: Obtain pilot-scale reference data corresponding to the target process segment. The target process segment is any process segment for the large-scale production of ternary cathode materials. The pilot-scale reference data includes a set of pilot process parameter values, a set of pilot quality test values, and a set of large-scale production quality target values. From the preset historical process parameter reference database, retrieve historical process parameter reference data that matches the pilot-scale reference data. The historical process parameter reference data includes a set of historical process parameter values ​​for pilot-scale production, a set of historical quality test values ​​for pilot-scale production, a set of historical process parameter values ​​for large-scale production, and a set of historical quality test values ​​for large-scale production. Using a pre-trained process parameter optimization model corresponding to the target process segment, at least based on the pilot-scale reference data and the matched historical process parameter reference data, a set of predicted values ​​for the large-scale production process parameters corresponding to the target process segment is obtained. The predicted set of process parameters for large-scale production is used as the current predicted set of process parameters for large-scale production, and the following verification steps are performed: Obtain a set of actual product quality values ​​for process verification, wherein the set of actual product quality values ​​for process verification is the product quality test results of large-scale process verification using the set of predicted values ​​of the current large-scale production process parameters; When the deviation between the actual value set of product quality verified by the process and the target value set of quality in large-scale production meets the predetermined deviation condition, the predicted value set of current large-scale production process parameters is determined as the optimized set of large-scale production process parameters for the target process segment. as well as When the deviation does not meet the predetermined deviation condition, the current set of predicted values ​​for large-scale production process parameters is adjusted, and the adjusted set of predicted values ​​for large-scale production process parameters is used as the new set of predicted values ​​for current large-scale production process parameters. The verification step is then performed again. The process parameter optimization model includes a residual network. The step of using a pre-trained process parameter optimization model corresponding to the target process segment to obtain a set of predicted values ​​for the large-scale production process parameters of the target process segment, based at least on the pilot-scale reference data and matching historical process parameter reference data, includes: Based on the corresponding difference between the set of pilot-scale process parameter values ​​in the pilot-scale reference data and the set of historical pilot-scale process parameter values ​​in the matched historical process parameter reference data, a pilot-scale process parameter deviation feature is generated. Based on the corresponding difference between the set of pilot-scale quality test values ​​in the pilot-scale reference data and the set of historical pilot-scale quality test values ​​in the matched historical process parameter reference data, a pilot-scale quality deviation feature is generated. Based on the corresponding difference between the set of target quality values ​​for large-scale production in the pilot-scale reference data and the set of historical quality detection values ​​for large-scale production in the matching historical process parameter reference data, a large-scale production quality deviation feature is generated. At least the pilot-scale process parameter deviation characteristics, the pilot-scale quality deviation characteristics, the large-scale production quality deviation characteristics, and the set of historical large-scale production process parameter values ​​are input into the residual network to obtain the output set of large-scale production process parameter residual correction values; and The set of historical process parameter values ​​for large-scale production is superimposed with the set of residual correction values ​​for large-scale production process parameters to obtain the set of predicted values ​​for large-scale production process parameters for the target process segment.

2. The process parameter optimization method of claim 1, wherein, Each historical process parameter reference data is mapped to a corresponding hash bucket according to its pre-determined hash code. The hash code is obtained by normalizing each historical process parameter reference data and then encoding it using the locality-sensitive hashing algorithm. The step of retrieving historical process parameter reference data from a preset historical process parameter reference database that matches the pilot-scale reference data includes: The pilot-scale reference data is normalized and encoded using the locality-sensitive hash algorithm to obtain the hash code corresponding to the pilot-scale reference data; Historical process parameter reference data mapped to hash buckets that match the hash codes corresponding to the pilot-scale reference data are identified as candidate historical process parameter reference data. Calculate the similarity between the pilot-scale to large-scale reference data and the reference data for each candidate historical process parameter; and Candidate historical process parameter reference data whose similarity meets the predetermined similarity conditions are determined as historical process parameter reference data that match the pilot-scale reference data.

3. The process parameter optimization method as described in claim 1 further includes: Obtain the initial theoretical values ​​of process parameters for large-scale production based on the scaling-up criterion; Based on the corresponding differences between the initial theoretical value set of the large-scale production process parameters and the historical process parameter value set in the matched historical process parameter reference data, a large-scale production process parameter deviation feature is generated. The step of inputting at least the pilot-scale process parameter deviation characteristics, the pilot-scale quality deviation characteristics, the large-scale production quality deviation characteristics, and the set of historical process parameter values ​​from the large-scale production into the residual network to obtain the output set of large-scale production process parameter residual correction values ​​includes: The pilot-scale process parameter deviation characteristics, the pilot-scale quality deviation characteristics, the large-scale production quality deviation characteristics, the large-scale production process parameter deviation characteristics, and the set of historical process parameter values ​​for large-scale production are input into the residual network to obtain the output set of large-scale production process parameter residual correction values.

4. The process parameter optimization method as described in claim 3, wherein, The residual network includes a historical experience branch residual network, an initial parameter branch residual network, and a fusion network. The step of inputting the pilot-scale process parameter deviation characteristics, the pilot-scale quality deviation characteristics, the large-scale production quality deviation characteristics, the large-scale production process parameter deviation characteristics, and the set of historical large-scale production process parameter values ​​into the residual network to obtain the output set of large-scale production process parameter residual correction values ​​includes: The pilot-scale process parameter deviation characteristics, the pilot-scale quality deviation characteristics, the large-scale production quality deviation characteristics, and the set of historical process parameter values ​​for large-scale production are input into the historical experience branch residual network to obtain the output set of historical experience branch residual correction values. At least the characteristics of the deviation of the large-scale production process parameters and the set of historical process parameter values ​​of the large-scale production are input into the initial parameter branch residual network to obtain the set of output initial parameter branch residual correction values; and The set of historical experience branch residual correction values ​​and the set of initial parameter branch residual correction values ​​are input into the fusion network to obtain the output set of scaled production process parameter residual correction values.

5. The process parameter optimization method as described in claim 4, wherein, The step of inputting at least the deviation characteristics of the scaled-up production process parameters and the set of historical process parameter values ​​from the scaled-up production process into the initial parameter branch residual network to obtain the set of initial parameter branch residual correction values ​​includes: The large-scale production process parameter deviation characteristics, the large-scale production quality deviation characteristics, and the set of historical large-scale production process parameter values ​​are input into the initial parameter branch residual network to obtain the output set of initial parameter branch residual correction values.

6. The process parameter optimization method as described in claim 3, wherein, The initial theoretical value set of the large-scale production process parameters is obtained through the following steps: From a number of preset parameter amplification theoretical formulas based on amplification criteria, select several parameter amplification theoretical formulas that match the target process segment. The target process segment includes any one of the following: mixing process segment, sintering process segment, pulverizing process segment, washing process segment, drying process segment, sieving process segment, demagnetizing process segment, batch mixing process segment. The amplification criteria include at least one of the following: fluid dynamics similarity criterion, power / volume similarity criterion, heat transfer / mass transfer similarity criterion, reaction kinetics similarity criterion. The set of pilot-scale process parameter values ​​is amplified and calculated based on the selected amplification theoretical formulas to obtain a set of multiple candidate large-scale production process parameter values. as well as The obtained sets of candidate large-scale production process parameter values ​​are weighted and fused with their corresponding weights to obtain the initial theoretical set of large-scale production process parameters. The corresponding weights are either preset values ​​or adaptive weight values ​​obtained through machine learning training based on the historical process parameter reference database.

7. The process parameter optimization method as described in claim 1, wherein, The adjustment of the current set of predicted values ​​for large-scale production process parameters includes: Using the process parameter optimization model, based on the pilot-scale reference data, the matched historical process parameter reference data, and the current set of predicted process parameters for large-scale production, an adjusted set of predicted process parameters for large-scale production is obtained.

8. The process parameter optimization method as described in claim 1, wherein, The adjustment of the current set of predicted values ​​for large-scale production process parameters includes: Based on the deviation, gradient path reasoning is performed along the associated edges in the pre-constructed ternary cathode material process knowledge graph, based on the positive / negative correlation between quality indicators and process parameters, to obtain a set of process parameter correction values. The ternary cathode material process knowledge graph includes process segment nodes, quality indicator nodes, process parameter nodes, and associated edges representing the relationship between the quality indicator nodes and the process segment nodes, as well as associated edges representing the positive or negative correlation between the process parameter nodes and the quality indicator nodes. Each process parameter correction value in the set of process parameter correction values ​​indicates the correction direction and magnitude of the corresponding process parameter. The current set of predicted values ​​for large-scale production process parameters is superimposed with the set of corrected values ​​for process parameters to obtain the adjusted set of predicted values ​​for large-scale production process parameters.

9. A process parameter optimization device for large-scale production of ternary cathode materials, comprising: One or more processors; as well as One or more memories storing executable instructions that, when executed by the one or more processors, perform the operations in the process parameter optimization method as described in any one of claims 1 to 8.

Images