Method and system for optimizing control parameters of ultrahigh-nickel ternary positive electrode material mixing equipment

Abstract

The invention discloses an ultrahigh nickel ternary positive electrode material mixing equipment control parameter optimization method and system, and the method comprises the steps: arranging multi-mode sensors on stirring blades and the inner wall of equipment, obtaining temperature, humidity and vibration acoustic data, building a basic field and a motion field in a finite element model, and carrying out the optimization of the control parameters of the ultrahigh nickel ternary positive electrode material mixing equipment; spatio-temporal joint estimation of the three key distributions in the mixing process is achieved, and a mixing state field reflecting the actual mixing behavior is obtained. And based on the mixing state field, dividing the equipment space into a strong flow area, a middle flow area and a weak flow area, and outputting the mixing quality risk through a pre-trained Bayesian network in combination with the fluidity index, the volume ratio and the stirring load characteristics of each distribution. And then, a particle swarm algorithm with an intelligent agent training mechanism is adopted, a historical control record is utilized to learn an influence rule of control parameters on the three distributions, a control parameter sequence of a future control window is generated, a sequence which enables a quality risk decrease value to be maximum is selected as a current optimal control strategy, and rolling optimization of the mixing equipment is realized.

Description

Technical Field

[0001] This invention relates to the field of control optimization technology, specifically to a method and system for optimizing control parameters of ultra-high nickel ternary cathode material mixing equipment. Background Technology

[0002] Ultra-high nickel ternary cathode materials, due to their high energy density, high compaction density, and excellent electrochemical performance, place higher demands on the uniformity of the mixing process, moisture content control, and agglomeration suppression. In traditional processes, the mixing process typically relies on fixed-speed or fixed-power stirring, lacking real-time sensing capabilities of the mixing chamber's internal state. This makes it difficult to accurately grasp key parameters such as temperature distribution, humidity distribution, and material flow state, leading to frequent occurrences of localized overheating, localized moisture content shifts, and agglomeration zone formation. While existing technologies have incorporated some sensors for process monitoring, they typically only acquire data from local points, failing to provide a holistic characterization of the dynamic behavior within the mixing chamber, and further hindering accurate assessment of the mixing state through a combination of spatial and temporal characteristics.

[0003] Furthermore, ultra-high nickel materials are extremely sensitive to moisture; abnormal moisture content can lead to risks such as oxidation and crystal structure damage. Therefore, precise control of humidity changes is essential during the mixing process. Simultaneously, the complex local flow field disturbances caused by the agitator blades make traditional single-point feedback ineffective in controlling the actual mixing state. Existing methods for optimizing the mixing process are mostly based on empirical parameter tuning or simple model prediction, failing to establish an effective correlation between mixing parameters and quality risks, and lacking an intelligent decision-making mechanism capable of dynamically generating control parameter sequences.

[0004] With the development of intelligent control and the upgrading of mixing equipment structure, the combination of multimodal sensing technology, dynamic field construction technology and intelligent optimization algorithm to achieve real-time perception, quality risk assessment and adaptive control optimization of the mixing process has become a key requirement for improving the quality of ultra-high nickel ternary materials. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is that existing mixing processes lack spatial perception and risk assessment capabilities regarding temperature, humidity, and flow state. The invention aims to achieve a predictable, optimizable, and continuously updated method for generating mixing control parameters, thereby improving mixing uniformity and quality stability.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for optimizing control parameters of an ultra-high nickel ternary cathode material mixing equipment, comprising:

[0008] During the mixing process, multi-modal sensors are arranged on the stirring blades and the inner wall of the equipment to acquire multi-source data of the mixing process.

[0009] Based on the multi-source data, the mixing state in space is jointly estimated to obtain a mixing state field characterizing the spatial distribution characteristics of the mixing cavity;

[0010] After comprehensively evaluating the quality risk of the mixing state field, the maximum reduction value of the quality risk is taken as the optimization target, and the control parameters of the equipment are predicted and optimized to obtain the control parameter sequence for reducing the quality risk.

[0011] The control parameter sequence is applied to the prediction and optimization process of the control parameters, and the mixing equipment is continuously updated based on the optimized control parameters.

[0012] As a preferred embodiment of the method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment described in this invention, the multi-source data includes temperature data collected by temperature sensors arranged on the stirring blades and the inner wall of the equipment.

[0013] Moisture data on changes in the moisture content of the mixture are collected by microwave moisture content sensors arranged on the stirring blades and the inner wall of the equipment.

[0014] Acoustic data collected by vibration and acoustic sensors arranged on the stirring blades and the inner wall of the equipment to determine the flow state, changes in cavity resonance, and particle agglomeration scale.

[0015] As a preferred embodiment of the control parameter optimization method for the ultra-high nickel ternary cathode material mixing equipment described in this invention, the joint estimation includes: classifying based on the temporal changes of the sensor's position in space; constructing a basic field using the multi-source data collected by the sensor on the inner wall of the equipment; and constructing a motion field using the multi-source data collected by the sensor in the stirring blade.

[0016] The basic field and the motion field are fused to obtain the mixed state field;

[0017] The construction process of the basic field includes: taking the internal space of the mixing cavity as the target area, performing finite element discretization on the space according to the preset grid size and equipment geometric features to obtain a mixing finite element model composed of multiple finite element elements;

[0018] The installation positions of various sensors arranged on the inner wall of the equipment are established in correspondence with the finite element units, and the multi-source data is assimilated into the corresponding finite element units according to the correspondence.

[0019] For each type of sensor data, based on the physical relationship between adjacent finite element elements, spatial interpolation and smoothing are performed on the observed values ​​in each finite element element, so that the distribution of the same type of sensor data in the entire finite element model is transformed from discrete values ​​to a continuously changing distribution.

[0020] The smoothed temperature distribution, humidity distribution, and vibration acoustic characteristic distribution are aggregated to obtain a basic field integrating the three distributions;

[0021] Simultaneously, in the aforementioned basic field, the confidence level for each distribution is calculated for each finite element: the confidence level for the i-th finite element is expressed as... ;

[0022] in, This represents the confidence level of the i-th finite element under the given temperature distribution. This represents the confidence level of the i-th finite element under the humidity distribution. This represents the confidence level of the i-th finite element under the vibration acoustic characteristic distribution.

[0023] Set index ,but ;

[0024] in, This represents the confidence level of the i-th finite element under the index distribution. The index x represents the finite element set containing the sensor; r represents... The element index in; express The centroid of the nearest finite element i is the distance between the centroid of finite element i and the centroid of finite element i. This represents the weight of index r, with a value of . exist In the middle, the result after normalization;

[0025] Based on the sensors on the stirring blades, and through the construction method of the basic field, an integrated result of the temperature distribution, humidity distribution, and vibration acoustic characteristic distribution at each moment is constructed, as well as the confidence level of each finite element on each distribution, which serves as the motion field.

[0026] Furthermore, the basic field and the motion field are sorted in time to obtain field sequences for the two fields.

[0027] As a preferred embodiment of the control parameter optimization method for the ultra-high nickel ternary cathode material mixing equipment described in this invention, the state fusion includes, based on the stirring blades and sensors on the inner wall of the equipment, constructing an integrated result of the temperature distribution, humidity distribution, and vibration acoustic characteristic distribution at each moment through the basic field construction method, as the mixing state field;

[0028] In the time series, based on the similarity between the base field and the mixing state field, and between the motion field and the mixing state field at the current time, the field is divided into three regions: the region with the largest similarity between the motion field and the mixing state field greater than or equal to a preset value of 1 is designated as the strong flow region dominated by the motion field; the region with the largest similarity between the motion field and the mixing state field greater than or equal to a preset value of 2 is designated as the weak flow region dominated by the base field; and the medium flow region is outside of these two regions.

[0029] If, under any distribution, there exists a finite element between a strong flow region and a weak flow region, and the judgments of preset value 1 and preset value 2 are satisfied in the two regions respectively, then under the current distribution, the confidence level of the current finite element in the two regions is obtained respectively, and the region with higher confidence level is selected for assignment.

[0030] Each region can contain multiple sub-regions, each of which is continuous and has a similarity that satisfies a preset value of 1. The size of each sub-region is greater than the preset minimum region constraint. Furthermore, there is no overlap between regions, and the union of the three regions forms the complete device space.

[0031] As a preferred embodiment of the method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment described in this invention, the comprehensive evaluation of quality risk includes calculating the flowability index and volume ratio of the strong flow region, medium flow region, and weak flow region under each distribution, based on temperature distribution, humidity distribution, and vibration acoustic characteristic distribution.

[0032] The fluidity index and volume ratio of each region under each distribution, as well as the torque data of the stirring blade at the current moment, are input into the pre-trained Bayesian network model. The Bayesian network outputs the quality risk of the mixing process, which is used to characterize the quality risk level of the ultra-high nickel ternary cathode material under the current mixing state.

[0033] As a preferred embodiment of the control parameter optimization method for the ultra-high nickel ternary cathode material mixing equipment described in this invention, the prediction optimization includes: dividing the control process into K control windows according to time, with the control parameters in each control window remaining consistent; assuming there are currently k remaining time windows, then using the control parameter sequence before the current time window, training the agent parameters of the particle swarm optimization algorithm for each time window; using the trained particle swarm optimization algorithm to generate k control parameters for the currently remaining time windows, forming a predicted control parameter sequence;

[0034] The optimal control parameter sequence is obtained by selecting the sequence that results in the greatest reduction in quality risk.

[0035] As a preferred embodiment of the control parameter optimization method for the ultra-high nickel ternary cathode material mixing equipment of the present invention, wherein: in the particle swarm algorithm, an individual particle represents a candidate sequence of control parameters, the position of the particle represents the specific value of the control parameter sequence corresponding to the particle, and the velocity represents the adjustment direction and magnitude of the control parameter sequence in the next generation.

[0036] The training of agent parameters for the particle swarm optimization algorithm in each time window includes: constructing historical control records by utilizing the sequence of control parameters executed in previous control windows and their corresponding changes in temperature distribution, humidity distribution, and vibration acoustic feature distribution; and training the neural network parameters of the agent based on the influence relationship of different control parameters on the three distribution change trends in the historical control records, so that the agent can generate the change results of the three distributions according to the three distributions and control parameters in each time window.

[0037] By accumulating over k time windows, the final quality risk of the three distributions under the predicted control parameter sequence is obtained.

[0038] A control parameter optimization system for an ultra-high nickel ternary cathode material mixing equipment using the method described in this invention is characterized by: a data acquisition unit that acquires multi-source data of the mixing process within the equipment by arranging multi-modal sensors on the stirring blades and the inner wall of the equipment.

[0039] The computing unit performs joint estimation of the mixing state in space based on the multi-source data to obtain a mixing state field characterizing the spatial distribution characteristics of the mixing cavity.

[0040] The analysis unit performs a comprehensive evaluation of the quality risk of the mixing state field; takes the maximum reduction value of the quality risk as the optimization target, and predicts and optimizes the control parameters of the equipment to obtain the control parameter sequence for reducing the quality risk.

[0041] The update unit applies the control parameter sequence to the prediction and optimization process of the control parameters, and performs rolling updates on the mixing equipment based on the optimized control parameters.

[0042] A computer device includes: a memory and a processor; the memory stores a computer program, wherein: when the processor executes the computer program, it implements the steps of the method described in any one of the present invention.

[0043] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the present invention.

[0044] The beneficial effects of this invention are as follows: This invention collects temperature, humidity, and vibration acoustic data using multimodal sensors and combines this with finite element discretization to construct a mixing state field, achieving visualization and spatial representation of the real dynamic behavior inside the mixing cavity. By analyzing the variation patterns of three distributions in strong flow, medium flow, and weak flow regions, it accurately predicts the quality risks of the mixing process, overcoming the technical limitations of traditional mixing processes that cannot quantify flow patterns and agglomeration trends. This invention further introduces a particle swarm optimization method based on an intelligent agent training mechanism. By learning the influence of historical control parameters on the three distributions, the optimization model can adaptively reflect the spatiotemporal coupling characteristics of the mixing process, thereby automatically generating the control parameter sequence with the largest risk reduction value, realizing dynamic rolling control of the mixing equipment. This method significantly improves the uniformity and stability of mixing, reduces the incidence of moisture deviation and agglomeration defects in ultra-high nickel ternary materials, and improves product quality consistency and the level of intelligence in the production process. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 The overall flowchart of the method for optimizing control parameters of an ultra-high nickel ternary cathode material mixing equipment provided by the present invention is shown.

[0047] Figure 2 The flowchart illustrates the implementation process of a method for optimizing control parameters in a mixing equipment for ultra-high nickel ternary cathode materials, as provided by this invention. Detailed Implementation

[0048] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0049] Reference Figure 1 and 2 As an embodiment of the present invention, a method for optimizing control parameters of an ultra-high nickel ternary cathode material mixing equipment is provided, comprising:

[0050] S1: During the mixing process, multi-modal sensors are arranged on the stirring blades and the inner wall of the equipment to acquire multi-source data of the mixing process.

[0051] The multi-source data includes temperature data collected by temperature sensors arranged on the agitator blades and the inner wall of the equipment; humidity data on changes in the moisture content of the mixture collected by microwave moisture content sensors arranged on the agitator blades and the inner wall of the equipment; and acoustic data collected by vibration and acoustic sensors arranged on the agitator blades and the inner wall of the equipment to determine the flow state, changes in cavity resonance, and particle agglomeration scale.

[0052] By arranging multi-modal sensors such as temperature, microwave moisture content, vibration, and acoustic sensors on the mixing blades and inner wall of the equipment, key thermal characteristics, moisture changes, and flow physics features during the mixing process can be collected simultaneously from different dimensions. This upgrades the mixing process from "local, discrete monitoring" to "full-process, dynamic, multi-source sensing," providing a real, continuous, and fusionable data foundation for the subsequent construction of the mixing state field.

[0053] Temperature data is used to reflect exothermic reactions or localized overheating trends; humidity data is used to identify uneven moisture evaporation or abnormal water absorption by materials; vibration and acoustic data can capture key information that cannot be obtained by a single sensor, such as material flowability, cavity formation, resonance shift, and changes in agglomeration scale. By jointly acquiring the above multi-source information, the spatial distribution and dynamic evolution characteristics of the mixing process can be comprehensively characterized, ensuring that subsequent state estimation, risk analysis, and control optimization have sufficient data support and physical reliability.

[0054] S2: Based on the multi-source data, the mixing state in space is jointly estimated to obtain a mixing state field that characterizes the spatial distribution characteristics of the mixing cavity.

[0055] The joint estimation includes classification based on the temporal changes in the spatial position of the sensors: constructing a basic field using the multi-source data collected by the sensors on the inner wall of the device to characterize the contribution of sensors whose spatial positions remain unchanged in time to the state field; and constructing a motion field using the multi-source data collected by the sensors in the stirring blades.

[0056] The basic field and the motion field are fused to obtain the mixed state field.

[0057] The construction process of the basic field includes: taking the internal space of the mixing chamber as the target area, discretizing the space using finite element methods according to a preset grid size and equipment geometry, to obtain a mixing finite element model composed of multiple finite element units. A correspondence is established between the installation positions of various sensors arranged on the inner wall of the equipment and the finite element units. Based on this correspondence, the multi-source data is assimilated into the corresponding finite element units, so that each finite element unit has observation information matching its position.

[0058] For each type of sensor data, based on the physical relationship between adjacent finite element elements, spatial interpolation and smoothing are performed on the observed values ​​in each finite element element, so that the distribution of the same type of sensor data in the entire finite element model transitions from discrete values ​​to a continuously changing distribution.

[0059] The smoothed temperature distribution, humidity distribution, and vibration acoustic characteristic distribution are aggregated to obtain the base field integrating the three distributions.

[0060] It's worth noting that sensors fixed to the inner wall of the equipment possess stable spatial coordinates, and their data can form a long-term benchmark observation of the mixing space. Meanwhile, sensors arranged on the agitator blades continuously scan different areas as the blades rotate, and their temporal variations provide local dynamic compensation information. By classifying the differences between these two types of sensors in the spatial and temporal dimensions, the "background state of the static position" can be separated from the "local disturbances that change with the trajectory," avoiding the mixing of observational information. This allows for a more accurate characterization of the diffusion, heat transfer, and flow behavior during the mixing process, establishing a consistent state basis for subsequent risk assessment and control optimization.

[0061] It's also important to know that the construction of the base field employs mature numerical field reconstruction methods such as finite element discretization, interpolation, and smoothing. This method can expand discrete observation points into a spatially continuous distribution, and has been widely validated in engineering calculations. Constructing the base field using fixed sensor data, and then fusing it with the temporal information from moving sensors, aligns with the physical logic of "a stable coordinate system providing a global baseline, and a moving coordinate system supplementing local high-frequency changes," ensuring that state representation does not depend on a single observation method. Furthermore, the fusion of the base field and the motion field forms a unified mixed state field, ensuring that subsequent region division, flow inference, risk prediction, and control parameter optimization are all based on consistent state variables. This guarantees that the computational logic of the entire control chain is self-consistent, traceable, and can be implemented in a closed loop.

[0062] Simultaneously, in the aforementioned basic field, the confidence level for each distribution is calculated for each finite element: the confidence level for the i-th finite element is expressed as... It's important to note that during the construction of the basic field, the mixing chamber is discretized into multiple finite element units (FEUs). The source of state data for each FEU is not consistent: some units have corresponding type of fixed sensors, and their state values ​​are directly derived from real-time measurements; while the state values ​​of the remaining units are inferred from the observations of adjacent units through spatial interpolation and smoothing calculations. Since there is an inherent difference in reliability between directly measured data and interpolated data, a confidence level is introduced as a measure of state reliability for each FEU during the construction of the basic field. When a FEU has a corresponding type of sensor at its location, its state data has a clear observation source, and the confidence level is considered to be at its highest. When a FEU is not directly covered by this type of sensor, its state data relies on spatial inference, and the confidence level is weakened accordingly based on the spatial relationship between the unit and the nearest sensor location.

[0063] in, This represents the confidence level of the i-th finite element under the given temperature distribution. This represents the confidence level of the i-th finite element under the humidity distribution. This represents the confidence level of the i-th finite element under the vibration acoustic characteristic distribution.

[0064] Set index ,but .

[0065] in, This represents the confidence level of the i-th finite element under the index distribution. This indicates that index x corresponds to the finite element set where the sensor is located. For the finite element set corresponding to the sensor installation location, the confidence level is 1 for the sensor type data; r represents... The element index in; express The centroid of the nearest finite element i is the distance between the centroid of finite element i and the centroid of finite element i. This represents the weight of index r, with a value of . exist In the middle, the result after normalization.

[0066] Because different types of sensors vary in spatial distribution, coverage, measurement sensitivity, and susceptibility to disturbance, directly using the interpolated field as equally weighted data can amplify noise, weaken effective observations, and even lead to erroneous spatial trend judgments. Therefore, by defining "sensor coverage unit confidence as 1" and gradually attenuating the non-covered area according to the distance from the nearest sensor, the state of each finite element is made consistent with the physical distance and can also reflect the credible contribution of the observation source. This establishes a confidence model that decreases with distance and increases with observation density.

[0067] This confidence level mechanism has significant advantages. First, it can suppress pseudo-continuities introduced by the interpolation process, preventing regions far from the sensor from being falsely reinforced in the field reconstruction and improving the realism of the state field. Second, different distributions (temperature, humidity, acoustics) have independent confidence levels, effectively distinguishing the reliability characteristics of different physical quantities, allowing for targeted weight allocation in subsequent fusion processes. Third, confidence levels are naturally suitable as adjustment factors for control optimization algorithms, such as automatically guiding optimization to focus on high-confidence regions in particle swarm optimization, improving the effectiveness and stability of the control strategy.

[0068] First, the confidence model based on "distance attenuation" is a mature data credibility modeling method, widely used in sensor networks, geographic information systems, and inverse problem solving. Its physical basis is highly consistent with the diffusion and heat transfer characteristics of the mixing chamber. Second, by calculating the distance using sensor coordinates and the centroid of the finite element method, the attenuation function can be numerically stable and is not dependent on specific equipment structures, making it suitable for different types of mixing equipment. Third, normalizing the weights by distance ensures the spatial comparability of the confidence scores, avoiding numerical instability caused by inconsistent weight scales in different regions. This allows the confidence scores to directly participate in state fusion, risk assessment, and optimization algorithms, forming a complete, coherent, and closed-loop computational logical chain.

[0069] Based on the sensors on the stirring blades, and through the method of constructing the basic field, an integrated result of the temperature distribution, humidity distribution, and vibration acoustic characteristic distribution at each moment is built, as well as the confidence level of each finite element on each distribution, which serves as the motion field.

[0070] Furthermore, the basic field and the motion field are sorted in time to obtain field sequences for the two fields.

[0071] The state fusion includes, based on the sensors on the stirring blades and the inner wall of the equipment, constructing an integrated result of the temperature distribution, humidity distribution and vibration acoustic characteristic distribution at each moment using the basic field construction method, as the mixing state field.

[0072] In the time series, the field is divided into three regions based on the similarity between the base field and the mixing state field, and between the motion field and the mixing state field at the current time, respectively: the region with the largest similarity between the motion field and the mixing state field greater than or equal to a preset value of 1 is designated as the strong flow region dominated by the motion field; the region with the largest similarity between the motion field and the mixing state field greater than or equal to a preset value of 2 is designated as the weak flow region dominated by the base field; and the medium flow region is outside the two regions.

[0073] If, under any distribution, a finite element exists between a strong flow region and a weak flow region, and satisfies preset values ​​1 and 2 in each region respectively, it can be classified into either the strong or weak flow region. In this case, under the current distribution, the confidence levels of the current finite element in both regions are obtained, and the region with the higher confidence level is selected for classification. Within each region, there can be multiple contiguous sub-regions, and the similarity of each sub-region satisfies preset value 1. The size of each sub-region is greater than a preset minimum region constraint (a preset constraint condition). Furthermore, there is no intersection between regions, and the union of the three regions constitutes the complete equipment space.

[0074] It's important to note that the mixing process inherently possesses two characteristics: a "gradual overall trend" and "strong local disturbances." The base field reflects the macroscopic distribution under fixed spatial coordinates, suitable as the overall background state; the motion field reflects the rapid local changes generated by the movement of the mixing blades, and is key to identifying strong flow behavior. Therefore, comparing the similarity between these two types of fields and the actual mixing state field to distinguish different flow-dominant mechanisms is essential for constructing a regional representation that can be used for risk assessment and control optimization.

[0075] The region with the highest similarity between the motion field and the mixing state field (greater than a preset value of 1) is defined as the strong flow region. Its advantage lies in the fact that the motion field is naturally more sensitive to high-speed flow, shearing, and local turbulence; regions with high similarity can accurately reflect high-dynamic behaviors such as agglomeration and breakup, and cavity changes. Regions with a similarity greater than a preset value of 2 are defined as the weak flow region, which can reflect the "background stable structure" described by the base field and identify trends such as temperature accumulation, insufficient humidity diffusion, and slow migration. The remaining portion is designated as the medium flow region, making the region representation compatible with various flow intensities and avoiding the judgment distortion caused by the dichotomous structure. This three-region division is both interpretable and practical, significantly improving the accuracy of subsequent risk inference and control optimization.

[0076] When a finite element method simultaneously satisfies the criteria for dividing into strong and weak flow regions, its assignment is determined by comparing the confidence levels of its corresponding distributions in the two regions. The aim is to resolve the problem of data conflicts arising from "similar appearances but unreliable accuracy." Higher confidence levels indicate that the region is more fully supported by actual observations from the corresponding sensor, resulting in higher data reliability. By relying on confidence levels rather than solely on numerical determination, region assignment avoids misclassifications caused by interpolation errors or local noise, making the region division more robust.

[0077] It is important to know that similarity, as a mature indicator for measuring the consistency between two types of fields, can be used in scenarios such as spatial clustering, image segmentation, and flow field recognition, and is equally applicable in the construction of mixed state. Using the continuous maximum region as the division range can ensure that the region has spatial connectivity and engineering interpretability. Confidence, as the basis for resolving conflicts, follows the principle of information reliability and is consistent with the physical phenomena in the mixing process. The overall division method has clear hierarchical logic and executability, and can be directly used for risk indicator calculation and controller optimization, realizing a traceable chain from "state field → flow region → risk judgment".

[0078] S3: After conducting a comprehensive evaluation of the quality risk of the mixed state field, take the maximum reduction value of the quality risk as the optimization target, and predict and optimize the control parameters of the equipment to obtain the control parameter sequence for reducing the quality risk.

[0079] The control parameter sequence is applied to the prediction and optimization process of the control parameters, and the mixing equipment is continuously updated based on the optimized control parameters.

[0080] The comprehensive evaluation of quality risks includes calculating the flowability index and volume percentage of the strong flow region, medium flow region, and weak flow region under each distribution, based on temperature distribution, humidity distribution, and vibration acoustic characteristic distribution.

[0081] The fluidity index and volume ratio of each region under each distribution, as well as the torque data of the stirring blade at the current moment, are input into the pre-trained Bayesian network model. The Bayesian network outputs the quality risk (probability) of the mixing process, which is used to characterize the quality risk level of the ultra-high nickel ternary cathode material under the current mixing state.

[0082] It's important to note that temperature distribution, humidity distribution, and vibration acoustic characteristic distribution correspond to heat accumulation, moisture migration, and flow agglomeration behavior during the mixing process, respectively. These are the three most critical variables affecting the quality of ultra-high nickel ternary materials. Using these three flowability indicators—distributed in the strong, medium, and weak flow regions—along with their volume percentage as inputs is to simultaneously characterize both the "spatial range of distribution" and the "intensity of dynamic change in distribution," thereby constructing a complete representation of the mixing state. Stirring torque is the only kinetic indicator reflecting material viscosity, agglomeration strength, and local reaction resistance. Using it along with the three distributions as inputs allows the risk assessment model to simultaneously obtain information from the flow field, temperature and humidity field, and mechanical load, avoiding bias caused by judging a single variable.

[0083] The feasibility of using a pre-trained Bayesian network as a risk inference model is based on three aspects: First, Bayesian networks are naturally suitable for handling conditional dependencies between multi-source variables, and can map the joint changes of temperature, humidity, acoustics, and torque to the probability of quality risk, forming an interpretable causal inference structure; second, temperature and humidity diffusion, agglomeration changes, and torque fluctuations are essentially probabilistic and uncertain processes, and using a probabilistic model can more accurately reflect the randomness and nonlinearity of the mixing process; third, the three distributions and torque can be trained with Bayesian structures and parameters using historical samples, making the model algorithmically feasible, logically continuous and stable, and outputting a consistent risk probability when the input changes, thereby providing clear and comparable quality risk indicators for subsequent control optimization.

[0084] The prediction optimization includes dividing the control process into K control windows over time (with fixed preset step sizes), ensuring consistent control parameters within each window; assuming there are currently k remaining time windows, training the agent parameters of the particle swarm optimization algorithm for each time window using the control parameter sequences preceding the current time window; using the trained particle swarm optimization algorithm to generate k control parameters for the current remaining time windows, forming the predicted control parameter sequence; and selecting the sequence with the greatest reduction in quality risk to obtain the current optimal control parameter sequence.

[0085] By dividing the mixing process into multiple control windows over time and maintaining consistent control parameters within each window, the control strategy can have a defined operating period, avoiding mechanical fluctuations and parameter oscillations caused by high-frequency adjustments. This also facilitates the formation of a structured sequence optimization problem at the algorithm level. Training the particle swarm intelligence agent using historical control parameter sequences prior to the current window allows it to learn the mapping relationship between "control parameter changes → three-distribution evolution → risk changes." This enables the optimization model to have dynamic memory of the mixing process, thereby improving the accuracy of future window predictions. Subsequently, in the remaining k control windows, the trained agent drives the particle swarm algorithm to generate candidate control parameter sequences. This allows the algorithm to assess the impact of different control strategies on future mixing risks based on the current state, ultimately selecting the sequence with the greatest risk reduction as the optimal control strategy, thus continuously advancing the mixing process towards more stable quality.

[0086] It's important to understand that segmenting the control process into fixed windows is a mature and widely used structure in rolling optimization and model predictive control, ensuring policy stability while the system updates in real time. Training the particle swarm intelligence using historical control records allows the algorithm to move beyond random iteration during the search process, instead relying on the actual statistical patterns of the mixing process for directional updates, significantly improving convergence speed and optimization quality. Serialized predictions enable the algorithm to perform overall planning across multiple future time steps, overcoming the traditional control trap of "short-term optimality leading to long-term deterioration." Using "risk reduction" as the selection criterion ensures that the optimization objective directly corresponds to the product's core quality indicators, resulting in a clear logical closed loop with engineering feasibility and interpretable control decisions.

[0087] Furthermore, in the particle swarm optimization algorithm, an individual particle represents a candidate sequence of control parameters, the particle's position indicates the specific value of the control parameter sequence corresponding to the particle, and the velocity indicates the adjustment direction and magnitude of the control parameter sequence in the next generation. This can map the mixing control problem into a typical sequence optimization problem, facilitating the overall adjustment and coordination of the parameter sequence through swarm intelligence mechanisms. Furthermore, by constructing historical control records containing the causal relationships between "executed control parameter sequences → three distribution changes," complex mixing coupling effects can be transformed into learnable samples, enabling the agent's model to extract the intrinsic laws governing the evolution of control parameters on temperature changes, humidity diffusion, and aggregation characteristics from historical behavior.

[0088] Building upon this foundation, training the agent's neural network enables it to predict the distribution changes in the next window based on the input three distribution states and candidate control parameters. This allows the particle swarm optimization algorithm to possess "look-ahead assessment capability" during the search process, meaning it no longer relies on blind trial-and-error searching but instead pre-judges the direction of influence and risk trend of each parameter sequence on the future mixing state. By accumulating k time windows, the continuous impact of the complete control sequence on the future mixing process can be simulated, thereby obtaining the final quality risk of the three distributions under the candidate sequences and achieving multi-step risk prediction.

[0089] The process of training the agent parameters for the particle swarm optimization algorithm in each time window includes: constructing historical control records by utilizing the sequence of control parameters executed in previous control windows and their corresponding changes in temperature distribution, humidity distribution, and vibration acoustic feature distribution; and training the agent's neural network parameters based on the influence relationship between different control parameters and the three distribution trends in the historical control records, so that the agent can generate the change results of the three distributions based on the three distributions and control parameters in each time window.

[0090] By accumulating over k time windows, the final quality risk of the three distributions under the predicted control parameter sequence is obtained.

[0091] It's important to understand that the Particle Swarm Optimization (PSO) algorithm is well-suited for sequence optimization in continuous parameter spaces. When combined with learnable agent networks, it significantly improves convergence speed and global search capabilities. By fitting the mapping between control parameters and distribution variations using neural networks, the complex challenges of multi-physics coupling and temporal influences in ultra-high nickel mixing processes can be effectively addressed, avoiding the difficulties in building traditional models. Furthermore, the prediction mechanism, trained on real historical data, possesses interpretability and physical consistency, ensuring that the final control parameter sequence can be stably executed in actual mixing equipment, guaranteeing a continuous reduction in quality risks.

[0092] In this embodiment, the neural network of the agent adopts a multi-layer feedforward neural network structure. Its purpose is to model the nonlinear relationship between control parameters and three distribution changes with minimal model complexity, enabling the particle swarm optimization algorithm to quickly call upon the agent's output during iteration, thus improving the real-time performance of prediction optimization. This structure maintains good generalization ability with limited samples while possessing stable forward inference speed, meeting the real-time requirements of the short control window of the mixing equipment. Its feasibility lies in the fact that the distribution changes of temperature, humidity, and vibration acoustic characteristics during the mixing process can all be abstracted as continuous state vectors. Feedforward networks excel at approximating such multivariate nonlinear mapping relationships, thus effectively learning the implicit patterns of "control action - distribution response" in historical control records.

[0093] In other alternative embodiments, the agent's neural network can also be replaced with a Long Short-Term Memory (LSTM) network, a Graph Neural Network (GNN), or a Convolutional-Recurrent Hybrid Network (CRN) to optimize for different mixing equipment scales and different physical coupling strengths. For example, when the three distributions have significant time lags, an LSTM structure can be used to enhance time-dependent modeling; when the finite element subdivision is finer and the spatial correlation is stronger, a GNN structure can be used to capture spatial propagation paths; when there are periodic perturbations in the equipment operation, a CNN-LSTM structure can be used to extract local patterns and global trends. All of the above alternatives maintain functional consistency, and their feasibility lies in the fact that these networks can simulate distribution evolution based on the input state, thereby meeting the estimation requirements for the next state in particle swarm optimization.

[0094] On the other hand, this embodiment also provides a control parameter optimization system for ultra-high nickel ternary cathode material mixing equipment, including: a data acquisition unit, which acquires multi-source data of the mixing process by means of multi-modal sensors arranged on the stirring blades and the inner wall of the equipment.

[0095] The calculation unit performs joint estimation of the mixing state in space based on the multi-source data to obtain a mixing state field that characterizes the spatial distribution characteristics of the mixing cavity.

[0096] The analysis unit performs a comprehensive evaluation of the quality risk of the mixed state field; takes the maximum reduction value of the quality risk as the optimization target, and predicts and optimizes the control parameters of the equipment to obtain the control parameter sequence for reducing the quality risk.

[0097] The update unit applies the control parameter sequence to the prediction and optimization process of the control parameters, and performs rolling updates on the mixing equipment based on the optimized control parameters.

[0098] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0099] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0100] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Additionally, computer-readable media can be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0101] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing control parameters of a mixing equipment for ultra-high nickel ternary cathode materials, characterized in that, include: During the mixing process, multi-modal sensors are arranged on the stirring blades and the inner wall of the equipment to acquire multi-source data of the mixing process. Based on the multi-source data, the mixing state in space is jointly estimated to obtain a mixing state field characterizing the spatial distribution characteristics of the mixing cavity; After comprehensively evaluating the quality risk of the mixing state field, the maximum reduction value of the quality risk is taken as the optimization objective, and the control parameters of the equipment are predicted and optimized to obtain the control parameter sequence for reducing the quality risk. The control parameter sequence is applied to the prediction and optimization process of the control parameters, and the mixing equipment is continuously updated based on the optimized control parameters.

2. The method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment as described in claim 1, characterized in that: The multi-source data includes temperature data collected by temperature sensors arranged on the stirring blades and the inner wall of the equipment. Moisture data on changes in the moisture content of the mixture are collected by microwave moisture content sensors arranged on the stirring blades and the inner wall of the equipment. Acoustic data collected by vibration and acoustic sensors arranged on the stirring blades and the inner wall of the equipment to determine the flow state, changes in cavity resonance, and particle agglomeration scale.

3. The method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment as described in claim 2, characterized in that: The joint estimation includes classification based on the temporal changes of the sensor's position in space: constructing the basic field using the multi-source data collected by the sensor on the inner wall of the device, and constructing the motion field using the multi-source data collected by the sensor in the stirring blade. The basic field and the motion field are fused to obtain the mixed state field; The construction process of the basic field includes: taking the internal space of the mixing cavity as the target area, performing finite element discretization on the space according to the preset grid size and equipment geometric features to obtain a mixing finite element model composed of multiple finite element elements; The installation positions of various sensors arranged on the inner wall of the equipment are established in correspondence with the finite element units, and the multi-source data is assimilated into the corresponding finite element units according to the correspondence. For each type of sensor data, based on the physical relationship between adjacent finite element elements, spatial interpolation and smoothing are performed on the observed values ​​in each finite element element, so that the distribution of the same type of sensor data in the entire finite element model is transformed from discrete values ​​to a continuously changing distribution. The smoothed temperature distribution, humidity distribution, and vibration acoustic characteristic distribution are aggregated to obtain a basic field integrating the three distributions; Simultaneously, in the aforementioned basic field, the confidence level for each distribution is calculated for each finite element: the confidence level for the i-th finite element is expressed as... ; in, This represents the confidence level of the i-th finite element under the given temperature distribution. This represents the confidence level of the i-th finite element under the humidity distribution. This represents the confidence level of the i-th finite element under the vibration acoustic characteristic distribution. Set index ,but ; in, This represents the confidence level of the i-th finite element under the index distribution. The index x represents the finite element set containing the sensor; r represents... The element index in; express The centroid of the nearest finite element i is the distance between the centroid of finite element i and the centroid of finite element i. This represents the weight of index r, with a value of . exist In the middle, the result after normalization; Based on the sensors on the stirring blades, and through the construction method of the basic field, an integrated result of the temperature distribution, humidity distribution, and vibration acoustic characteristic distribution at each moment is constructed, as well as the confidence level of each finite element on each distribution, which serves as the motion field. Furthermore, the basic field and the motion field are sorted in time to obtain field sequences for the two fields.

4. The method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment as described in claim 3, characterized in that: The state fusion includes, based on the sensors on the stirring blades and the inner wall of the equipment, constructing an integrated result of the temperature distribution, humidity distribution and vibration acoustic characteristic distribution at each moment using the basic field construction method, as the mixing state field; In the time series, based on the similarity between the base field and the mixing state field, and between the motion field and the mixing state field at the current time, the field is divided into three regions: the region with the largest similarity between the motion field and the mixing state field greater than or equal to a preset value of 1 is designated as the strong flow region dominated by the motion field; the region with the largest similarity between the motion field and the mixing state field greater than or equal to a preset value of 2 is designated as the weak flow region dominated by the base field. And the medium flow area outside the two zones; If, under any distribution, there exists a finite element between a strong flow region and a weak flow region, and the judgments of preset value 1 and preset value 2 are satisfied in the two regions respectively, then under the current distribution, the confidence level of the current finite element in the two regions is obtained respectively, and the region with higher confidence level is selected for assignment. Each region can contain multiple sub-regions, each of which is continuous and has a similarity that satisfies a preset value of 1. The size of each sub-region is greater than the preset minimum region constraint. Furthermore, there is no overlap between regions, and the union of the three regions forms the complete device space.

5. The method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment as described in claim 4, characterized in that: The comprehensive evaluation of the quality risk includes calculating the flowability index and volume percentage of the strong flow region, medium flow region, and weak flow region under each distribution, based on temperature distribution, humidity distribution, and vibration acoustic characteristic distribution. The fluidity index and volume ratio of each region under each distribution, as well as the torque data of the stirring blade at the current moment, are input into the pre-trained Bayesian network model. The Bayesian network outputs the quality risk of the mixing process, which is used to characterize the quality risk level of the ultra-high nickel ternary cathode material under the current mixing state.

6. The method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment as described in claim 5, characterized in that: The prediction optimization includes dividing the control process into K control windows according to time, with the control parameters in each control window remaining consistent; assuming there are currently k remaining time windows, then using the control parameter sequence before the current time window, training the agent parameters of the particle swarm optimization algorithm for each time window; using the trained particle swarm optimization algorithm to generate k control parameters for the current remaining time window, forming the predicted control parameter sequence. The optimal control parameter sequence is obtained by selecting the sequence that results in the greatest reduction in quality risk.

7. The method for optimizing control parameters of the ultra-high nickel ternary cathode material mixing equipment as described in claim 6, characterized in that: In the particle swarm optimization algorithm, an individual particle represents a candidate sequence of control parameters, the position of the particle represents the specific value of the control parameter sequence corresponding to the particle, and the velocity represents the adjustment direction and magnitude of the control parameter sequence in the next generation. The training of agent parameters for the particle swarm optimization algorithm in each time window includes: constructing historical control records by utilizing the sequence of control parameters executed in previous control windows and their corresponding changes in temperature distribution, humidity distribution, and vibration acoustic feature distribution; and training the neural network parameters of the agent based on the influence relationship of different control parameters on the three distribution change trends in the historical control records, so that the agent can generate the change results of the three distributions according to the three distributions and control parameters in each time window. By accumulating over k time windows, the final quality risk of the three distributions under the predicted control parameter sequence is obtained.

8. A control parameter optimization system for an ultra-high nickel ternary cathode material mixing equipment using the method described in any one of claims 1-7, characterized in that: The data acquisition unit acquires multi-source data on the mixing process by using multi-modal sensors arranged on the stirring blades and the inner wall of the equipment. The computing unit performs joint estimation of the mixing state in space based on the multi-source data to obtain a mixing state field characterizing the spatial distribution characteristics of the mixing cavity. The analysis unit performs a comprehensive evaluation of the quality risk of the mixing state field; takes the maximum reduction value of the quality risk as the optimization target, and predicts and optimizes the control parameters of the equipment to obtain the control parameter sequence for reducing the quality risk. The update unit applies the control parameter sequence to the prediction and optimization process of the control parameters, and performs rolling updates on the mixing equipment based on the optimized control parameters.

9. A computer device, comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images