Iron phosphate synthesis parameter self-adaptive control system based on intelligent algorithm

Abstract

The invention belongs to the technical field of chemical process intelligent control, and discloses an iron phosphate synthesis parameter self-adaptive control system based on an intelligent algorithm. The system is composed of a target quality sensing module, a reaction data acquisition and preprocessing module, a sensitive section identification module, a process boundary modeling module, a multi-strategy control model library module, a strategy evaluation and structure evolution module and a control execution and feedback module. According to the method, a multi-module cooperation mechanism including target scoring modeling, dynamic sensitive segment identification and reaction boundary modeling is constructed, so that whole-process online identification and segmented precise control of the crystal form, the particle size and the crystallinity of a product in the iron phosphate synthesis process are realized; compared with a traditional control system depending on experience setting and univariate feedback, the system has the advantages of being higher in state sensing depth, more accurate in crystal form conversion process recognition, finer in reaction stage regulation and control and the like, and the problems of crystal form deviation, particle size drift and inter-batch quality fluctuation are effectively solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent control of chemical processes, and particularly relates to a ferriphosphate synthesis parameter adaptive control system based on an intelligent algorithm. BACKGROUND

[0002] As a precursor of lithium iron phosphate positive electrode material, the microcrystalline structure, particle size distribution and crystallinity of ferriphosphate largely determine the capacity density, rate performance and cycle stability of the final battery; especially under the background of higher requirements of new energy batteries for high specific energy, long service life and thermal stability, the process precision of ferriphosphate becomes a key link to guarantee the quality of the material; at present, the wet synthesis is still the mainstream route for ferriphosphate production, which forms a precipitate by reacting iron salt and phosphorus source under certain temperature, pH, stirring and feeding conditions, and then obtains the target product through crystallization, washing and drying.

[0003] However, the wet reaction system of ferriphosphate has characteristics such as strong nonlinearity, multivariable coupling and dynamic disturbance sensitivity; the phenomena such as crystal nucleus generation, crystal growth and particle agglomeration in the reaction process are jointly affected by factors such as temperature gradient, feeding rate, raw material concentration change and mixing efficiency, resulting in easy fluctuation of product particle size distribution and difficulty in accurate control of crystal structure; the traditional control strategy is often based on fixed threshold setting or empirical rule design, lacks continuous sensing ability for the dynamic evolution process of the product, and cannot adjust and control the path in real time according to the crystal state or conversion rate; in addition, these control systems usually rely on a single model for operation, the controller structure is fixed, lacks strategy evaluation, switching and evolution mechanism, and it is difficult to restore precision control once the set working condition is deviated. SUMMARY

[0004] The purpose of the present application is to provide a ferriphosphate synthesis parameter adaptive control system based on an intelligent algorithm to solve the problems raised in the background.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme: a ferriphosphate synthesis parameter adaptive control system based on an intelligent algorithm, which is composed of a target quality sensing module, a reaction data acquisition and preprocessing module, a sensitive section identification module, a process boundary modeling module, a multi-strategy control model library module, a strategy evaluation and structure evolution module and a control execution and feedback module; The target quality sensing module is used to acquire and construct an ideal parameter model of the particle size distribution, crystallinity and crystal type characteristics of the target product, and to complete the target score based on the real-time acquired product data by using the KL divergence formula; The reaction data acquisition and preprocessing module is used to acquire parameters such as temperature, pH, conductivity, feeding concentration and stirring rate in the reaction process, and to perform data preprocessing through the sliding window standardization formula; The sensitive segment identification module is used to identify the sensitive segments of crystal form changes based on the target score sequence and product time series using the dynamic time warping (DTW) formula, and output the priority segments for regulation; The process boundary modeling module is used to build a process model based on the reaction rate relationship and use the multi-component reaction rate kinetics formula to determine the control boundaries of each parameter; A multi-strategy control model library module is used to construct at least three types of control strategy models with different structures and output control action candidates based on the current state vector. The models all generate strategy actions based on the strategy mapping function formula; The strategy evaluation and structure evolution module is used to score and rank candidate control strategies based on the current target score and response characteristics, and uses a multi-factor scoring function formula to determine the strategy retention and structure update mechanism; The control execution and feedback module is used to execute the current optimal strategy output control action, adjust the controller parameters based on the controller gradient adaptive update formula, and form a feedback closed loop.

[0006] Preferably, the target quality perception module includes: (1) Collect product data with excellent crystal form, concentrated particle size distribution and moderate crystallinity from historical production samples, and establish an ideal parameter model. This model is used to describe the distribution curve of the target product in the particle size space, the crystallinity range and the crystal stability threshold, and has traceability and standard constraints. By normalizing the representative samples, a probability distribution template for scoring is generated, forming a basic standard reflecting the characteristics of the ideal product, ensuring that the subsequent scoring has a unified reference system and physical meaning; (2) When the system is running, the data on the particle size distribution and crystallization performance of the current batch of products are collected in real time, and matched and compared with the aforementioned standard model. The difference between the probability distribution of the current product and the ideal model distribution is quantitatively analyzed by introducing the KL divergence formula, and the scoring result is output; the score reflects the degree of match between the current process output and the target crystal form characteristics, and can be used as a direct basis for subsequent control strategy calls to achieve refined closed-loop adjustment driven by the score; KL divergence scoring formula (Kullback-Leibler Divergence) ; Where, : The target product is The ideal probability distribution of particle size intervals or crystal form dimensions; : Real-time probability distribution of the current product in the same dimension; : total number of discrete distribution intervals; : The quality deviation score between the current product and the daily standard model; Technical effects: Accurate measurement of the difference between the target product and the real-time product in particle size / crystallization behavior, as a scoring feedback signal for strategy optimization driving; Source: KL divergence principle in information theory, suitable for distribution difference measurement in probability space.

[0007] Preferably, the reaction data acquisition and preprocessing module comprises: (1) By deploying multiple types of sensors at key positions of the reaction kettle, real-time acquisition of core parameter data during the reaction process is carried out, including temperature, pH value, conductivity, feed concentration and stirring rate. The acquired parameters have multi-dimensional heterogeneity, including both continuous time series data and batch processing characteristic information sampled intermittently. The system sets a sampling period and a synchronization mechanism to ensure that different data streams have a unified time reference, providing stable and reliable input for subsequent state modeling and control feedback; (2) To solve the problems of inconsistent dimensions, numerical scale differences and abnormal disturbances in the original collected data, this module introduces a sliding window standardization mechanism to perform mean and variance statistics on various process parameters within a local time window, and completes the normalization conversion accordingly. This processing method not only retains the original dynamic trend, but also eliminates atypical noise interference, so that different types of parameters can participate in state modeling and strategy calculation in a unified numerical space, thereby enhancing the perception stability of the control system to state fluctuations; The expression of the sliding window standardization mechanism is: ; In the formula, : The original sensor parameter value (temperature, pH, etc.) collected at time point t; : The normalized standard value; , : The mean and standard deviation within the window; Technical effects: Time domain standardization processing of sensor data to eliminate non-stationarity and dimension differences, and to improve the robustness of the subsequent identification module; Source: Sliding standardization method commonly used in time series modeling.

[0008] Preferably, the sensitive section identification module comprises: (1) Based on the constructed target score sequence and the real-time product performance change curve, dynamic comparison analysis of the key period is carried out. The score sequence reflects the fitting degree between the product quality and the target model, and the product performance sequence includes particle size evolution, crystallization degree and conductivity performance. The two types of curves are aligned through the common time axis to establish the time sequence correlation basis of score fluctuation and product performance change as a prerequisite for identifying the key section of fluctuation; (2) On the basis of the constructed time sequence correlation, the system introduces a dynamic time warping mechanism to identify the position section with significant nonlinear deviation between the score curve and the product performance curve; this mechanism can automatically align the variation position and highlight the sensitive area of rapid evolution of product crystal form or index mutation. The system outputs the regulatory priority section label based on the identification results to guide the subsequent control strategy to allocate resources preferentially to the process fluctuation sensitive window, achieving dynamic and accurate regulation; The dynamic time warping (DTW) formula is: ; In the formula, : the time point of the target score sequence is; : the time point of the product time sequence is; : the DTW cumulative shortest distance at the coordinate : the initial condition is , and the rest are ; ; ; Technical effect: accurately identify the section position sensitive to crystal form change, for dynamic key regulation of the downstream strategy module; Source: dynamic time warping algorithm (for nonlinear time alignment), which has been widely used in time sequence sensitive section extraction.

[0009] Preferably, the process boundary modeling module comprises: (1) Based on the main reaction path of the iron phosphate synthesis process, a rate modeling system is established with iron salt concentration, phosphate concentration, temperature and pH as key variables; by analyzing experimental data, the mathematical dependence relationship between the generation rate and multiple reaction conditions is constructed, which is used to quantify the influence of process input parameters on product conversion efficiency, and provides a reliable theoretical basis for parameter regulation; (2) The system combines the target crystal form stable interval with the historical product performance, and calculates the safe regulation range of each key variable using the established reaction rate model; the determined boundary value covers the dynamic upper and lower limits of temperature, feed concentration and pH variables, which is used to constrain the adjustable space of the control strategy, ensuring that the reaction process runs stably within the target quality range and avoids abnormal deviation; The multi-component reaction rate boundary formula is: ; ; In the formula, : reaction rate (mol·L⁻¹·s⁻¹); , , : concentration of each reactant; , , : measured reaction order; : rate constant; : frequency factor; : activation energy (J / mol); : gas constant; : reaction temperature (K); Technical effects: Accurate modeling of reaction behavior and determination of adjustable control boundaries for each reactant concentration, temperature, and stirring rate provide parameter space constraints for strategy generation; Source: Classical chemical kinetics and Arrhenius rate theory.

[0010] Preferably, the multi-strategy control model library module comprises: (1) To improve the environmental adaptability and strategy robustness of the control system, this module presets to construct at least three types of control strategy models with different structural characteristics, typically including rule-based controllers, traditional PID controllers, and neural network controllers with learning ability; each model can be independently run and has a heterogeneous control path to meet the collaborative needs of various control mechanisms under complex reaction conditions; To ensure the stability and effectiveness of the control system under parameter perturbation, environmental fluctuations, or model uncertainty, this module introduces a robustness design principle when constructing control strategy models; each model considers the sensitivity of external disturbances to control output during training or construction, and adjusts the strategy mapping structure, introduces redundant action paths, or sets a tolerance range to improve its adaptability to abnormal states; during system operation, error tolerance analysis and output fluctuation monitoring mechanisms are used to evaluate the stability of strategy output under different interference scenarios, thus selecting the optimal strategy that balances control accuracy and robustness, and achieving multi-scenario adaptive control.

[0011] Strategy robustness specifically refers to: When the concentration of raw materials fluctuates (such as inconsistent concentrations of iron or phosphorus sources); Or environmental factors change (such as temperature changes, sensor measurement errors); Or there are slight nonlinear disturbances in the intermediate process (such as changes in reaction rate); The control system's strategy model (such as rule control, neural network control, etc.) can still generate reasonable control actions to maintain crystal quality and reaction stability without output anomalies or system failure.

[0012] (2) During operation, the system activates each type of control model and generates control action candidates based on the current process state vector; each model internally maps the current state input to the corresponding adjustment action output through a strategy mapping function, forming a multi-strategy response set; the optimal action will be selected from this set based on the target score and predicted results, achieving a balance between response speed and control accuracy.

[0013] The policy mapping function expression is: ; In the formula, : the control action output by the th policy model under the state ; : the policy function with parameters (which can be a DNN, a tree model, or a rule system); : the state vector at the current time; : the policy model number (at least three types of model structures); Technical effects: multiple structure heterogeneous control candidates are generated based on the current system state, improving the diversity and generalization ability of control actions. Source: the policy function mapping idea comes from the policy optimization structure of reinforcement learning and control theory.

[0014] Preferably, the policy evaluation and structure evolution module comprises: (1) receiving candidate control actions output from each control model, and combining the current process state, historical feedback curve, and target score result to quantitatively evaluate the regulation effect of each policy; the system performs weighted processing on control accuracy, response speed, and system stability through a multi-factor scoring function, outputs the comprehensive score corresponding to each policy, and ensures that the selection logic has comparability and interpretability; The multi-factor scoring function and the structure updating formula are: ; ; In the formula, : the th policy score; : the improvement degree of the quality score before and after the policy; : the policy stability score; : the change amount of yield or rate; : the weighting coefficient; : the th policy model parameter; : the learning rate; : the gradient of the policy parameter; Technical effects: dynamic comprehensive scoring realizes the survival of the fittest of policies, guiding the self-evolution and updating of model structures. Source: policy screening and structure evolution principle in multi-objective reinforcement learning; (2) Based on the scoring results, the system ranks the candidate control strategies according to the scores, and selects some strategies according to the set evolution mechanism to remain for the next round of control; for the control structure with long-term lag in scoring, the strategy replacement or structure fine-tuning process is automatically triggered to form a dynamically evolving control model set; this mechanism improves the adaptability and evolution ability of the overall control system, and ensures continuous optimization of the control path under complex working conditions.

[0015] Preferably, the control execution and feedback module comprises: (1) The optimal control strategy confirmed by the strategy ranking module is received, and the corresponding control action is issued to the execution layer to adjust the key variables in the reaction process, such as temperature, pH, feeding rate and stirring speed; the system is provided with a control signal buffer and an actuator linkage mechanism to ensure the continuity and response stability of the control command during the switching process of multiple models, avoiding system oscillation caused by mutations or disturbances; (2) After the execution of the control action, the system synchronously monitors the deviation between the actual response result and the target expectation, and adjusts the internal parameters of the current strategy model based on the controller gradient adaptive update mechanism; this mechanism optimizes the controller performance through error back propagation, and feeds back the update result to the strategy evaluation module to form a cross-module closed-loop linkage; this process realizes the continuous learning and dynamic self-optimization ability of the control system to product fluctuations; The expression formula of the controller gradient adaptive update mechanism is: ; In the formula, The current controller parameters; The learning rate; : The loss function under the current strategy control (for example delayed feedback + parameter penalty); The gradient of the controller loss function; Technical effect: Based on the continuous update closed-loop mechanism of feedback control, the strategy deviation is corrected in real time to ensure the stability and continuous optimization ability of the control; Source: Adaptive control and gradient update mechanism in control theory.

[0016] The beneficial effects of the present application are as follows: 1. The present application realizes online identification and segmented precision control of the product crystal form, particle size and crystallinity in the synthesis process of iron phosphate by constructing a multi-module cooperative mechanism including target scoring modeling, dynamic sensitive segment identification and reaction boundary modeling; compared with the traditional control system relying on experience setting and single variable feedback, the present system has higher state perception depth, more accurate identification of crystal form transformation process and more precise regulation of reaction stages, effectively solves the problems of crystal form deviation, particle size drift and batch-to-batch quality fluctuation, and significantly improves the consistency and controllability of iron phosphate materials.

[0017] 2、The application has the ability of flexible switching and self-optimization for different working conditions by controlling at least three structural isomers in the control model library and assisting with the strategy scoring and ranking mechanism and the structure evolution mechanism; through the cooperative action of the strategy mapping function and the scoring function, the system can realize real-time selection and update reservation among multiple strategies, solve the problem of single strategy and easy failure in the prior art, significantly enhance the fault tolerance of the control system to process disturbance and raw material fluctuation, improve the robustness and long-term effectiveness of the control strategy under complex reaction conditions, and meet the stability requirements of industrial continuous production.

[0018] 3、The application realizes the dual adaptive evolution of the parameter layer and the structure layer by constructing a complete closed-loop control logic chain, starting from self-scoring driving, executing the optimal action after strategy selection, and automatically triggering the controller gradient update based on the feedback result; a causal closed loop is formed between the control output and the product evaluation, the system can continuously correct the internal model error and control deviation, and overcome the problem that the traditional static control cannot self-learn; the mechanism strengthens the linkage of each module in the control chain, ensures the long-term stability, evolution ability and learning ability of the product quality control, and embodies the significant system-level innovation advantage. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The application is a phosphorus acid iron synthesis parameter adaptive control system based on intelligent algorithm. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0021] As shown in the figure, Figure 1 The application provides a phosphorus acid iron synthesis parameter adaptive control system based on intelligent algorithm, which is composed of a target quality perception module, a reaction data acquisition and preprocessing module, a sensitive section identification module, a process boundary modeling module, a multi-strategy control model library module, a strategy evaluation and structure evolution module, and a control execution and feedback module. The target quality perception module is used to acquire and construct an ideal parameter model of the particle size distribution, crystallinity and crystal form characteristics of the target product, and complete target scoring based on the real-time acquired product data by using the KL divergence formula. The reaction data acquisition and preprocessing module is used for acquiring parameters such as temperature, pH, conductivity, feed concentration and stirring rate during the reaction process, and performing data preprocessing through a sliding window standardization formula. The sensitive section identification module is used for identifying a sensitive section of crystal form change based on a target score sequence and a product time sequence through a dynamic time warping (DTW) formula, and outputting a regulation priority section. The process boundary modeling module is used for constructing a process model based on a reaction rate relationship, and determining the regulation boundary of each parameter by using a multi-component reaction rate kinetics formula. The multi-strategy control model library module is used for constructing at least three types of control strategy models with different structures, and outputting control action candidates according to a current state vector, wherein the models are based on a strategy mapping function formula to generate strategy actions. The strategy evaluation and structure evolution module is used for scoring and sorting candidate control strategies according to a current target score and response characteristics, and determining a strategy retention and structure updating mechanism by using a multi-factor score function formula. The control execution and feedback module is used for executing a control action output by a current optimal strategy, adjusting controller parameters based on a controller gradient self-adaptive updating formula, and forming a feedback closed loop.

[0022] Embodiment one: particle size structure adaptive control based on a target score model In this embodiment, a ternary co-precipitation process is used to synthesize iron phosphate by wet method, the precursors of the reaction are FeSO4 and H3PO4, a temperature and pH controlled reaction device is used, particle size distribution, crystallinity and crystal structure spectrum data during the formation of the precipitate are collected by the system, and a KL divergence score function is used to construct a target product model. The real-time collected data are compared with the target distribution to generate a score sequence, which is fed back to the strategy evaluation module to adjust the priority of the strategy. When the score is lower than a threshold value, the system actively switches the regulation strategy and re-fits the stirring rate and feed concentration parameters, so as to realize continuous convergence of the precipitate particle size distribution from a bimodal structure to a unimodal structure, effectively suppress agglomeration and improve particle size consistency.

[0023] Embodiment two: DTW identification of sensitive sections and construction of a reaction boundary model The embodiment monitors the temperature, pH, conductivity, reaction heat data and product crystal form change trend of the whole process of iron phosphate synthesis in a continuous reaction device; the system uses the dynamic time warping (DTW) method to compare the product score sequence with the crystal form pattern evolution time sequence, automatically identifies the sensitive section before the crystal form mutation, and outputs the time window; then the multi-component reaction kinetics equation is used to extract the change relationship of the reaction rate in the section with stirring speed and pH, and a process parameter control boundary function is constructed to provide effective control boundary input for the multi-strategy controller; in the sensitive section, the crystal form conversion rate is precisely disturbed and controlled to convert the final product from the non-target crystal form to the LiFePO4 crystal structure, thereby improving the first product yield.

[0024] Example Three: Multi-strategy control and feedback evolution closed-loop self-optimization example In this embodiment, the system loads three types of control strategies: model predictive control (MPC), reinforcement learning strategy control (DRL), and fuzzy control; before each strategy output, the system generates control action candidates based on the current state vector and score history through a strategy mapping function, and evaluates the strategy control effect (including crystal form deviation, energy consumption, response speed, etc.) with a multi-factor scoring function; the strategy with the lowest score enters the structure update process, and the internal weights are adjusted or the control structure is replaced through a controller gradient adaptive update function; after the controller executes the control action, the actual crystal form feedback and reaction response error are used to automatically update the scores of each strategy and feed back to the evaluation module, forming a closed-loop control of strategy-execution-feedback-update, which significantly improves the strategy stability and quality consistency of the system under long-term operation.

[0025] The target quality perception module refers to collecting product data with excellent crystal form, concentrated particle size distribution, and moderate crystallinity from historical production samples, establishing an ideal parameter model, which is used to describe the distribution curve of the target product in the particle size space, the crystallinity range, and the crystal form stability threshold, and has traceability and standard constraint; by normalizing representative samples, a probability distribution template for scoring is generated, forming a basic standard reflecting the characteristics of ideal products, ensuring that subsequent scoring has a unified reference system and physical meaning; during system operation, the particle size distribution and crystallization performance data of the current batch of products are collected in real time and compared with the aforementioned standard model; the difference between the probability distribution of the current product and the ideal model distribution is quantitatively analyzed by introducing the KL divergence formula, and the scoring result is output. The score reflects the matching degree between the current process output and the target crystal form characteristics, and can be used as a direct basis for subsequent control strategy calls, realizing fine closed-loop adjustment driven by scoring.

[0026] Among them, the reaction data acquisition and preprocessing module refers to the real-time acquisition of core parameter data in the reaction process through the deployment of multiple types of sensors at key positions of the reaction kettle, including temperature, pH value, conductivity, feed concentration and stirring rate. The collected parameters have multi-dimensional heterogeneity, including both continuous time series data and batch processing characteristic information collected intermittently. The system sets a sampling period and a synchronization mechanism to ensure that different data streams have a unified time reference and provide stable and reliable input for subsequent state modeling and control feedback. To solve the problem of inconsistent dimensions, numerical scale differences and abnormal disturbances in the original collected data, this module introduces a sliding window standardization mechanism to calculate the mean and variance of each type of process parameter within a local time window and complete the normalization conversion accordingly. This processing method not only retains the original dynamic trend, but also eliminates atypical noise interference, so that different types of parameters can participate in state modeling and strategy calculation in a unified numerical space, thereby enhancing the perception stability of the control system to state fluctuations.

[0027] Among them, the sensitive section identification module refers to the dynamic comparison and analysis of the key period based on the constructed target score sequence and the real-time product performance change curve. The score sequence reflects the fitting degree between the product quality and the target model, and the product performance sequence includes particle size evolution, crystallization degree and conductivity. The two types of curves are aligned through the common time axis to establish the time sequence correlation basis of score fluctuation and product performance change as a prerequisite for identifying the key section of fluctuation. On the basis of the established time sequence correlation, the system introduces a dynamic time warping mechanism to identify the position section where there is a significant nonlinear deviation between the score curve and the product performance curve. This mechanism can automatically align the variation position and highlight the sensitive area of rapid evolution of product crystal form or index mutation. The system outputs the regulation priority section label based on the identification result to guide the subsequent control strategy to allocate resources preferentially to the process fluctuation sensitive window, realizing dynamic and accurate regulation.

[0028] Among them, the process boundary modeling module refers to establishing a rate modeling system with iron salt concentration, phosphate concentration, temperature and pH as key variables based on the main reaction path of the iron phosphate synthesis process. By analyzing experimental data, the mathematical dependence relationship between the generation rate and multiple reaction conditions is constructed to quantify the influence of process input parameters on product conversion efficiency and provide a reliable theoretical basis for parameter regulation. The system combines the target crystal form stable interval and the historical product performance to calculate the safe regulation range of each key variable using the established reaction rate model. The determined boundary values cover the dynamic upper and lower limits of temperature, feed concentration and pH variables to constrain the adjustable space of the control strategy, ensuring that the reaction process operates stably within the target quality range and avoids abnormal deviation.

[0029] Among them, the multi-strategy control model library module refers to in order to improve the environmental adaptability and strategy robustness of the control system, this module presets to build at least three types of control strategy models with different structural characteristics, typically including rule-based controller, traditional PID controller and neural network controller with learning ability; each model can be independently run and has a heterogeneous control path to meet the collaborative needs of multiple control mechanisms under complex reaction conditions; during operation, the system activates each type of control model and generates control action candidates according to the current process state vector; each model internally maps the current state input to the corresponding adjustment action output through a strategy mapping function, forming a multi-strategy response set; the optimal action will be selected from the response set according to the target score and prediction results, realizing the collaborative guarantee of response speed and control accuracy.

[0030] Among them, the strategy evaluation and structure evolution module refers to receiving candidate control actions output from each control model, and combining the current process state, historical feedback curve and target score results to quantitatively evaluate the regulation effect of each strategy; the system outputs the comprehensive score corresponding to each strategy by weighting the control accuracy, response speed and system stability through a multi-factor scoring function, ensuring that the optimization selection logic is comparable and interpretable; based on the scoring results, the system sorts the candidate control strategies according to the score from high to low, and selects some strategies to be retained for the next round of control according to the set evolution mechanism; for control structures with long-term lag in scoring, the strategy replacement or structure fine-tuning process is automatically triggered to form a dynamically evolving control model set; this mechanism improves the adaptability and evolution ability of the overall control system, ensuring continuous optimization of the control path under complex conditions.

[0031] Among them, the control execution and feedback module refers to receiving the optimal control strategy confirmed by the strategy sorting module and issuing the corresponding control action to the execution layer for adjusting key variables such as temperature, pH, feeding rate and stirring speed during the reaction process; the system has a control signal buffer and actuator linkage mechanism to ensure the continuity and response stability of the control command during multi-model switching, avoiding system oscillation caused by mutations or disturbances; after the control action is executed, the system synchronously monitors the deviation between the actual response result and the target expectation, and adjusts the internal parameters of the current strategy model based on the controller gradient adaptive update mechanism; this mechanism optimizes the controller performance through error backpropagation and feeds back the update results to the strategy evaluation module, forming a cross-module closed-loop linkage; this process realizes the continuous learning and dynamic self-optimization ability of the control system to product fluctuations.

[0032] It is to be understood that the terminology used herein such as first and second, and the like, is only used to distinguish one entity or action from another entity or action, and does not necessarily require or imply any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0033] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. An adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm, characterized by: The system includes; The target quality perception module is used to collect and build an ideal parameter model for the particle size distribution, crystallinity and crystal form characteristics of the target product, and complete the target scoring using the KL divergence formula based on the real-time collected product data; The reaction data acquisition and preprocessing module is used to collect parameters such as temperature, pH, conductivity, feed concentration, and stirring rate during the reaction process, and perform data preprocessing using a sliding window standardization formula; Sensitive segment identification module, which is used to identify the sensitive segments of crystal form changes based on the target score sequence and product time series through the dynamic time warping formula, and output the priority segments for regulation; The process boundary modeling module is used to build a process model based on the reaction rate relationship and use the multi-component reaction rate kinetics formula to determine the control boundaries of each parameter; A multi-strategy control model library module is used to construct at least three types of control strategy models with different structures and output control action candidates based on the current state vector. The models all generate strategy actions based on the strategy mapping function formula; The strategy evaluation and structure evolution module is used to score and rank candidate control strategies based on the current target score and response characteristics, and uses a multi-factor scoring function formula to determine the strategy retention and structure update mechanism; The control execution and feedback module is used to execute the current optimal strategy output control action, adjust the controller parameters based on the controller gradient adaptive update formula, and form a feedback closed loop.

2. The adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm according to claim 1, characterized in that: The target quality perception module includes: (1) Collect product data with excellent crystal form, concentrated particle size distribution and moderate crystallinity from historical production samples, and establish an ideal parameter model. This model is used to describe the distribution curve of the target product in the particle size space, the crystallinity range and the crystal stability threshold, and has traceability and standard constraints. By normalizing the representative samples, a probability distribution template for scoring is generated, forming a basic standard reflecting the characteristics of the ideal product, ensuring that the subsequent scoring has a unified reference system and physical meaning; (2) When the system is running, the data on the particle size distribution and crystallization performance of the current batch of products are collected in real time, and matched and compared with the aforementioned standard model. By introducing the KL divergence formula, the difference between the probability distribution of the current product and the ideal model distribution is quantitatively analyzed, and the scoring results are output.

3. The adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm according to claim 1, characterized in that: The reaction data acquisition and preprocessing module includes: (1) By deploying multiple types of sensors at key locations in the reactor, core parameter data during the reaction process are collected in real time, including temperature, pH value, conductivity, feed concentration, and stirring rate. The collected parameters are multi-dimensional and heterogeneous, including both continuously changing time series data and batch feature information of intermittent sampling. The system sets the sampling cycle and synchronization mechanism to ensure that different data streams have a unified time base; (2) In order to solve the problems of inconsistent dimensions, differences in numerical scales and abnormal disturbances in the original collected data, this module introduces a sliding window normalization mechanism to perform mean and variance statistics on various process parameters within a local time window, and complete the normalization conversion based on this.

4. The adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm according to claim 1, characterized in that: The sensitive section identification module includes: (1) Based on the constructed target scoring sequence and the real-time product performance change curve, a dynamic comparative analysis of key periods is carried out. The scoring sequence reflects the degree of fit between the product quality and the target model, and the product performance sequence includes particle size evolution, crystallinity, and electrical conductivity. (2) Based on the constructed temporal association, the system introduces a dynamic time warping mechanism to identify the position segments where there is a significant nonlinear offset between the scoring curve and the product performance curve.

5. The adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm according to claim 1, characterized in that: The process boundary modeling module includes: (1) Based on the main reaction pathways of the ferric phosphate synthesis process, a rate modeling system was established with iron salt concentration, phosphate concentration, temperature, and pH as key variables. By analyzing experimental data, a mathematical dependency relationship between the generation rate and multiple reaction conditions was constructed to quantify the impact of process input parameters on product conversion efficiency, providing a reliable theoretical basis for parameter control. (2) The system combines the target crystal stability range with historical product performance and uses the established reaction rate model to calculate the safe control range of each key variable; the determined boundary values ​​cover the dynamic upper and lower limits of temperature, feed concentration, and pH variables.

6. The adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm according to claim 1, characterized in that: The multi-strategy control model library module includes: (1) To improve the environmental adaptability and strategy robustness of the control system, this module pre-sets the construction of at least three types of control strategy models with different structural characteristics, typically including rule-based controllers, traditional PID controllers, and neural network controllers with learning capabilities; each model can operate independently and has heterogeneous control paths to meet the collaborative needs of diverse control mechanisms under complex reaction conditions; (2) During operation, the system dynamically activates various control models based on the current process state vector and generates control action candidates; each model uses a strategy mapping function to map the current state input to the corresponding adjustment action output, forming a multi-strategy response set.

7. The adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm according to claim 1, characterized in that: The strategy evaluation and structure evolution module includes: (1) Receive candidate control actions from each control model output, and quantitatively evaluate the control effect of each strategy by combining the current process status, historical feedback curve and target score results; the system uses a multi-factor scoring function to weight control accuracy, response speed and system stability, and outputs a comprehensive score corresponding to each strategy to ensure that the selection logic is comparable and explainable; (2) Based on the scoring results, the system sorts the candidate control strategies according to their scores and selects some strategies to be retained for the next round of control according to the set evolution mechanism; for control structures with long-term lagging scores, the strategy replacement or structure fine-tuning process is automatically triggered to form a dynamically evolving control model set.

8. The adaptive control system for ferric phosphate synthesis parameters based on an intelligent algorithm according to claim 1, characterized in that: The control execution and feedback module includes: (1) Receive the optimal control strategy confirmed by the strategy ranking module and send its corresponding control action to the execution layer to adjust the key variables in the reaction process, such as temperature, pH, feeding rate and stirring speed; the system is equipped with a control signal buffer and actuator linkage mechanism to ensure the continuity and response stability of the control command during the multi-model switching process; (2) After the control action is executed, the system synchronously monitors the deviation between the actual response result and the target expectation, and adjusts the internal parameters of the current strategy model based on the controller gradient adaptive update mechanism; this mechanism optimizes the controller performance through error backpropagation and feeds the updated results back to the strategy evaluation module to form a cross-module closed-loop linkage; this process realizes the control system's continuous learning and dynamic self-optimization capabilities for product fluctuations.

Images