Optimization method for drug preparation

Abstract

The present invention relates to an optimization method for the preparation of a medicine. A machine learning model of a deep neural network is used to screen and optimize the active ingredients and excipients of a medicine, and a relationship model between ingredients and efficacy and stability is established; a dynamic formulation module is used to adjust the proportion and dosage of ingredients in real time according to different medicine preparation conditions including temperature and pH value; an adaptive reaction condition adjustment system is constructed to monitor the reaction conditions including temperature, pressure, pH value, and stirring speed in the preparation process in real time; a dynamic mixing system is introduced, and microfluidic technology is used to control the metering and mixing of medicine ingredients; the medicine preparation process is decomposed into multiple key steps including primary reaction, purification, and concentration, and each step is optimized separately; a real-time quality monitoring system is constructed, and the physical and chemical properties of the medicine including particle size distribution, solubility, and purity are monitored in real time through sensors and imaging equipment; a full-process optimization integration platform is provided.

Description

Technical Field

[0001] The present invention relates to the field of pharmaceutical preparation, and in particular to an optimization method for pharmaceutical preparation. Background Art

[0002] China Invention Patent Publication No. CN115719103A proposes a pharmaceutical process scheduling optimization method and device for solid preparation workshops. The method obtains production parameter information based on historical pharmaceutical information, and establishes a pharmaceutical scheduling model on this basis to optimize the maximum completion time and pharmaceutical cost in the production process. The method also establishes a scheduling encoding and decoding model between pharmaceutical batch process information and pharmaceutical equipment information, and solves these models to obtain the best scheduling plan, and finally realizes the association between pharmaceutical process information and pharmaceutical equipment information.

[0003] However, although this method has solved the scheduling problem in the pharmaceutical process to a certain extent and optimized production efficiency and cost, it still has some significant deficiencies and drawbacks in the overall optimization of drug preparation. First, this method mainly focuses on the scheduling optimization of the pharmaceutical process, but lacks multi-level and multi-factor optimization of the specific drug preparation process. The drug preparation process involves multiple complex chemical reactions, physical changes, and biological treatment steps, each of which is affected by multiple factors, such as temperature, pressure, pH value, reaction time, stirring speed, etc. Existing methods only rely on historical data and production parameter information to optimize scheduling, and do not fully consider these dynamically changing reaction conditions and how to adjust these conditions in real time during the actual production process to achieve the best results.

[0004] Secondly, the methods of the prior art rely on the quality and quantity of historical data during the optimization process, which may lead to a significant reduction in the optimization effect when the data is insufficient or the data quality is poor. Since the data in the pharmaceutical preparation process is multi-dimensional and complex, relying solely on historical data to establish a scheduling model may not be able to fully and accurately reflect all potential variables and interactions in the drug preparation process, which will lead to insufficient effectiveness and accuracy of the scheduling model. In addition, the optimization methods of the prior art mainly focus on minimizing the maximum completion time and production costs in the optimization goals of the scheduling process, but lack optimization considerations for steps such as purification, concentration and mixing of pharmaceutical ingredients. The key to pharmaceutical preparation is how to improve the quality and stability of the drug while optimizing the balance between the various processes. A single focus on optimizing production speed and cost may ignore the quality requirements of the preparation itself. Summary of the invention

[0005] The purpose of the present invention is to provide an optimized method for preparing a medicament, thereby solving some of the drawbacks pointed out in the background technology.

[0006] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: Optimization method for preparation of medicaments:

[0007] Includes: S1, ingredient optimization and dynamic deployment:

[0008] S1.1. Use a deep neural network machine learning model to screen and optimize the active ingredients and excipients of the drug, and establish a relationship model between ingredients and efficacy and stability;

[0009] S1.2, using a dynamic formulation module to adjust the proportion and dosage of ingredients in real time according to different preparation conditions including temperature and pH value;

[0010] S2. Adaptive optimization of reaction conditions:

[0011] S2.1. Based on S1, an adaptive reaction condition adjustment system is constructed to monitor the reaction conditions including temperature, pressure, pH value and stirring speed in the preparation process in real time;

[0012] S2.2, real-time adjustment of reaction conditions through adaptive neuro-fuzzy inference system;

[0013] S3, Dynamic mixing and reaction monitoring:

[0014] S3.1. Introduce a dynamic mixing system and use microfluidic technology to control the metering and mixing of pharmaceutical ingredients;

[0015] S3.2, and the integrated online monitoring module detects the flow rate, concentration and temperature key parameters of the mixed liquid in real time, and adjusts the mixing conditions according to the monitoring data;

[0016] S4, step-by-step optimization synthesis strategy:

[0017] S4.1. Based on the optimization results of S1 to S3, the drug preparation process is decomposed into multiple key steps including primary reaction, purification, and concentration, and each step is optimized separately;

[0018] S4.2. Use genetic algorithm to comprehensively consider the coupling relationship between each step and optimize the time, temperature and reaction dose parameters of each step;

[0019] S5. Real-time quality monitoring and feedback adjustment:

[0020] S5.1. Build a real-time quality monitoring system to monitor the physical and chemical properties of the drug, including particle size distribution, solubility, and purity, in real time through sensors and imaging equipment;

[0021] S5.2, combine machine learning models to predict the quality of the drug. If the quality deviates from expectations, immediately send feedback signals to S2 and S3 to adjust the reaction conditions and mixing strategies to correct the deviation;

[0022] S6, full process optimization and integration platform:

[0023] S6.1. Introduce an integrated platform to uniformly manage and optimize the data of each module in S1 to S5, including component screening, reaction condition adjustment, dynamic mixing, and quality monitoring.

[0024] Furthermore, the establishment of a relationship model between ingredients and efficacy and stability includes:

[0025] S1. First, design a multi-layer deep neural network structure, where the input layer receives the feature vector of the drug composition:

[0026] X=[x 1 ,x 2 ,…,x n ]

[0027] Each component feature vector is represented by several physical and chemical properties, including molecular structure, polarity and solubility; through multiple hidden layers H i =f(W i ·H i-1 +b i ), where f is a custom nonlinear activation function, W i and b i are the weight matrix and bias vector of the i-th layer respectively;

[0028] The output layer provides efficacy ε and stability The prediction index of the network is minimized by the loss function:

[0029]

[0030] Conduct training, including is a custom difference function used to measure the gap between the actual value and the predicted value, γ j (X) is the adaptive weight coefficient, which is used to adjust the contribution of different component combinations;

[0031] S2. The next component screening and combination optimization method includes using the trained deep neural network DNN model to perform a combination analysis on potential active ingredients and excipients and construct an objective function:

[0032]

[0033] Where C = [c 1 ,c 2 ,…,c k ] represents the combination of ingredients, p and q are optimization parameters, Ψ is the efficacy and stability scoring function of the combination of ingredients, and the solution is To obtain the component combination; then through the evolutionary algorithm:

[0034]

[0035] in represents the component combination in the tth generation, is the fitness increment of the ingredient combination, κ is the learning rate parameter, and the selection of ingredient combinations is adaptively adjusted to find formulations targeting different drug targets;

[0036] S3. Finally, the deep neural network DNN model outputs the weight matrix M = [m ij ], where m ij is the weight of the influence of the i-th ingredient on the j-th efficacy or stability index, through the formula:

[0037]

[0038] where α ij Based on the correlation coefficient between ingredients and efficacy, the key ingredients and their interactions that have the greatest impact on efficacy and stability are identified, and optimization suggestions based on scientific analysis are provided using the multidimensional relationship model λ(M).

[0039] Furthermore, the process of constructing the adaptive reaction condition adjustment system includes:

[0040] S1. First, the adaptive weighted average method and Kalman filter technology are used to perform data fusion and set the comprehensive reaction condition function Θ(t) to describe the comprehensive state of each reaction condition in the entire preparation process:

[0041]

[0042] Among them, ω i , ν j and χ k They are the adaptive weight functions of temperature, pressure and pH sensors, representing the importance of each sensor data at different times, T i (t), P j (t) and pH k (t) represents the temperature, pressure and pH value data at different times;

[0043] S2. Then, by learning historical data and real-time sensor data, the nonlinear effects of various reaction conditions on the quality and stability of the drug are controlled; adaptive optimization is used to set the optimization function:

[0044]

[0045] Where X(t) is the state vector of the reaction conditions, including temperature, pressure, pH value and stirring speed, αi is the adjustment weight of the corresponding condition, controlling its influence on the overall optimization, β is the weighting coefficient, which is used in the second-order derivative term to capture the acceleration change of the reaction condition, and Γ(X) is the objective function of the drug quality;

[0046] S3. Finally, edge computing and distributed control architecture are used to deploy data processing and decision-making parts on the local side of the manufacturing equipment, and local computing resources are used to analyze sensor data in real time. The control response is optimized through the local control response function R(t), which describes the system's rapid response and adjustment to real-time data:

[0047]

[0048] Among them, ζ and ξ are adjustment coefficients, which are used to balance the contribution of data fusion and regulation optimization. Indicates the rate of change of the comprehensive reaction condition data, is the rate of change of the optimization function.

[0049] Furthermore, the dynamic mixing system comprises the following steps:

[0050] S1. First, microfluidics technology is used to measure and mix the components with micron-level precision. The real-time dynamic changes of the component flow rate and mixing ratio are described by the control function Φ(t), which is defined as:

[0051]

[0052] Among them, Φ(t) is the measurement error function of the overall reaction process, α i is the sensitivity coefficient adjusted by the component flow rate, indicating the contribution weight of different components to the total error; Q i (t) represents the flow rate of the i-th component, λ i is the target flow rate of the corresponding component, indicating the ideal component flow rate during the preparation process;

[0053] S2. Then, the component flow rate and mixing ratio are adjusted in combination with the real-time monitoring data, and a dynamic response function Ψ(t) is set to characterize the adjustment of the component flow rate and mixing ratio at different times. The expression is:

[0054]

[0055] Among them, Ψ(t) is the feedback response function of the real-time adjustment process, β j is the control coefficient for adjusting the response speed, which is used to weigh the sensitivity of different components to the adjustment speed. j (t) is the real-time flow rate of the jth component, indicating the actual flow rate of the component during the preparation process, μ j(t) is the expected flow rate of the jth component, which represents the ideal flow rate required for the component according to the current conditions;

[0056] S3. Finally, the integration of microfluidic system and microreactor is used to coordinate the mixing and reaction process of components, and an integrated optimization function Ω(x) is constructed to describe the reaction efficiency under the conditions of entering the reaction stage after mixing. The formula is:

[0057]

[0058] Among them, Ω(x) is the reaction efficiency optimization function, κ is the reaction rate constant, which represents the reaction rate under specific conditions; γ is the component degradation coefficient, which describes the predicted degradation or loss of the component during the transmission process, e -γx is the component concentration attenuation factor, v(x) is the fluid velocity field, which indicates the flow velocity of the reactants inside the microreactor, is the divergence of the velocity field, reflecting the diffusion characteristics of the fluid; is the space area inside the reactor.

[0059] Furthermore, in the primary reaction step, a deep neural network is used to predict the yield and quality of the reaction product, a relationship model between the reaction conditions and the product quality is constructed, and a reaction optimization function is defined:

[0060]

[0061] in is the primary reaction optimization function, which represents the comprehensive reaction efficiency within a given time t; T is the reaction temperature, P is the reaction pressure, S is the stirring speed; Y is the yield of the reaction product; α, β, γ are the weight coefficients of temperature, pressure and stirring speed in the optimization process, which are used to adjust the influence of each variable on the product yield.

[0062] Furthermore, in the purification step, the purification process is optimized by a multi-stage separation model including multi-stage extraction, fractionation or chromatographic separation, and a multi-objective optimization function is constructed:

[0063]

[0064] in is the purification process optimization function, which is used to evaluate the overall efficiency of the purification process; ρ is the solvent concentration, F is the flow rate, E is the energy consumption, and τ is the purification time; C is the purity of the purified product, and μ, v, and λ are the weight coefficients for optimizing the solvent concentration, flow rate, and energy consumption, respectively, indicating the importance of each optimization target in the overall optimization process;

[0065] Furthermore, in the optimization process of the concentration step, a concentration kinetic model is established, the operating parameters of the evaporation or concentration equipment are optimized, and an adaptive control strategy is introduced to optimize the concentration efficiency function:

[0066]

[0067] in is the optimization function for the concentration process, used to evaluate the efficiency of the concentration stage; T v is the evaporation temperature, P v is the vacuum degree, κ is the concentration time, δ is the solvent recovery rate; M is the solid content of the concentrate, indicating the final product concentration of the concentration process; ξ, ζ, η are the optimization weight coefficients of each parameter, reflecting the important proportion of evaporation temperature, vacuum degree and solvent recovery rate in the optimization process.

[0068] The pharmaceutical preparation optimization method of the present invention has the following beneficial effects:

[0069] By adopting advanced technologies such as deep neural networks, dynamic mixing systems, and adaptive reaction condition adjustment, the present invention can accurately control various key parameters in the drug preparation process, such as ingredient ratios, temperature, pressure, pH value, etc.; reduce costs: in the concentration step, by establishing a concentration kinetic model and optimizing parameters such as evaporation and vacuum degree, the present invention maximizes the solvent recovery rate and the concentration efficiency of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 The present invention is a flowchart of the optimized method for preparing the pharmaceutical agent.

[0071] Figure 2 The present invention provides a flow chart of the relationship model between ingredients, efficacy and stability.

[0072] Figure 3 A process flow chart of an adaptive reaction condition adjustment system is constructed for the present invention. DETAILED DESCRIPTION

[0073] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0074] Combined with Figure 1Flowchart, the optimization method for preparing the medicament of the present invention comprises the following steps: S1, ingredient optimization and dynamic deployment: S1.1, using a machine learning model of a deep neural network to screen and optimize the active ingredients and excipients of the medicament, and establish a relationship model between ingredients and efficacy and stability; specifically, by collecting a large amount of experimental data of the medicament, including the types and proportions of active ingredients and excipients, and corresponding parameters such as efficacy (such as effectiveness, absorbability, bioavailability, etc.) and stability (such as physical stability, chemical stability, etc.), these data are input into the deep neural network model for training. The design of the deep neural network (DNN) model adopts a multi-layer structure, the input layer receives the feature vectors of the active ingredients and excipients, these features include molecular structure, physical and chemical properties, etc., and the output layer is used to predict efficacy and stability indicators. Through training, the model learns the effects of different ingredient combinations on the performance of the medicament, and generates a multi-dimensional ingredient-efficacy-stability relationship model. This model can not only accurately predict the effects of existing ingredient combinations, but also be used to explore new ingredient combinations and screen out the optimal ingredient formula to achieve the desired efficacy and stability goals. S1.2, using a dynamic deployment module to adjust the proportion and dosage of ingredients in real time according to different drug preparation conditions including temperature and pH value; the dynamic deployment module monitors key reaction conditions (such as temperature and pH value) in the drug preparation process by integrating a real-time monitoring system and a control system, and adjusts the proportion and dosage of ingredients according to the monitoring results. Based on the output of the machine learning model and real-time data feedback, such as temperature and pH value changes detected by sensors, this module adjusts the addition rate and proportion of ingredients through an adaptive control algorithm (such as model predictive control MPC) to optimize the preparation conditions and reaction environment of the drug, ensuring that the drug is always in the optimal state under changing conditions and achieving stable efficacy and quality goals.

[0075] S2. Adaptive optimization of reaction conditions: S2.1. Based on S1, an adaptive reaction condition adjustment system is constructed to monitor the reaction conditions including temperature, pressure, pH value, and stirring speed in the preparation process in real time. The system adopts multi-sensor integration technology and installs high-precision sensors at key positions of reaction vessels or preparation equipment to obtain data on reaction conditions in real time. For example, temperature sensors monitor the thermal changes in the reaction environment, pressure sensors detect the changes in air pressure generated during the reaction, pH sensors monitor the changes in the acidity and alkalinity of the solution in real time, and stirring speed sensors measure the stirring state and speed of the reactants. The data collected by all sensors are transmitted to the central control system through the data integration module. The central control system adopts an adaptive algorithm to fuse and analyze multi-source data, thereby comprehensively monitoring the real-time reaction conditions in the preparation process of the drug.

[0076] S2.2, adjust the reaction conditions in real time through the adaptive neural fuzzy inference system; the system uses the adaptive neural fuzzy inference system (ANFIS), combined with the self-learning ability of neural networks and the reasoning ability of fuzzy logic, to intelligently control and adjust the reaction conditions. The ANFIS system first learns the nonlinear relationship between reaction conditions and drug quality from historical data and current real-time data, establishes a rule base and model, and the model input includes a variety of real-time monitoring data such as temperature, pressure, pH value and stirring speed. These inputs are converted into fuzzy sets through the fuzzy logic system, and reasoning is performed through the fuzzy rule set to output the optimal reaction condition adjustment strategy. On this basis, the ANFIS system combines the adaptive mechanism to continuously optimize and adjust the reaction conditions. By adjusting the key parameters such as the heating and cooling rate of the equipment, the opening of the pressure regulating valve, and the speed of the agitator in real time, it ensures that the drug preparation process is always under the optimal reaction conditions, thereby maximizing the reaction efficiency, product quality and stability of the drug.

[0077] S3. Dynamic mixing and reaction monitoring: S3.1. Introduce a dynamic mixing system and use microfluidics to control the metering and mixing of pharmaceutical ingredients. The system uses microfluidics to control the flow rate and volume of each pharmaceutical ingredient through micron-level fluid channels and precision micro pumps, thereby achieving high-precision ingredient metering and mixing. The application of microfluidics can ensure that each ingredient is delivered to the mixing area at a precise ratio and flow rate. A variety of microvalves and diverters are set up inside the microfluidic chip. By programming and controlling the switching and flow distribution of these microstructures, dynamic mixing of multiple ingredients is achieved, ensuring the uniformity and reactivity of the medicine at the microscale, and improving mixing efficiency and stability.

[0078] S3.2, the integrated online monitoring module detects the flow rate, concentration and temperature key parameters of the mixed liquid in real time, and adjusts the mixing conditions according to the monitoring data; the online monitoring module monitors the key parameters of the mixed liquid in real time through an integrated sensor network, including a flow rate sensor to measure the flow rate of the agent, a concentration sensor to monitor the concentration distribution of each component in the mixed liquid, and a temperature sensor to sense the temperature change of the mixed liquid, and these data are transmitted to the central control system in real time. The central control system analyzes the current mixing state based on the monitoring data, and adjusts the mixing process in real time by adjusting the micropump flow rate, valve opening and mixing path parameters in the microfluidic system to ensure that the flow rate, concentration and temperature of the mixed liquid are in the optimal state, thereby achieving the best reaction effect and product quality. In this way, the dynamic mixing system can achieve efficient and accurate component metering and mixing in the preparation process of the agent, and at the same time combine real-time monitoring and adjustment to ensure the optimization of the uniformity, stability and reaction efficiency of the agent.

[0079] S4. Step-by-step optimization of synthesis strategy: S4.1. Based on the optimization results of S1 to S3, the drug preparation process is decomposed into multiple key steps including primary reaction, purification, and concentration, and each step is optimized separately; specifically, according to the results of component optimization, reaction condition optimization, mixing and monitoring obtained in the previous steps, the preparation process is decomposed and analyzed, and the entire drug preparation process is divided into multiple key steps such as primary reaction, purification, and concentration; in the primary reaction stage, by using reaction kinetic models and machine learning algorithms (such as deep neural networks or random forest regression), the reaction conditions (such as temperature, pressure, reaction time, stirring speed) are predicted and optimized to maximize the yield and purity of the target product while minimizing the formation of by-products; in the purification stage, the separation efficiency and purity are optimized through multi-stage separation technology (such as multi-stage extraction, fractionation or chromatographic separation) and real-time monitoring; in the concentration stage, by establishing a concentration kinetic model and an adaptive control strategy, the concentration parameters (such as temperature, pressure, vacuum degree, etc.) are dynamically adjusted to ensure that the concentration efficiency and solvent recovery rate are maximized, thereby improving the product quality and process efficiency of each step.

[0080] S4.2. Genetic algorithms are used to comprehensively consider the coupling relationship between each step and optimize the time, temperature, and reaction dosage parameters of each step. The operating parameters of each step are globally optimized through genetic algorithms. Genetic algorithms simulate the biological evolution process and use operations such as selection, crossover, and mutation to find the global optimal solution, which can effectively deal with the coupling and mutual influence between each step. The goal of genetic algorithms is to minimize the total production time, energy consumption, and raw material usage, while maximizing the product yield and purity. The fitness function is used to evaluate the advantages and disadvantages of each parameter combination. For example, when optimizing the primary reaction, the genetic algorithm evaluates the effects of different reaction time, temperature, and reaction dosage combinations on product yield and purity. In the purification stage, the effects of different solvents, temperatures, and flow rate conditions on purification efficiency are evaluated. In the concentration stage, the effects of different evaporation temperatures, vacuum degrees, and time on concentration effects are evaluated. Through multiple iterations and evolutions, the genetic algorithm gradually finds the optimal combination of parameters for each step to ensure that the entire preparation process is carried out under global optimal conditions, thereby improving the overall quality, stability, and preparation efficiency of the drug.

[0081] S5. Real-time quality monitoring and feedback adjustment: S5.1. Build a real-time quality monitoring system to monitor the physical and chemical properties of the drug in real time through sensors and imaging equipment; specifically, the system is equipped with a series of high-precision sensors and imaging equipment to monitor the key quality indicators of the drug during the preparation process, including the particle size distribution, solubility, purity and other physical and chemical properties of the drug particles. Particle size distribution monitoring uses technologies such as laser scattering, dynamic light scattering or microscopic imaging to obtain real-time information on the size and distribution of particles to ensure that the particle size is uniform, which is beneficial to the solubility and bioavailability of the drug; solubility testing uses UV-Vis spectrometers or near-infrared spectrometers to measure the dissolution of the drug in the solution to ensure that it meets the predetermined solubility standards; purity testing uses high-performance liquid chromatography (HPLC), gas chromatography (GC) and other equipment to monitor the content of active ingredients and impurities in the drug to ensure that the purity of the final product meets the specified quality standards. S5.2. Combine machine learning models to predict the quality of pharmaceuticals. If the quality deviates from expectations, feedback signals are immediately sent to S2 and S3 to adjust the reaction conditions and mixing strategies to correct the deviations. Specifically, the constructed machine learning model (such as a deep neural network or a support vector machine) is trained with a large amount of historical data to learn the complex relationship between the physical and chemical properties of the pharmaceutical and its quality. The model continuously receives data input from the real-time monitoring system, such as particle size distribution, solubility, purity, etc., and predicts the overall quality of the pharmaceutical based on the current monitoring results. If the machine learning model detects that the pharmaceutical quality indicators deviate from expectations, such as uneven particle size, solubility below the standard, or purity that does not meet the requirements, the system will immediately send a feedback signal to the reaction condition adjustment system S2 and the mixing strategy control module S3, and adjust the reaction conditions (such as temperature, pressure, pH value, stirring speed) and mixing ratio, flow rate and other parameters accordingly according to the actual situation to correct the current deviations and ensure that the pharmaceutical preparation process is always kept in the best state, thereby ensuring that the quality and stability of the final pharmaceutical meet the expected requirements.

[0082] S6. Full-process optimization and integration platform: Introduce an integrated platform to uniformly manage and optimize the data of each module in S1 to S5, including ingredient screening, reaction condition adjustment, dynamic mixing, and quality monitoring. Specifically, the integrated platform uses advanced data integration and analysis technologies to centrally manage all key data sources in the pharmaceutical preparation process by establishing a central data management system. The data sources include the output data of the ingredient screening module, such as the types and proportions of active ingredients and excipients and their optimized combination schemes; real-time data of the reaction condition adjustment module, such as real-time monitoring results and adjustment records of reaction parameters such as temperature, pressure, pH value, and stirring speed; mixing condition data of the dynamic mixing module, including flow rate, ingredient concentration, mixing uniformity, etc.; data of the quality monitoring module, including physical and chemical property monitoring data such as particle size distribution, solubility, and purity; the integrated platform aggregates these data from different modules in real time through standardized data interfaces, and performs data cleaning, preprocessing, storage, and analysis to ensure data Integrity, consistency and traceability; the platform integrates advanced data analysis tools and machine learning algorithms, which can conduct in-depth analysis and correlation mining of multi-source data, identify the interactions and optimization space between modules, and establish a multi-factor optimization model for the drug preparation process; the platform's optimization engine continuously learns and updates data, optimizes ingredient screening strategies, reaction condition adjustment schemes, mixing control strategies and quality monitoring indicators in real time, and ensures that the entire preparation process is carried out under global optimal conditions, thereby improving the production efficiency, product quality and consistency of the drugs. The integrated platform also has a digital decision support function, which can provide operators with real-time optimization suggestions based on the analysis results or directly adjust the parameter settings of related modules, minimize human intervention and operational errors, and significantly improve the intelligence level and production efficiency of drug preparation.

[0083] Embodiment 1:

[0084] See also Figure 2 In this example, the research and development department of "Pharmaceutical Company A" develops a new anti-inflammatory drug. In order to ensure that the drug has the best efficacy and stability in clinical use, a deep neural network (DNN) model is used to establish a relationship model between the drug ingredients and the efficacy and stability, thereby optimizing the ingredient formula.

[0085] First, a multi-layer deep neural network structure was designed, where the input layer receives the feature vector of the drug composition:

[0086] X=[x 1 ,x 2 ,…,x n ]

[0087] Each component feature vector is represented by a variety of physical and chemical properties, including molecular structure, polarity, solubility, etc. Assume that a combination of three main active ingredients is being studied, and the feature vectors of each component are x 1 、x 2 and x 3 Their characteristic values ​​include molecular structure parameters (such as molecular weight of 300-500 g / mol), polarity parameters (dielectric constant range of 10-30), and solubility (solubility in water of 0.1-10 mg / mL).

[0088] Next, the hidden layer of the network is calculated by the formula:

[0089] H i =f(W i ·H i-1 +b i )

[0090] Calculate, where f is a custom nonlinear activation function, ReLU function (RectifiedLinearUnit) is selected, W i and b i are the weight matrix and bias vector of the i-th layer respectively. Assume that the model has three hidden layers, and the weight matrix W of each layer 1 ,W 2 ,W 3 The sizes are 3x10, 10x10, 10x2, and the bias vector b 1 ,b 2 ,b 3 The sizes are 10, 10, and 2 respectively. Through the processing of these hidden layers, the network is able to capture the complex nonlinear relationship of drug components.

[0091] At the output layer, the model provides efficacy ε and stability The efficacy of the drug is measured by the inhibition rate of the drug on inflammatory markers in laboratory experiments, ranging from 0-100%, while the stability The degradation rate of the drug under different temperature and humidity conditions is expressed in the range of 0-1%. The network minimizes the loss function:

[0092]

[0093] Conduct training, including is the difference function, which is used to measure the gap between the efficacy and stability predicted by the model and the actual experimental data. The mean square error (MSE) is selected as the difference function; γ j (X) is the adaptive weight coefficient, which is used to adjust the contribution of different component combinations.

[0094] The model is trained using large-scale experimental data. For example, for 200 different combinations of ingredients, the efficacy and stability data of each combination under different experimental conditions are collected. By inputting these data into a deep neural network, the network learns and optimizes the ingredient combination. The model finally learned by the network is set to predict an efficacy of 80% and a stability of 0.95 under a given input feature X, and after multiple iterations of training, the model's loss function is It is close to 0.02, which indicates that the model can predict drug efficacy and stability well.

[0095] In the process of continuing to develop new anti-inflammatory drugs, the composition of the drug is further optimized to ensure that the drug can maintain high efficiency and stability under different conditions. A previously trained deep neural network (DNN) model was used to analyze and optimize potential active ingredients and excipient combinations. To achieve this goal, an objective function was constructed It is used to evaluate the efficacy and stability of different ingredient combinations. The objective function is defined as:

[0096]

[0097] Where C = [c 1 ,c 2 ,…,c k ] indicates the combination of ingredients, c 1 ,c 2 ,…,c k are the concentrations of different active ingredients and excipients, respectively. p and q are optimization parameters that control the efficacy ε and stability. The sensitivity of Ψ is the efficacy and stability scoring function of the component combination. Usually, Ψ will comprehensively consider the efficacy and stability indicators and set Among them, α and β are scoring weights, reflecting the evaluation of the relative importance of efficacy and stability, and Ω represents the integral area under different experimental conditions. In order to ensure that the combination of ingredients achieves the best effect, it is necessary to solve That is, find the combination of all ingredients that The biggest combination.

[0098] Next, the combination of ingredients is further optimized through evolutionary algorithms. The core of evolutionary algorithms is to gradually optimize the combination of ingredients through iterations. Specifically, in the tth generation, the update formula of the component combination is:

[0099]

[0100] in, represents the component combination in the tth generation, is the fitness increment of the component combination, and κ is the learning rate parameter, which is used to control the step size of each iteration. The learning rate κ is set in the range of 0.01 to 0.1 to ensure that the optimization process can converge quickly while maintaining sufficient accuracy.

[0101] Set the initially selected component combination to The corresponding component concentrations are 20%, 30% and 50% respectively. Through calculation, the initial value of the objective function is obtained This means that the efficacy and stability are ideal at this time. After multiple iterations of the evolutionary algorithm, the composition is adjusted to At this time, the maximum value of the objective function is This shows that this combination is superior in terms of efficacy and stability.

[0102] In further tests, the performance of the optimized combination of ingredients under different experimental conditions was verified. The results showed that the drug maintained high anti-inflammatory effects and good stability in different temperature and pH environments, and side effects were significantly reduced.

[0103] In the next stage of the development of new anti-inflammatory drugs, the specific impact of the combination of ingredients is further analyzed and optimized to more clearly understand the contribution of each ingredient to efficacy and stability. In the previous step, the deep neural network (DNN) model has been used to optimize the combination of ingredients and find a combination with superior efficacy and stability. Next, the weight matrix M = [m ij ], where m ij It represents the influence weight of the i-th ingredient on the j-th efficacy or stability index.

[0104] For further optimization, a multidimensional relationship model λ(M) was defined, which is used to identify the key components and their interactions that have the greatest impact on drug efficacy and stability. Specifically, the expression of λ(M) is:

[0105]

[0106] Among them, α ij is the correlation coefficient between component and efficacy, reflecting the importance of each component to a specific efficacy or stability index; m ij The weights of the components output by the DNN model. In order to identify the most critical combination of components in practical applications, the weight matrix M is specifically analyzed. It is assumed that three main components and two key efficacy and stability indicators are studied, so the weight matrix is:

[0107]

[0108] In this matrix, m11 =0.4 represents the influence weight of component 1 on efficacy index 1, m 12 =0.6 represents the influence weight of component 1 on the stability index, and so on. Next, adjust the weight of each element and set α ij The value range is 0.5 to 1.5. ij for:

[0109] α 11 =1.0,α 12 =1.2,α 21 =0.8,α 22 =1.5,α 31 =1.1,α 32 =0.9

[0110] Substituting these values ​​into the expression of the relational model λ(M), we get:

[0111] λ(M)=1.0·ln(1+0.4 2 )+1.2·ln(1+0.6 2 )+0.8·ln(1+0.2 2 )+1.5·ln(1+0.8 2 )+1.1

[0112] ·ln(1+0.5 2 )+0.9·ln(1+0.3 2 )

[0113] After calculation, we get:

[0114] λ(M)≈1.0·ln(1.16)+1.2·ln(1.36)+0.8·ln(1.04)+1.5·ln(1.64)+1.1·ln(1.25)+0.9

[0115] ln(1.09)

[0116] After further calculation, it is found that λ(M)≈0.148+0.309+0.031+0.48+0.223+0.085=1.276.

[0117] By calculating the value of λ(M), we can identify component 3 (because α 31 =1.1 and m 31 =0.5) has a greater impact on the efficacy, component 2 (because α 22 =1.5 and m 22=0.8) has a significant impact on stability. Based on this analysis, the proportion of ingredient 3 was further increased in subsequent optimization, while the concentration of ingredient 2 was adjusted to enhance the anti-inflammatory effect and stability of the drug. After the optimization analysis of this multidimensional relationship model, Pharmaceutical Company A found that the new combination of ingredients made the drug show stronger efficacy and higher stability in preclinical experiments, which was significantly better than the original combination, further verifying the effectiveness of the optimization suggestions based on scientific analysis.

[0118] Embodiment 2:

[0119] Combination Figure 3 In this embodiment, in the process of continuing to optimize the preparation of new anti-inflammatory drugs, an adaptive reaction condition adjustment system is constructed to ensure that various reaction conditions can be maintained in the optimal state during the entire drug preparation process to maximize the efficacy and stability of the drug. First, the adaptive weighted average method and Kalman filtering technology are used for data fusion to comprehensively analyze key reaction conditions such as temperature, pressure and pH value.

[0120] The comprehensive reaction condition function Θ(t) is defined to describe the comprehensive state of each reaction condition in the entire preparation process, and its mathematical expression is:

[0121]

[0122] Among them, ω i , ν j and χ k They are the adaptive weight functions of temperature, pressure and pH sensors, representing the importance of each sensor data at different times, T i (t), P j (t) and pH k (t) represents the temperature, pressure and pH value data at different times.

[0123] Assume that the pharmaceutical team is monitoring the data of three temperature sensors, two pressure sensors and one pH sensor. At a certain moment, the data measured by the temperature sensor is T 1 (t) = 37°C, T 2 (t) = 38°C, T 3 (t) = 36.5°C; the data measured by the pressure sensor is P 1 (t) = 1.2 atm, P 2 (t) = 1.1 atm; the data measured by the pH sensor is pH 1 (t) = 7.4. In order to determine the importance of each data point, an adaptive weight function is defined, such as ω i (T i (t)) = 0.3 + 0.01 × (T i (t)-36.5), νj (P j (t)) = 0.5 + 0.02 × (P j (t)-1.0), χ k (pH k (t))=0.4+0.05×|pH k (t)-7.0|, where ω i , ν j , χ k The value range of is set to 0.1 to 1.0 to reflect the changing importance of different conditions to the reaction.

[0124] After substituting specific data, calculate each weight value: ω 1 =0.3+0.01×(37-36.5)=0.35,ω 2 =0.3+0.01×(38-36.5)=0.45,ω 3 =0.3+0.01×(36.5-36.5)=0.3;ν 1 =0.5+0.02×(1.2-1.0)=0.54,ν 2 =0.5+0.02×(1.1-1.0)=0.52; χ 1 =0.4+0.05×|7.4-7.0|=0.6.

[0125] According to these weight values, substitute them into the comprehensive reaction condition function Θ(t) for calculation:

[0126]

[0127] For observations set within the time interval [0,10] minutes, we calculate:

[0128]

[0129] Θ(t)=46.66×10=466.6

[0130] The result of this comprehensive reaction condition function shows that within 10 minutes, the current reaction condition (temperature, pressure, pH value) comprehensive state is 466.6. Kalman filtering technology is used to smooth these data to remove noise and ensure the reliability and accuracy of the monitoring data. Then, the result of this function is used to adjust the reaction conditions in real time, such as by adjusting the temperature control device, pressure control valve and pH regulator to keep the reaction conditions within the optimal range.

[0131] Next, in order to further optimize the production process of the new anti-inflammatory drug, the nonlinear effects of various reaction conditions on the quality and stability of the drug were controlled by learning historical data and real-time sensor data. Adaptive optimization technology was used to set the optimization function It is used to describe the effect of reaction conditions on the quality of the drug. Specifically, the expression of the optimization function is:

[0132]

[0133] Where X(t) is the state vector of the reaction conditions, including temperature, pressure, pH value and stirring speed; α i is the adjustment weight of the corresponding condition, which controls the influence of each condition on the overall optimization. Its value range is 0.1 to 2.0 to adapt to the importance of different conditions to the reaction results; β is the weighting coefficient, which ranges from 0.01 to 0.1 and is used for the second-order derivative term to capture the acceleration change of the reaction conditions; Γ(X) is the objective function of drug quality, which evaluates the quality indicators of the drug under the current reaction conditions, such as efficacy and stability.

[0134] The goal is to optimize temperature, pressure, and pH to maximize the efficacy of the drug. First, through historical data analysis and laboratory measurements, the current condition state vector is determined as follows: temperature T(t) = 37°C, pressure P(t) = 1.2atm, pH value pH(t) = 7.4, stirring speed S(t) = 1000rpm. Using these conditions, the first-order derivative is calculated and the second derivative For example, at the first time point (1 minute), the temperature change rate is The rate of change of pressure is The pH change rate is The rate of change of stirring speed is

[0135] Substitute these values ​​into the optimization function and set α i =[1.5,1.2,1.0,0.8] and β = 0.05, and set Γ(X) = 100-5(T-37) 2 -10(P-1.2) 2 -15(pH-7.4) 2 -0.1(S-1000) 2 , which is used to evaluate the impact of current reaction conditions on the quality of the drug, where Γ(X) ranges from 0 to 100, and the closer it is to 100, the closer the drug quality is to the target.

[0136] Compute the value of the optimized function:

[0137]

[0138]

[0139]

[0140] get Indicates the optimization result of the current reaction conditions. During the optimization process, the reaction conditions are continuously adjusted (such as by controlling the temperature regulator, pressure control valve, pH regulator and agitator speed) to maximize If in a new round of iterations, by fine-tuning the temperature to 37.5°C, the pressure to 1.25 atm, the pH to 7.35 and the stirring speed to 1050 rpm, the new optimization function value is improved to 22.5, indicating that the adjustment direction is correct.

[0141] Through the adaptive optimization process, Pharmaceutical Company A was able to find the combination of reaction conditions that would achieve the best drug quality and stability in the shortest possible time.

[0142] Finally, to further improve the efficiency of the pharmaceutical process and the precise control of reaction conditions, edge computing and distributed control architecture are used to deploy data processing and decision-making parts on the local side of the preparation equipment to achieve real-time data analysis and control response. By utilizing the computing resources of local devices, data from various sensors, including temperature, pressure, pH value, and stirring speed, can be quickly analyzed, thereby achieving real-time optimization of various reaction conditions in the pharmaceutical preparation process.

[0143] To this end, a local control response function is constructed It is used to describe the system's rapid response and adjustment to real-time data. The expression of this response function is:

[0144]

[0145] Among them, ζ and ξ are adjustment coefficients, which control the balance between data fusion and regulation optimization; the value range of ζ is set to 0.1 to 1.0 to adjust the contribution of comprehensive reaction condition data, and the value range of ξ is set to 0.01 to 0.1 to adjust the change rate of the optimization function; It indicates the rate of change of comprehensive reaction condition data, reflecting the changes of temperature, pressure, pH value and other data; The rate of change of the optimization function is used to evaluate the impact of adjustment of reaction conditions on the quality of the drug.

[0146] Assume that the change rate of the current comprehensive reaction condition function Θ(t) is calculated to be 0.5, and the optimization function The rate of change of is 0.3. In this step, we choose ζ = 0.5 and ξ = 0.05 and substitute these values ​​into the local control response function:

[0147]

[0148] Further calculation yields:

[0149]

[0150] The key is to set the comprehensive response time of the reaction process within the first 10 minutes.

[0151]

[0152] This response value of 1.325 indicates the system's response speed and adjustment effect to the current reaction condition changes. Based on this response value, the control system at the local end can make decisions quickly, such as adjusting the temperature regulator setting to fine-tune to a more precise temperature range, such as from 37.5°C to 37.8°C; or adjusting the pressure valve setting to adjust the pressure from 1.25atm to 1.23atm. The rapid response of these adjustments can optimize the reaction conditions in the preparation process in real time, thereby ensuring the best quality of the drug.

[0153] With the introduction of edge computing and distributed control architecture, data transmission delays can be significantly reduced, and the real-time performance and response speed of the system can be improved. At the same time, the application of this architecture reduces dependence on central servers and reduces the load and cost of data processing. Through this local optimization strategy, it was found that the time efficiency of drug production was improved by 15% and the production cost was reduced by 10%.

[0154] Embodiment 3:

[0155] In this embodiment, in the process of further optimizing the preparation of the novel anti-inflammatory drug, a dynamic mixing system is used to ensure that the flow rate and mixing ratio of each component can be accurately controlled throughout the preparation process of the drug. The core of this system is to use microfluidics technology to achieve micron-level precision in component metering and mixing, and to describe the real-time dynamic changes of component flow rate and mixing ratio through the control function Φ(t).

[0156] Specifically, a control function Φ(t) is defined to measure the metering error in the overall reaction process. The mathematical expression of this function is:

[0157]

[0158] Among them, Φ(t) is the measurement error function of the overall reaction process; α i Q is the sensitivity coefficient for adjusting the component flow rate, which is used to indicate the contribution weight of different components to the total error. Its value range is 0.1 to 1.0, ensuring that the sensitivity adjustment is adapted to the importance of different components. i (t) represents the flow rate of the i-th component, λ iis the target flow rate of the corresponding component, indicating the ideal component flow rate during the preparation process.

[0159] Assume that three components are being mixed, and the target flow rates of these components are set to λ 1 =0.8mmL / min,λ 2 =1.0mL / min,λ 3 =0.6mL / min. In order to control the mixing process in real time, the microfluidic device is used to take the component flow rate monitoring data as input. For example, the component flow rate at a certain moment is Q 1 (t) = 0.75 mL / min, Q 2 (t) = 1.05 mL / min, Q 3 (t) = 0.55 mL / min. Based on these data, the sensitivity coefficient α is defined 1 =0.9,α 2 =0.7,α 3 =0.5, which is used to reflect the influence of each component on the total error.

[0160] Substituting these values ​​into the control function Φ(t), we can calculate:

[0161]

[0162] In the actual monitoring process, it is set that at the first minute, the component flow rate change rate is Substituting these data into the above formula, we can calculate:

[0163]

[0164] This result shows that the metering error of the mixing process is 1.38095 in the first minute. By monitoring this error value, the valve opening and pressure settings in the microfluidic device are adjusted in real time to reduce the deviation of the component flow rate. For example, by adjusting the flow rate control valve of the first component, the flow rate is increased to 0.8mL / min; the flow rate of the second component is adjusted to 1.0mL / min, and the flow rate of the third component is fine-tuned to 0.6mL / min. Through these real-time adjustments, the metering error function Φ(t) gradually approaches zero, indicating that the component flow rate and mixing ratio gradually reach the optimal state.

[0165] In order to further optimize the dynamic mixing process of drug preparation, in addition to accurately controlling the flow rate of each component, it is also necessary to dynamically adjust the component flow rate and mixing ratio to ensure that the final quality and stability of the drug are optimal. To this end, a dynamic response function Ψ(t) was set in combination with real-time monitoring data to characterize the adjustment of component flow rate and mixing ratio at different times.

[0166] The expression of the dynamic response function Ψ(t) is:

[0167]

[0168] Among them, Ψ(t) is the feedback response function of the real-time adjustment process, β j R is the control coefficient for adjusting the response speed, which is used to weigh the sensitivity of different components to the adjustment speed. Its value range is 0.1 to 1.0 to reflect the importance of each component in the entire mixing process. j (t) is the real-time flow rate of the jth component, indicating the actual flow rate of the component during the preparation process; μ j (t) is the expected flow rate of the jth component, which represents the ideal flow rate required for the component according to the current conditions.

[0169] The flow rate adjustment of three main components is being monitored, where the expected flow rates of these components are set as μ 1 (t) = 0.8 mL / min, μ 2 (t) = 1.0 mL / min, μ 3 (t) = 0.6 mL / min. At a certain moment, the actual monitored component flow rate is R 1 (t) = 0.78 mL / min, R 2 (t) = 1.03 mL / min, R 3 (t) = 0.55 mL / min. Set the control coefficient β to adjust the response speed 1 =0.7, β 2 =0.5, β 3 =0.9, these coefficients reflect the importance and sensitivity of different components in adjusting speed.

[0170] Substitute these values ​​into the dynamic response function Ψ(t) for calculation:

[0171]

[0172] Set the change rate of the component flow rate within the first minute to be Substituting these data into the expression, we get:

[0173]

[0174] This result shows that in the first minute, the adjustment response value of the component flow rate and mixing ratio is 1.24892. Based on this response value, it can be judged whether the current adjustment speed and direction meet the requirements. If the value of Ψ(t) is too high, it means that the adjustment of the component flow rate is not accurate enough, and the response speed and sensitivity need to be further optimized. For example, β will be increased according to the actual situation. 1to 0.8 to respond more quickly to changes in the flow rate of component 1; or adjust the device so that the actual flow rate of component 2 is closer to its expected value of 1.0 mL / min.

[0175] Through these dynamic adjustments, the value of Ψ(t) is gradually reduced to approach zero, which means that the flow rate and ratio adjustment of the components in the mixing process have reached an ideal state. After a series of adjustments and optimizations, the uniformity and stability of the drug mixing have been significantly improved.

[0176] In order to further improve the efficiency and effect of the drug preparation process, the microfluidic system is integrated with the microreactor to better control the mixing and reaction process of the components. Through this integration, the coordinated control of the mixing of components and the reaction process can be achieved to ensure that the drug reaches the optimal state when entering the reaction stage. To this end, an integrated optimization function Ω(x) is constructed to describe the reaction efficiency of the mixed components in the reaction stage. The mathematical expression of this function is:

[0177]

[0178] Among them, Ω(x) is the reaction efficiency optimization function, which is used to evaluate the overall reaction efficiency after the mixture enters the reaction stage; κ is the reaction rate constant, which ranges from 0.01 to 0.1 and represents the reaction rate under specific conditions; γ is the component degradation coefficient, which ranges from 0.001 to 0.01 and describes the degradation or loss of the component during the transmission process; e -γx is the component concentration attenuation factor, which reflects the attenuation of the component when it is transmitted inside the reactor; v(x) is the fluid velocity field, which indicates the flow velocity of the reactant inside the microreactor; is the divergence of the velocity field, reflecting the diffusion characteristics of the fluid in the reactor; is the space area inside the reactor.

[0179] Set the reaction process of the microreactor being monitored, the internal space area of ​​the reactor The reaction chamber has a volume of 10 mL. The reaction rate constant κ = 0.05 and the component degradation coefficient γ = 0.005 are set. Under this condition, it is observed that the fluid velocity field v(x) of the reactants is uniform in the reactor and is set to 1 cm / s.

[0180] To calculate the optimization function Ω(x), we first calculate the divergence of the velocity field The value is set to 0.1s-1, indicating that the diffusion characteristics of the reactants inside the reactor are good. Next, substitute these values ​​for the integral calculation:

[0181]

[0182] because The reaction chamber is 10 mL, and the integration is performed over the entire volume of the reactor. Assuming that the fluid velocity field v(x) in the reactor is uniformly distributed, the integration can be simplified to:

[0183]

[0184] Further calculation:

[0185]

[0186] The shape of the reactor is assumed to be a cuboid with side lengths of L = 2 cm, W = 2 cm, and H = 2.5 cm, so the volume Due to e -0.005x The change in the microreactor is not significant and can be approximated as a constant:

[0187] Ω(x)≈0.005·e -0.005·5 ·10=0.005·e -0.025 10

[0188] Calculated:

[0189] Ω(x)≈0.005·0.9753·10=0.048765

[0190] This result shows that the current reaction efficiency optimization value is 0.048765. Based on this optimization value, it can be judged whether the current mixing and reaction conditions are close to the optimal state. If the value of Ω(x) is low, it means that the reaction efficiency is not high, which is due to uneven diffusion of the components inside the reactor or severe degradation of the components. To improve the reaction efficiency, the operating conditions of the reactor can be adjusted, such as increasing the fluid velocity or optimizing the internal structure of the reactor to promote mixing.

[0191] Embodiment 4:

[0192] In this example, in order to further optimize the production process of the new anti-inflammatory drug, a deep neural network (DNN) was used in the primary reaction step to predict the yield and quality of the reaction product, thereby constructing a relationship model between the reaction conditions (temperature, pressure and stirring speed) and the product quality. To achieve this goal, a reaction optimization function was defined It is used to measure the overall efficiency of the reaction under specific conditions. The expression of the optimization function is:

[0193]

[0194] in, is the primary reaction optimization function, which represents the comprehensive reaction efficiency within a given time t; T is the reaction temperature, P is the reaction pressure, S is the stirring speed; Y is the yield of the reaction product; α, β, γ are the weight coefficients of temperature, pressure and stirring speed in the optimization process, respectively, and their value range is set to 0.1 to 1.0, which is used to adjust the influence of each variable on the product yield.

[0195] Suppose a reaction process is being optimized, the goal of which is to maximize the yield of the reaction product within 20 minutes. The initial reaction conditions selected are temperature T = 80 ° C, pressure P = 2.0 atm, and stirring speed S = 300 rpm. Under these conditions, the DNN model is used to predict the change of product yield with temperature, pressure and stirring speed.

[0196] The output of the DNN model shows that the product yield Y under the current conditions is 75% at 10 minutes and 85% at 20 minutes. For further optimization, the reaction optimization function is used. The weight coefficients α=0.7, β=0.5, and γ=0.3 were set to reflect the high relative importance of temperature on product yield, while the influence of pressure and stirring speed was relatively small.

[0197] When calculating the optimization function, it is necessary to determine the derivative value of each variable. For example, through experiments and model predictions, the derivative of the temperature change at 10 minutes on the product yield is obtained: Derivative of pressure change on product yield Derivative of stirring speed change on product yield

[0198] Substitute these values ​​into the optimization function for calculation:

[0199]

[0200] Further calculation yields:

[0201]

[0202]

[0203]

[0204] Optimizing functions It indicates the overall efficiency of the reaction under the current reaction conditions. By analyzing this value, it is found that temperature has the greatest impact on the product yield, so in the next experiment, the temperature is increased to 85°C, while the pressure is slightly lowered to 1.8 atm and the stirring speed is increased to 350 rpm to further improve the product yield.

[0205] Under the new conditions, the DNN model made predictions again, and the results showed that the product yield increased to 90% at 20 minutes.

[0206] Embodiment 5:

[0207] In this example, in the process of optimizing the production of a new anti-inflammatory drug, the purification of the drug was performed to ensure the high purity and quality of the final product. To achieve this goal, a multi-stage separation model was introduced, including multi-stage extraction, fractionation, and chromatography separation methods to improve the overall efficiency of the purification process. In this process, a multi-objective optimization function was constructed. Used to evaluate the overall efficiency of the purification process and find the optimal balance between multiple objectives.

[0208] The mathematical expression of the optimization function is:

[0209]

[0210] in, is the purification process optimization function, which is used to evaluate the overall efficiency of the entire purification process; ρ is the solvent concentration, F is the flow rate, E is the energy consumption, and τ is the purification time; C is the purity of the purified product, indicating the purity of the final product under different purification conditions; μ, ν, λ are the weight coefficients for optimizing the solvent concentration, flow rate, and energy consumption, respectively, and their values ​​range from 0.1 to 1.0, indicating the importance of each optimization objective in the overall optimization process.

[0211] The optimization of the purification process is set to achieve a purity of more than 99% after purification, and to minimize energy consumption and shorten purification time while ensuring high purity. The initial conditions are set as follows: solvent concentration ρ = 0.8mmol / L, flow rate F = 5mL / min, energy consumption E = 50J / s, and target purification time τ = 60 minutes. Through experiments and preliminary tests, the purity C of the product under these conditions is 98%.

[0212] In order to further optimize the purification process, a multi-objective optimization function was used to balance the effects of solvent concentration, flow rate, and energy consumption on purification efficiency. The weight coefficients were set to μ = 0.5, ν = 0.3, and λ = 0.2. These coefficients reflect that solvent concentration has the greatest impact on product purity, followed by flow rate, and energy consumption is relatively minimal.

[0213] Next, by monitoring the experimental data, the derivatives of each variable on the product purity are calculated. For example, the effect of solvent concentration on purity is The effect of flow rate on purity is The influence of energy consumption on purity is

[0214] Substitute this data into the optimization function:

[0215]

[0216] Further calculation yields:

[0217]

[0218]

[0219]

[0220] This optimization function value Indicates the overall efficiency of the purification process under the current conditions. For further optimization, the solvent concentration was increased to 0.85 mol / L, the flow rate was increased to 6 mL / min, and the energy consumption was tried to be controlled within 45 J / s. Under the new experimental conditions, after 30 minutes, the product purity reached 99.2%, and the energy consumption was reduced.

[0221] Through repeated adjustments and optimization, the optimal purification conditions were gradually found, which significantly reduced energy consumption and shortened purification time while ensuring high purity of the drug. Ultimately, the application of this multi-stage separation optimization model made the production process of the new anti-inflammatory drug more efficient and stable, providing reliable process support for the company's large-scale production.

[0222] In order to further improve the production process of the drug, the optimization is focused on the concentration step. The concentration process is crucial for the development of new anti-inflammatory drugs because it affects the concentration of active ingredients in the drug and the recovery rate of the solvent, which directly affects the quality and production cost of the final product. In order to optimize the concentration process, a concentration kinetic model was established and an adaptive control strategy was introduced to optimize the concentration efficiency function. To evaluate the efficiency of the concentration stage.

[0223] The mathematical expression of the optimization function is:

[0224]

[0225] in, is the optimization function of the concentration process, which is used to evaluate the overall efficiency of the concentration stage; T v is the evaporation temperature, ranging from 40℃ to 80℃; P v is the vacuum degree, ranging from 0.1 to 0.5 atm; κ is the concentration time, ranging from 30 to 90 minutes; δ is the solvent recovery rate, ranging from 70% to 95%; M is the solid content of the concentrate, indicating the final product concentration of the concentration process; ξ, ζ, η are the weight coefficients of evaporation temperature, vacuum degree and solvent recovery rate in the optimization process, respectively, and their values ​​range from 0.1 to 1.0, reflecting the importance of each parameter to the concentration efficiency.

[0226] The setting is optimizing the evaporation concentration process, the goal of which is to concentrate the solution to 90% solid content and recover the solvent as much as possible. The initial setting conditions are: evaporation temperature T v =60℃, vacuum degree P v =0.3atm, concentration time κ=60 minutes, solvent recovery rate δ=80%. Through experiments and preliminary tests, it was determined that the solid content M of the concentrate under these conditions was 85%.

[0227] To further optimize the enrichment step, use the optimize function To balance the effects of temperature, vacuum degree and solvent recovery rate on concentration efficiency. The weight coefficients are set to ξ=0.6, ζ=0.3, η=0.1. These coefficients reflect that the evaporation temperature has the greatest impact on the solid content, followed by the vacuum degree, while the solvent recovery rate has a relatively small impact.

[0228] Next, by monitoring the experimental data, the derivatives of each variable with respect to the solid content are calculated. For example, the effect of evaporation temperature on solid content is The effect of vacuum degree on solid content is The effect of solvent recovery rate on solid content is

[0229] Substitute this data into the optimization function:

[0230]

[0231] Further calculation yields:

[0232]

[0233] This optimization function value It indicates the comprehensive efficiency of the concentration process under the current conditions. For further optimization, the evaporation temperature was increased to 65°C, the vacuum degree was reduced to 0.25 atm, and the solvent recovery rate was maintained above 85%. Under the new experimental conditions, after 40 minutes of concentration time, the solid content of the concentrate reached 90% and the solvent recovery rate reached 92%.

[0234] Through repeated adjustments and optimizations, the optimal concentration conditions were gradually found, ensuring that the concentration process of the agent maximized the solid content while improving the solvent recovery rate and reducing energy consumption and time costs.

[0235] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An optimized method for preparing a drug, characterized in that The following steps are involved: S1. Ingredient optimization and dynamic deployment: S1.

1. Use a deep neural network machine learning model to screen and optimize the active ingredients and excipients of the drug, and establish a relationship model between ingredients and efficacy and stability; S1.2, using a dynamic formulation module to adjust the proportion and dosage of ingredients in real time according to different preparation conditions including temperature and pH value; S2. Adaptive optimization of reaction conditions: S2.

1. Based on S1, an adaptive reaction condition adjustment system is constructed to monitor the reaction conditions including temperature, pressure, pH value and stirring speed in the preparation process in real time; S2.2, real-time adjustment of reaction conditions through adaptive neuro-fuzzy inference system; S3, Dynamic mixing and reaction monitoring: S3.

1. Introduce a dynamic mixing system and use microfluidic technology to control the metering and mixing of pharmaceutical ingredients; S3.2, and the integrated online monitoring module detects the flow rate, concentration and temperature key parameters of the mixed liquid in real time, and adjusts the mixing conditions according to the monitoring data; S4, step-by-step optimization synthesis strategy: S4.

1. Based on the optimization results of S1 to S3, the drug preparation process is decomposed into multiple key steps including primary reaction, purification, and concentration, and each step is optimized separately; S4.

2. Use genetic algorithm to comprehensively consider the coupling relationship between each step and optimize the time, temperature and reaction dose parameters of each step; S5. Real-time quality monitoring and feedback adjustment: S5.

1. Build a real-time quality monitoring system to monitor the physical and chemical properties of the drug, including particle size distribution, solubility, and purity, in real time through sensors and imaging equipment; S5.2, combine machine learning models to predict the quality of the drug. If the quality deviates from expectations, immediately send feedback signals to S2 and S3 to adjust the reaction conditions and mixing strategies to correct the deviation; S6, full process optimization and integration platform: S6.

1. Introduce an integrated platform to centrally manage and optimize the data of each module from S1 to S5, including component screening, reaction condition adjustment, dynamic mixing, and quality monitoring; The dynamic mixing system comprises the following steps: S1, first use microfluidics technology to measure and mix the ingredients with micron-level precision; S2. Then, the component flow rate and mixing ratio are adjusted in combination with the real-time monitoring data, and a dynamic response function Ψ(t) is set to characterize the adjustment of the component flow rate and mixing ratio at different times. The expression is: Among them, Ψ(t) is the feedback response function of the real-time adjustment process, β j is the control coefficient for adjusting the response speed, which is used to weigh the sensitivity of different components to the adjustment speed. j (t) is the real-time flow rate of the jth component, indicating the actual flow rate of the component during the preparation process, μ j (t) is the expected flow rate of the jth component, which represents the ideal flow rate required for the component according to the current conditions; S3. Finally, the integration of microfluidic system and microreactor is used to coordinate the mixing and reaction process of components, and an integrated optimization function Ω(x) is constructed to describe the reaction efficiency under the conditions of entering the reaction stage after mixing. The formula is: Among them, Ω(x) is the reaction efficiency optimization function, κ is the reaction rate constant, which represents the reaction rate under specific conditions; γ is the component degradation coefficient, which describes the predicted degradation or loss of the component during the transmission process, e -γx is the component concentration attenuation factor, v(x) is the fluid velocity field, which indicates the flow velocity of the reactants inside the microreactor, is the divergence of the velocity field, reflecting the diffusion characteristics of the fluid; v is the spatial area inside the reactor.

2. The method for optimizing the preparation of a drug according to claim 1, characterized in that The establishment of the relationship model between ingredients, efficacy and stability includes: S1. First, design a multi-layer deep neural network structure, where the input layer receives the feature vector of the drug composition: X=[x1,x2,…,x n ] Each component feature vector is represented by several physical and chemical properties, including molecular structure, polarity and solubility; through multiple hidden layers H i =f(W i ·H i-1 +b i ), where f is a custom nonlinear activation function, W i and b i are the weight matrix and bias vector of the i-th layer respectively; The output layer provides efficacy ε and stability The prediction index of the network is minimized by the loss function: Conduct training, including is a custom difference function used to measure the gap between the actual value and the predicted value, γ j (X) is the adaptive weight coefficient, which is used to adjust the contribution of different component combinations; S2. Then the ingredient screening and combination optimization method includes using the trained deep neural network DNN model to perform combination analysis on potential active ingredients and excipients; S3. Finally, the deep neural network DNN model outputs the weight matrix M of the effect of component combination on efficacy and stability during the screening and optimization process. ij ], where m ij is the weight of the influence of the i-th ingredient on the j-th efficacy or stability index, through the formula: where α ij Based on the correlation coefficient between ingredients and efficacy, the key ingredients and their interactions that have the greatest impact on efficacy and stability are identified, and optimization suggestions based on scientific analysis are provided using the multidimensional relationship model λ(M).

3. The method for optimizing the preparation of a drug according to claim 1, characterized in that The process of constructing the adaptive reaction condition regulation system includes: S1. First, the adaptive weighted average method and Kalman filter technology are used to perform data fusion and set the comprehensive reaction condition function Θ(t) to describe the comprehensive state of each reaction condition in the entire preparation process: Among them, ω i , ν j and χ k They are the adaptive weight functions of temperature, pressure and pH sensors, representing the importance of each sensor data at different times, T i (t), P j (t) and pH k (t) represents the temperature, pressure and pH value data at different times; S2. Then, by learning historical data and real-time sensor data, the nonlinear effects of various reaction conditions on the quality and stability of the drug are controlled; adaptive optimization is used to set the optimization function: Where X(t) is the state vector of the reaction conditions, including temperature, pressure, pH value and stirring speed, α i is the adjustment weight of the corresponding condition, controlling its influence on the overall optimization, β is the weighting coefficient, which is used in the second-order derivative term to capture the acceleration change of the reaction condition, and Γ(X) is the objective function of the drug quality; S3. Finally, edge computing and distributed control architecture are adopted to deploy data processing and decision-making parts on the local side of the preparation equipment, and local computing resources are used to analyze sensor data in real time.

4. The method for optimizing the preparation of a drug according to claim 1, characterized in that In the primary reaction step, a deep neural network is used to predict the yield and quality of the reaction product, a relationship model between the reaction conditions and the product quality is constructed, and a reaction optimization function is defined: in is the primary reaction optimization function, which represents the comprehensive reaction efficiency within a given time t; T is the reaction temperature, P is the reaction pressure, S is the stirring speed; Y is the yield of the reaction product; α, β, γ are the weight coefficients of temperature, pressure and stirring speed in the optimization process, which are used to adjust the influence of each variable on the product yield.

5. The method for optimizing the preparation of a drug according to claim 1, characterized in that In the purification step, the purification process is optimized by a multi-stage separation model including multi-stage extraction, fractionation or chromatographic separation, and a multi-objective optimization function is constructed: in is the purification process optimization function, which is used to evaluate the overall efficiency of the purification process; ρ is the solvent concentration, F is the flow rate, E is the energy consumption, and τ is the purification time; C is the purity of the purified product, and μ, v, and λ are the weight coefficients for optimizing the solvent concentration, flow rate, and energy consumption, respectively, indicating the importance of each optimization objective in the overall optimization process.

6. The method for optimizing the preparation of a pharmaceutical preparation according to claim 1, characterized in that In the optimization process of the concentration step, a concentration kinetic model is established, the operating parameters of the evaporation or concentration equipment are optimized, and an adaptive control strategy is introduced to optimize the concentration efficiency function: in is the optimization function for the concentration process, used to evaluate the efficiency of the concentration stage; T v is the evaporation temperature, P v is the vacuum degree, κ is the concentration time, δ is the solvent recovery rate; M is the solid content of the concentrate, indicating the final product concentration of the concentration process; ξ, ζ, η are the optimization weight coefficients of each parameter, reflecting the important proportion of evaporation temperature, vacuum degree and solvent recovery rate in the optimization process.

Images