A method for preparing a veterinary compound tilmicosin solution

Abstract

The application provides a preparation method of a veterinary compound tilmicosin solution, comprising the following steps: according to the energy adjustment requirement in the temperature compensation signal, a coupling relationship model of temperature adjustment and stirring intensity is established, corresponding stirring speed data is obtained, the stirring speed data comprises a stirring speed change sequence and a stirring power change sequence, linear smoothing processing based on a moving average filtering principle is performed on a mutation point and fluctuation noise existing in the stirring speed data, and a target stirring speed is obtained.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for preparing a veterinary compound tilmicosin solution. Background Technology

[0002] Efficient delivery of veterinary drug formulations is crucial for disease control in animal husbandry. Tilmicosin, a broad-spectrum antibacterial drug, has significant efficacy in treating respiratory and digestive tract infections in livestock and poultry. The development of its compound formulations is essential for improving clinical treatment outcomes. However, existing methods for preparing tilmicosin solutions often focus on the selection of a single solvent system, neglecting the impact of interactions between compound components on the stability of the final product. This leads to precipitation or stratification during storage, severely affecting drug bioavailability. Tilmicosin itself has extremely low solubility in water, a physicochemical property that directly restricts the development of aqueous formulations. When it needs to be compounded with other active ingredients, the dissolution behavior of different components in the mixed solvent varies significantly. Some components may preferentially precipitate under specific solvent ratios, while others may compete with the solvent for dissolution. This mismatch in dissolution behavior can lead to the risk of phase separation during formulation, further resulting in uneven distribution of the active ingredients in the solution. For example, in actual production, if tilmicosin is added first and then other components, microcrystals may form due to excessively high local concentrations. Once these crystals form, they are difficult to dissolve, and even subsequent adjustments to the solvent ratio or extended stirring time cannot completely dissolve them, ultimately leading to inaccurate formulation content.

[0003] More complexly, the operating parameters during the preparation process have a cascading effect on the dissolution equilibrium. Changes in the order of adding ingredients can lead to instantaneous changes in the local solvent environment; insufficient stirring speed may cause the solute to aggregate due to failure to disperse in time; and temperature fluctuations will alter the solubility profiles of each component. These factors intertwine and jointly determine the homogeneity and stability of the final solution. Therefore, establishing a systematic preparation process that optimizes the composition of the mixed solvent system, precisely controls the order of adding each component, and rationally sets stirring and heating parameters to ensure that all active ingredients in the compound achieve true molecular-level dispersion in the solution and maintain a stable state over a long period has become a key issue in the preparation of veterinary compound tilmicosin solution. Summary of the Invention

[0004] This invention provides a method for preparing veterinary compound tilmicosin solution, mainly comprising: The component dissolution data of the target drug component is obtained. The component dissolution data includes the saturated solubility information and dissolution rate information of each component in different solvent systems. The component dissolution data is classified into multiple categories based on the solubility difference characteristics and dissolution kinetic characteristics using a classification algorithm. A component solubility classification model is established. Each component is divided into high solubility category, medium solubility category and low solubility category using the component solubility classification model. A first dissolution feature set containing solubility classification labels and dissolution behavior characteristic parameters is obtained. Based on the solubility classification labels and solubility behavior feature parameters in the first solubility feature set, an association mapping relationship between solubility features and solvent ratio is established, and corresponding solvent ratio data in the historical experimental database is obtained. The solvent ratio data includes the proportion of main solvent, the proportion of co-solvent, and the proportion of diluent. A regression algorithm is used to construct a solvent ratio prediction model based on the solubility behavior feature parameters. The solvent ratio prediction model is used to perform regression prediction calculation on the solvent ratio data to obtain an initial solvent ratio including the initial proportion of main solvent, the initial proportion of co-solvent, and the initial proportion of diluent. For each component material corresponding to the initial solvent ratio, the feeding sequence data of each component material is obtained. The feeding sequence data includes the addition time node and addition amount information of each component. A dynamic concentration simulation model is established to simulate the feeding process. The local concentration characteristics of each time node in the feeding sequence data are extracted. If the peak concentration value corresponding to the local concentration characteristic is greater than the preset concentration threshold, the feeding sequence data is rearranged according to the concentration gradient optimization principle to obtain the target feeding sequence. The actual feeding operation is carried out according to the target feeding sequence. Temperature fluctuation data during the mixing process is monitored and acquired in real time using a temperature sensor array. The temperature fluctuation data includes temperature change curves at various spatial locations of the mixing system. A temperature stability assessment model is established to determine whether the temperature fluctuation data exceeds the preset temperature threshold range. If the temperature deviation value in the temperature fluctuation data exceeds the preset temperature threshold range, the temperature fluctuation data is reversed and compensated based on the thermodynamic balance compensation principle to generate a temperature compensation signal. Based on the energy regulation requirements in the temperature compensation signal, a coupling relationship model between temperature regulation and stirring intensity is established, and the corresponding stirring speed data is obtained. The stirring speed data includes the stirring speed change sequence and the stirring power change sequence. The abrupt change points and fluctuation noise in the stirring speed data are linearly smoothed based on the moving average filtering principle to obtain the target stirring speed. The mixed solvent is continuously stirred at the target stirring speed to obtain the processed dissolution equilibrium data, which includes a solution transmittance change curve and a microscopic particle distribution image. The amount of crystal precipitation is extracted from the dissolution equilibrium data using an image recognition algorithm and a particle size analysis algorithm. If the amount of crystal precipitation is greater than a preset precipitation threshold, the component dissolution data is re-acquired and the component solubility classification model parameters are updated for iterative processing until the amount of crystal precipitation is less than or equal to the preset precipitation threshold, thus obtaining the target drug solution.

[0005] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for preparing veterinary compound tilmicosin solution. It addresses the intertwined challenges of multi-component drug formulation, including significant differences in component solubility, improper feeding sequence leading to localized supersaturation, temperature fluctuations causing crystallization, and unstable stirring resulting in uneven dissolution. By acquiring saturated solubility and dissolution rate data of each component in different solvent systems, a classification algorithm is used to establish a component solubility classification model, dividing the components into high, medium, and low solubility categories and forming a set of solubility characteristics. Furthermore, combined with historical experimental data, a regression algorithm is used to construct a solvent ratio prediction model, outputting the main solvent, co-solvent, and dilution ratio. The initial ratio of the agent was determined; then, the local concentration characteristics of each time node during the feeding process were extracted through a dynamic concentration simulation model. Based on the concentration gradient optimization principle, the nodes exceeding the threshold were rearranged in time to obtain the target feeding sequence. The mixing temperature fluctuation was monitored in real time using a temperature sensor array. A temperature compensation signal was generated based on the thermodynamic balance compensation principle. The stirring intensity was coupled with the stirring speed and smoothed using a moving average filter. Finally, the crystal precipitation amount was iteratively updated and the classification model parameters were updated by combining transmittance and particle size analysis. This significantly improved the dissolution uniformity, stability and batch consistency of the compound tilmicosin solution. Attached Figure Description

[0006] Figure 1 This is a flowchart of a method for preparing a veterinary compound tilmicosin solution according to the present invention.

[0007] Figure 2 This is a schematic diagram illustrating a method for preparing a veterinary compound tilmicosin solution according to the present invention.

[0008] Figure 3 This is another schematic diagram of a method for preparing a veterinary compound tilmicosin solution according to the present invention. Detailed Implementation

[0009] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0010] like Figures 1-3 The preparation method of a veterinary compound tilmicosin solution in this embodiment may specifically include: S101. Obtain component dissolution data of the target drug component. The component dissolution data includes saturated solubility information and dissolution rate information of each component in different solvent systems. The component dissolution data is classified into multiple categories based on solubility difference characteristics and dissolution kinetic characteristics using a classification algorithm. A component solubility classification model is established. Each component is divided into high solubility category, medium solubility category and low solubility category using the component solubility classification model. A first dissolution feature set containing solubility classification labels and dissolution behavior characteristic parameters is obtained.

[0011] The process involves acquiring component solubility data for the target drug components, including saturated solubility and solubility rate information for each component in different solvent systems. Based on the saturated solubility and solubility rate information, solubility difference characteristics and dissolution kinetics characteristics of each component are extracted to obtain an initial solubility feature vector. If the initial solubility feature vector contains missing values, imputation is performed using mean imputation to obtain a complete solubility feature vector. The complete solubility feature vector is then subjected to dimensionality reduction using principal component analysis (PCA) to extract principal component features, resulting in a dimensionality-reduced solubility feature vector. This dimensionality-reduced solubility feature vector is then subjected to multi-class classification using a classification algorithm to construct a component solubility classification model, and the classification prediction results output by the model are obtained. Finally, if the classification prediction results meet a preset confidence threshold, each component is categorized into high-solubility, medium-solubility, and low-solubility categories based on the prediction results. Obtain the high solubility category, the medium solubility category, and the low solubility category; generate corresponding solubility classification labels based on the high solubility category, the medium solubility category, and the low solubility category; extract the solubility behavior parameters corresponding to each component; and obtain a first solubility feature set containing the solubility classification labels and the solubility behavior parameters.

[0012] For example, in a drug component solubility analysis scenario, the saturated solubility and solubility rate of the target drug component in different solvent systems such as water and ethanol are first obtained. By comparing the saturated solubility of the component in water with that in ethanol, solubility difference characteristics are extracted. Simultaneously, solubility kinetic characteristics are extracted based on the rate of change of solubility of the component over a specific time period, thereby constructing an initial solubility feature vector. This approach comprehensively quantifies the thermodynamic and kinetic behavior of drug molecules under different environments, providing fundamental data support for subsequent analyses.

[0013] In one possible implementation, the initial dissolution feature vector may contain missing data due to experimental measurement errors or equipment malfunctions. If a missing value is detected in the solubility feature of a certain component, the values ​​of that feature for all known components in the same solvent system are extracted, their arithmetic mean is calculated, and this mean is used to fill in the missing positions, thereby obtaining a complete dissolution feature vector. This mean-filling method can effectively maintain the overall stability of the data distribution and avoid information loss due to sample removal.

[0014] It should be noted that the complete dissolution feature vector usually contains a large number of redundant variables. This vector is input into a principal component analysis algorithm, where orthogonal transformations convert the original high-dimensional correlated variables into a few linearly independent principal component features. Assuming there are 20 original features, after dimensionality reduction, the top 5 principal components with a cumulative variance contribution rate of 90% are retained, resulting in the dimensionality-reduced dissolution feature vector. This significantly reduces computational complexity while preserving the core dissolution feature information.

[0015] For example, after obtaining the dimensionality-reduced solubility feature vector, it is input into a support vector machine (SVM) classification algorithm to perform multi-class classification processing, constructing a component solubility classification model. The model outputs the probability value of each component belonging to different solubility categories as the classification prediction result. If the probability of a component being predicted as a high-solubility category reaches a preset confidence threshold of 85%, it is strictly classified into the high-solubility category; similarly, medium-solubility and low-solubility categories are also classified. Setting a confidence threshold can significantly reduce the false positive rate and improve the reliability of the classification results.

[0016] In one possible implementation, based on the above classification results, a label 1 is assigned to the high solubility category, a label 2 to the medium solubility category, and a label 3 to the low solubility category, generating corresponding solubility classification labels. Subsequently, solubility behavior parameters such as the maximum dissolution rate and time to peak concentration of each component are extracted during the dissolution process and concatenated with the classification labels to obtain a first solubility feature set containing the classification labels and solubility behavior parameters. This feature set provides valuable structured input for subsequent drug formulation design and drug release behavior prediction.

[0017] S102. Based on the solubility classification labels and solubility behavior feature parameters in the first solubility feature set, establish the correlation mapping relationship between solubility features and solvent ratio, obtain the corresponding solvent ratio data in the historical experimental database, the solvent ratio data includes the proportion of main solvent, the proportion of co-solvent, and the proportion of diluent, use a regression algorithm to construct a solvent ratio prediction model based on the solubility behavior feature parameters, and perform regression prediction calculation on the solvent ratio data through the solvent ratio prediction model to obtain the initial solvent ratio including the initial proportion of main solvent, the initial proportion of co-solvent, and the initial proportion of diluent.

[0018] The first set of solubility features is used to obtain solubility classification labels and solubility behavior feature parameters. Historical solvent ratio data, including the proportions of the main solvent, co-solvent, and diluent, is retrieved from a historical experimental database to obtain an initial training sample set. Based on the solubility behavior feature parameters and historical solvent ratio data in the initial training sample set, a linear regression algorithm is used to construct a solvent ratio prediction model, obtaining the predicted ratio value output by the model. The difference between the predicted ratio value and the historical solvent ratio data is calculated to obtain a ratio deviation value. If the ratio deviation value is greater than a preset deviation threshold, the corresponding sample data in the initial training sample set is removed to obtain an updated training sample set. The linear regression algorithm is used to retrain the model based on the updated training sample set to determine the target solvent ratio prediction model. The target solvent ratio prediction model is used to perform regression prediction calculations on the currently input solubility behavior feature parameters to obtain an initial solvent ratio including the initial proportions of the main solvent, co-solvent, and diluent. The initial solvent ratio includes the initial proportions of the main solvent, the co-solvent, and the diluent. A summation calculation is performed to obtain the summation result. If the summation result is not equal to a preset total ratio threshold, the initial solvent ratio is normalized to obtain the target solvent ratio.

[0019] For example, in the design of drug solvent systems, the first step is to integrate the initial set of solubility characteristics acquired earlier with historical experience data. Specifically, this involves extracting historical solvent ratio data from the company's internal historical experimental database that shares the same solubility classification label as the current drug component. Assuming a drug component is labeled as highly soluble, its corresponding historical ratio records are extracted, including a main solvent ratio of 40%, a co-solvent ratio of 30%, and a diluent ratio of 30%. These historical ratio data are then correlated and bound with solubility behavior characteristic parameters such as the component's maximum dissolution rate and time to peak concentration to construct an initial training sample set.

[0020] In one possible implementation, based on the initial training sample set mentioned above, an initial solvent ratio prediction model is constructed using a linear regression algorithm. Dissolution behavior feature parameters are used as input independent variables, and historical solvent ratio data are used as output dependent variables. The linear mapping relationship between the features and the ratio is fitted using the least squares method. After the model's initial training is complete, the feature parameters from the training set are input back into the model, outputting the corresponding predicted ratio values. Subsequently, the difference between the predicted ratio values ​​and the actual historical solvent ratio data is calculated to obtain the ratio deviation value. Assuming that the predicted ratio of the main solvent for a certain sample is 55, while the actual historical ratio is 40, the calculated ratio deviation value is 15. If the system's preset deviation threshold is 10, since 15 is greater than 10, the sample is determined to be abnormal or high-noise data and is removed from the initial training sample set, thus obtaining a cleaned and updated training sample set.

[0021] It should be noted that after obtaining the updated training sample set, the linear regression algorithm is called again for secondary model training to determine the final target solvent ratio prediction model. The dissolution behavior characteristics of the drug to be analyzed are input into this target model, regression prediction calculation is performed, and the output includes an initial solvent ratio of 45 for the main solvent, 35 for the co-solvent, and 25 for the diluent. Then, these three initial ratio values ​​are summed, resulting in a ratio sum of 105. If the preset total ratio threshold is strictly set to 100, since 105 is not equal to 100, the initial solvent ratio must be normalized. By dividing each initial ratio by the sum of 105 and multiplying by the total sum threshold of 100, the proportion of each solvent is redistributed, and the final output is the target solvent ratio that meets the total amount constraint.

[0022] S103. For each component material corresponding to the initial solvent ratio, obtain the feeding sequence data of each component material. The feeding sequence data includes the addition time node and addition amount information of each component. Establish a dynamic concentration simulation model to simulate the feeding process. Extract the local concentration characteristics of each time node in the feeding sequence data. If the peak concentration value corresponding to the local concentration characteristic is greater than the preset concentration threshold, perform time-series rearrangement processing on the feeding sequence data based on the concentration gradient optimization principle to obtain the target feeding sequence.

[0023] Obtain the initial solvent ratio for each component, and extract the initial feeding sequence for each component, which includes the addition time point and amount of each component. Based on the addition time point and amount, perform simulation calculations on the initial feeding sequence using a dynamic concentration simulation model to obtain the local concentration value corresponding to each addition time point. Extract the maximum value from the local concentration values ​​to determine the local peak value corresponding to the maximum value. If the local peak value is greater than a preset concentration threshold, calculate the concentration difference between the local peak value and the concentration threshold. Based on the concentration difference, use a random forest algorithm to perform temporal rearrangement processing on the addition time points in the initial feeding sequence to obtain rearranged time points. Construct a new feeding sequence using the rearranged time points and the addition amount, and determine the new feeding sequence as the target feeding sequence.

[0024] For example, after obtaining the initial solvent ratio including the main solvent, co-solvent, and diluent, it is necessary to further determine the order in which each component is added. For a specific polymer dissolution process, the extracted initial addition sequence might be set as follows: add 50 kg of main solvent at minute 0, add 30 kg of co-solvent at minute 10, and add 20 kg of diluent at minute 20. This initial addition sequence clarifies the timing and amount of each component added, providing a basic operating benchmark for the subsequent dissolution process.

[0025] Specifically, the aforementioned addition time points and amounts were used as input data and imported into a dynamic concentration simulation model for simulation calculations. This model, based on fluid diffusion dynamics, simulates the mass transfer process inside the mixing container, thus outputting the local concentration values ​​corresponding to each addition time point. Assuming the simulation results show that at the 12-minute mark, because the co-solvent has just been added and has not yet fully diffused, the simulated local concentration reaches its maximum. This maximum value was extracted, and the local peak concentration was determined to be 85.

[0026] In one possible implementation, a preset concentration threshold of 70 is set. Since the local peak value of 85 is greater than the concentration threshold of 70, it indicates that the current feeding sequence is highly likely to cause local concentration oversaturation, leading to solute aggregation. In this case, the concentration difference between the local peak value and the concentration threshold is calculated to be 15. Subsequently, this concentration difference, the dissolution behavior characteristics of each component, and the initial addition time point are used as input features and fed into a pre-trained random forest algorithm. The random forest algorithm constructs multiple decision trees to perform parallel evaluation and regression prediction of the temporal features, outputting optimized rearranged time points.

[0027] For example, the algorithm output adjusts the addition time of the co-solvent to the 15th minute and advances the addition time of the diluent to the 5th minute. Finally, a new dosing sequence is constructed by rearranging these time points and the original addition amounts, and this new sequence is determined as the target dosing sequence. Using this target dosing sequence effectively avoids the risk of uneven dissolution caused by excessively high local concentrations, significantly improving the mixing uniformity and overall dissolution efficiency of the multi-component solvent system.

[0028] S104. Perform actual feeding operation according to the target feeding sequence, and use a temperature sensor array to monitor and obtain temperature fluctuation data in real time during the mixing process. The temperature fluctuation data includes temperature change curves at various spatial locations of the mixing system. Establish a temperature stability assessment model to determine whether the temperature fluctuation data exceeds the preset temperature threshold range. If the temperature deviation value in the temperature fluctuation data exceeds the preset temperature threshold range, perform reverse compensation calculation on the temperature fluctuation data based on the thermodynamic balance compensation principle to generate a temperature compensation signal.

[0029] The feeding operation is initiated by a target feeding sequence. Sensors are used to collect the initial temperature signal under mixed conditions, obtaining the initial temperature signal of a spatial point. Kalman filtering is performed on the initial temperature signal of the spatial point to obtain a denoised temperature fluctuation value. This denoised temperature fluctuation value is then obtained and spliced ​​using a time series reconstruction method to generate a temperature change line for the spatial point. This temperature change line is compared with a pre-established evaluation model to determine if it exceeds a preset threshold range. If it does, the portion exceeding the threshold range is extracted to determine the temperature deviation value. This temperature deviation value is then obtained, and a reverse calculation is performed based on the thermodynamic balance compensation principle to obtain a reverse compensation amount. A digital-to-analog conversion is performed on this reverse compensation amount to generate a temperature compensation signal.

[0030] In one possible implementation, after the target feeding sequence is initiated, the materials inside the reactor begin to mix. At this time, thermocouple sensors placed at different depths on the inner wall of the reactor collect the initial temperature signal under mixing conditions in real time. Due to the mechanical friction of the stirring blades and the release of the heat of solution of the materials, the collected initial temperature signal usually contains high-frequency noise. Using this initial temperature signal as input data, a Kalman filter algorithm is used for state estimation and noise covariance matrix update, and the denoised temperature fluctuation value is output.

[0031] For example, the sensor collects the original temperature of 85 degrees at a certain spatial point. After Kalman filtering, transient interference caused by stirring is eliminated, and the true temperature fluctuation value is 82 degrees.

[0032] For example, after obtaining the denoised temperature fluctuation values, a time series reconstruction method using a sliding window mechanism is employed to concatenate the discrete temperature fluctuation values ​​in chronological order, generating a temperature change line reflecting the heat accumulation process at that spatial point. This temperature change line is then input into a pre-established evaluation model based on a dynamic time warping algorithm. The model compares the input temperature change line with a standard reaction temperature curve to determine if it exceeds a preset threshold range. Assuming the preset normal temperature fluctuation threshold range is 75 to 80 degrees Celsius, if the currently reconstructed temperature change line reaches 82 degrees Celsius within a certain time period, it is determined to exceed the threshold range, and the 2 degrees exceeding the threshold range are extracted as a temperature deviation value.

[0033] It should be noted that, for the extracted temperature deviation value, a reverse calculation is performed based on the principle of thermodynamic balance compensation.

[0034] Specifically, the temperature deviation of 2 degrees and the specific heat capacity of the current material are input into the reverse heat conduction model. By solving the heat conservation equation, the excess heat that needs to be removed is calculated, thus obtaining the reverse compensation amount. For example, the calculated flow rate of the cooling medium is 50 liters per minute. Finally, based on this reverse compensation amount, digital-to-analog conversion is performed to convert the digital cooling medium flow rate into an analog voltage signal, generating a temperature compensation signal. This signal directly drives the regulating valve of the cooling jacket, completing the closed-loop regulation of the heat balance of the reaction system.

[0035] S105. Based on the energy regulation requirements in the temperature compensation signal, establish a coupling relationship model between temperature regulation and stirring intensity, and obtain the corresponding stirring speed data. The stirring speed data includes a stirring speed change sequence and a stirring power change sequence. Perform linear smoothing processing on the abrupt changes and fluctuation noise in the stirring speed data based on the moving average filtering principle to obtain the target stirring speed.

[0036] The energy regulation amount contained in the temperature compensation signal is obtained. A coupling relationship model between the temperature regulation amount and the stirring intensity is constructed based on the energy regulation amount, and the initial stirring speed data is output through the coupling relationship model. The initial stirring speed data is obtained, and a rotational speed change sequence and a power change sequence are extracted from it. First-order difference calculations are performed on the rotational speed change sequence and the power change sequence to obtain a speed difference sequence. Based on the speed difference sequence, if the absolute value of the difference in the speed difference sequence is greater than a preset abrupt change threshold, the corresponding data point is determined to be an abrupt change point, and an abnormal speed sequence containing the abrupt change point is extracted. For the abnormal speed sequence, a moving average filtering algorithm is used to perform linear smoothing processing to eliminate the abrupt change points and fluctuation noise in the abnormal speed sequence, resulting in a smooth speed sequence. The smooth speed sequence is obtained, and the smoothness value corresponding to the smooth speed sequence is calculated. If the smoothness value is less than a preset smoothness threshold, the smooth speed sequence is determined to be the target stirring speed. Based on the target stirring speed, the timestamp information corresponding to the target stirring speed is extracted, and a stirrer control command sequence is generated based on the timestamp information and output.

[0037] Specifically, after obtaining the energy regulation amount contained in the temperature compensation signal, a coupling relationship model between the temperature regulation amount and the stirring intensity needs to be constructed. This model receives the energy regulation amount as input features and uses a multinomial regression algorithm to fit the conversion relationship between thermal energy and mechanical energy, thereby outputting the initial stirring speed data.

[0038] For example, when the input energy regulation is 5000 joules, the initial stirring speed data output by the coupling model has a base value of 120 revolutions per minute, corresponding to a power of 800 watts. The rotational speed change sequence and power change sequence over a continuous time period are extracted from this initial data. A first-order difference calculation is performed on the rotational speed change sequence, i.e., the difference in rotational speed between two adjacent time points is calculated to obtain the speed difference sequence.

[0039] For example, a preset mutation threshold of 15 revolutions per minute is set. The velocity difference sequence is iterated; if the absolute value of the difference between adjacent time points reaches 20 revolutions per minute, exceeding the mutation threshold, then that data point is identified as a mutation point. In this case, the sequence containing the 10 data points before and after the mutation point is extracted as the abnormal velocity sequence. For this abnormal velocity sequence, a moving average filtering algorithm is used to perform linear smoothing. The sliding window size is set to 5, and the arithmetic mean of the data within the window is calculated sequentially to replace the original center point value, thereby eliminating mutation points and high-frequency fluctuation noise in the abnormal velocity sequence, resulting in a smooth velocity sequence.

[0040] It should be noted that after obtaining the smoothed speed sequence, its smoothness value needs to be further calculated to verify the processing results. The smoothness value is obtained by calculating the sequence variance. If the calculated smoothness value is 8, which is less than the preset smoothness threshold of 10, the smoothed speed sequence is determined to meet the stable operation requirements and is confirmed as the target stirring speed. Based on the target stirring speed, its corresponding timestamp information is extracted. Based on this timestamp information, a stirrer control command sequence containing the start time, target speed, and duration is generated. Finally, this command sequence is output to drive the mixing equipment to perform the corresponding stirring action.

[0041] S106. The mixed solvent is continuously stirred at the target stirring speed to obtain the processed dissolution equilibrium data. The dissolution equilibrium data includes the solution transmittance change curve and the microscopic particle distribution image. The amount of crystal precipitation is extracted from the dissolution equilibrium data using image recognition algorithm and particle size analysis algorithm. If the amount of crystal precipitation is greater than the preset precipitation threshold, the component dissolution data is re-acquired and the component solubility classification model parameters are updated for iterative processing until the amount of crystal precipitation is less than or equal to the preset precipitation threshold, thereby obtaining the target drug solution.

[0042] Continuous stirring is performed on the initial mixed solvent at a target stirring speed to obtain dissolution equilibrium data, including a transmittance curve and an initial microparticle distribution image. Background noise is removed from the initial microparticle distribution image using an image recognition algorithm to obtain the target microparticle distribution image. A particle size analysis algorithm is used to perform pixel area statistics and transmittance comparison on the target microparticle distribution image and the transmittance curve to extract the current crystal precipitation amount. If the current crystal precipitation amount is greater than a preset precipitation threshold, the current component dissolution data is re-acquired, and the component solubility classification model parameters are updated based on the current component dissolution data to obtain an updated component solubility classification model. The updated component solubility classification model is used to reclassify the current component dissolution data to determine a new target stirring speed. Continuous stirring of the initial mixed solvent is then performed again at the new target stirring speed until the final extracted crystal precipitation amount is less than or equal to the preset precipitation threshold, thus obtaining the target drug solution.

[0043] In one possible implementation, when acquiring dissolution equilibrium data, the transmittance change curve of the solution is collected by an online optical sensor installed on the side wall of the reactor, and the initial microscopic particle distribution image is captured by a built-in microscope camera.

[0044] Specifically, Gaussian filtering is used to remove background noise from the initial microparticle distribution image, which contains bubbles and impurities. The input is the initial image containing noise. By setting the filtering window size to a 5x5 matrix, a weighted average is calculated on the pixels to output a smooth image of the target microparticle distribution with clear edges.

[0045] For example, when extracting the current crystal precipitation amount, a watershed particle size analysis algorithm is used to process the target micro-particle distribution image. The target micro-particle distribution image is input, and the algorithm identifies particle edge contours and calculates the total area of ​​pixels in connected regions. Simultaneously, the transmittance value of the solution transmittance curve at the current time point is extracted. The pixel area ratio and transmittance value are input into a pre-constructed multiple linear regression model for comparison and analysis, outputting the mass concentration of the current crystal precipitation amount, for example, determining the current crystal precipitation amount to be 15 grams per liter.

[0046] In one possible implementation, a preset precipitation threshold of 5 grams per liter is set. When the current crystal precipitation amount is detected to reach 15 grams per liter, which is greater than this threshold, a model update mechanism is triggered.

[0047] Specifically, the current temperature, concentration, and stirring speed are collected as the current component solubility data, and these are used as training samples input into a support vector machine classification model. The hyperplane weight parameters within the model are adjusted using a gradient descent algorithm, outputting an updated component solubility classification model. This updated model is then used to reassess the current state, outputting a new target stirring speed of 300 revolutions per minute, and a control command is issued to continue stirring until the crystal precipitation decreases to 4 grams per liter, completing the preparation of the target drug solution.

[0048] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for preparing a veterinary compound tilmicosin solution, characterized in that, The method includes: The component dissolution data of the target drug component is obtained. The component dissolution data includes the saturated solubility information and dissolution rate information of each component in different solvent systems. The component dissolution data is classified into multiple categories based on the solubility difference characteristics and dissolution kinetic characteristics using a classification algorithm. A component solubility classification model is established. Each component is divided into high solubility category, medium solubility category and low solubility category using the component solubility classification model. A first dissolution feature set containing solubility classification labels and dissolution behavior characteristic parameters is obtained. Based on the solubility classification labels and solubility behavior feature parameters in the first solubility feature set, an association mapping relationship between solubility features and solvent ratio is established, and corresponding solvent ratio data in the historical experimental database is obtained. The solvent ratio data includes the proportion of main solvent, the proportion of co-solvent, and the proportion of diluent. A regression algorithm is used to construct a solvent ratio prediction model based on the solubility behavior feature parameters. The solvent ratio prediction model is used to perform regression prediction calculation on the solvent ratio data to obtain an initial solvent ratio including the initial proportion of main solvent, the initial proportion of co-solvent, and the initial proportion of diluent. For each component material corresponding to the initial solvent ratio, the feeding sequence data of each component material is obtained. The feeding sequence data includes the addition time node and addition amount information of each component. A dynamic concentration simulation model is established to simulate the feeding process. The local concentration characteristics of each time node in the feeding sequence data are extracted. If the peak concentration value corresponding to the local concentration characteristic is greater than the preset concentration threshold, the feeding sequence data is rearranged according to the concentration gradient optimization principle to obtain the target feeding sequence. The actual feeding operation is carried out according to the target feeding sequence. Temperature fluctuation data during the mixing process is monitored and acquired in real time using a temperature sensor array. The temperature fluctuation data includes temperature change curves at various spatial locations of the mixing system. A temperature stability assessment model is established to determine whether the temperature fluctuation data exceeds the preset temperature threshold range. If the temperature deviation value in the temperature fluctuation data exceeds the preset temperature threshold range, the temperature fluctuation data is reversed and compensated based on the thermodynamic balance compensation principle to generate a temperature compensation signal. Based on the energy regulation requirements in the temperature compensation signal, a coupling relationship model between temperature regulation and stirring intensity is established, and the corresponding stirring speed data is obtained. The stirring speed data includes the stirring speed change sequence and the stirring power change sequence. The abrupt change points and fluctuation noise in the stirring speed data are linearly smoothed based on the moving average filtering principle to obtain the target stirring speed. The mixed solvent is continuously stirred at the target stirring speed to obtain the processed dissolution equilibrium data, which includes a solution transmittance change curve and a microscopic particle distribution image. The amount of crystal precipitation is extracted from the dissolution equilibrium data using an image recognition algorithm and a particle size analysis algorithm. If the amount of crystal precipitation is greater than a preset precipitation threshold, the component dissolution data is re-acquired and the component solubility classification model parameters are updated for iterative processing until the amount of crystal precipitation is less than or equal to the preset precipitation threshold, thus obtaining the target drug solution.

2. The method for preparing a veterinary compound tilmicosin solution according to claim 1, characterized in that, The process involves acquiring component dissolution data of the target drug components, including saturated solubility information and dissolution rate information of each component in different solvent systems. A classification algorithm is used to perform multi-category classification processing on the component dissolution data based on solubility difference characteristics and dissolution kinetic characteristics, establishing a component solubility classification model. This model is then used to classify each component into high-solubility, medium-solubility, and low-solubility categories, resulting in a first set of dissolution features containing solubility classification labels and dissolution behavior characteristic parameters, including: Acquire component dissolution data of the target drug component, wherein the component dissolution data includes saturated solubility information and dissolution rate information of each component in different solvent systems; based on the saturated solubility information and the dissolution rate information, extract the solubility difference characteristics and dissolution kinetic characteristics of each component to obtain an initial dissolution feature vector; The initial dissolution feature vector is obtained. If the initial dissolution feature vector has missing values, the missing values ​​are imputed using the mean imputation method to obtain the complete dissolution feature vector. The complete dissolution feature vector is obtained, and the dimensionality reduction of the complete dissolution feature vector is performed by the principal component analysis algorithm to extract the principal component features and obtain the dimensionality-reduced dissolution feature vector. The reduced-dimensional dissolution feature vector is obtained, and a multi-class classification process is performed on the reduced-dimensional dissolution feature vector using a classification algorithm to construct a component solubility classification model and obtain the classification prediction result output by the component solubility classification model. The classification prediction results are obtained. If the classification prediction results meet the preset confidence threshold, the components are classified according to the classification prediction results to obtain high solubility category, medium solubility category and low solubility category. Obtain the high solubility category, the medium solubility category, and the low solubility category; generate corresponding solubility classification labels based on the high solubility category, the medium solubility category, and the low solubility category; extract the solubility behavior parameters corresponding to each component; and obtain a first solubility feature set containing the solubility classification labels and the solubility behavior parameters.

3. The method for preparing a veterinary compound tilmicosin solution according to claim 1, characterized in that, The process involves establishing a mapping relationship between solubility features and solvent ratios based on the solubility classification labels and solubility behavior feature parameters in the first solubility feature set, obtaining corresponding solvent ratio data from the historical experimental database, whereby the solvent ratio data includes the proportion of the main solvent, the proportion of the co-solvent, and the proportion of the diluent. A regression algorithm is then used to construct a solvent ratio prediction model based on the solubility behavior feature parameters. This model is used to perform regression prediction calculations on the solvent ratio data to obtain an initial solvent ratio including the initial proportions of the main solvent, the co-solvent, and the diluent, comprising: Obtain the solubility classification labels and solubility behavior feature parameters contained in the first solubility feature set, and obtain the corresponding historical solvent ratio data from the historical experimental database. The historical solvent ratio data includes the proportion of main solvent, the proportion of co-solvent, and the proportion of diluent to obtain the initial training sample set. Based on the dissolution behavior feature parameters and historical solvent ratio data contained in the initial training sample set, a solvent ratio prediction model is constructed using a linear regression algorithm to obtain the predicted ratio value output by the solvent ratio prediction model. The difference between the predicted ratio and the historical solvent ratio data is calculated to obtain the ratio deviation value. If the ratio deviation value is greater than the preset deviation threshold, the corresponding sample data in the initial training sample set is removed to obtain an updated training sample set. The linear regression algorithm is used to retrain the model based on the updated training sample set to determine the target solvent ratio prediction model; The target solvent ratio prediction model is used to perform regression prediction calculations on the currently input dissolution behavior characteristic parameters to obtain an initial solvent ratio that includes the initial ratio of the main solvent, the initial ratio of the co-solvent, and the initial ratio of the diluent. The initial solvent ratio includes the initial proportions of the main solvent, the co-solvent, and the diluent. A summation calculation is performed to obtain the summation result. If the summation result is not equal to a preset total ratio threshold, the initial solvent ratio is normalized to obtain the target solvent ratio.

4. The method for preparing a veterinary compound tilmicosin solution according to claim 1, characterized in that, For each component material corresponding to the initial solvent ratio, the feeding sequence data of each component material is obtained. The feeding sequence data includes the addition time node and addition amount information of each component. A dynamic concentration simulation model is established to simulate the feeding process. Local concentration features of each time node in the feeding sequence data are extracted. If the peak concentration value corresponding to the local concentration feature is greater than a preset concentration threshold, the feeding sequence data is time-series rearranged based on the concentration gradient optimization principle to obtain the target feeding sequence, including: Obtain the components corresponding to the initial solvent ratio, and extract the initial feeding sequence of each component, which includes the addition time point and addition amount of each component; Based on the addition time point and the addition amount, a dynamic concentration simulation model is used to perform simulation calculations on the initial feeding sequence to obtain the local concentration value corresponding to each addition time point; Extract the maximum value from the local concentration values ​​and determine the local peak value corresponding to the maximum value; If the local peak value is greater than the preset concentration threshold value, then the concentration difference between the local peak value and the concentration threshold value is calculated. Based on the concentration difference, the random forest algorithm is used to rearrange the addition time points in the initial feeding sequence to obtain the rearranged time points; A new feeding sequence is constructed by using the rearranged time points and the added amount, and the new feeding sequence is determined as the target feeding sequence.

5. The method for preparing a veterinary compound tilmicosin solution according to claim 1, characterized in that, The actual feeding operation is carried out according to the target feeding sequence. A temperature sensor array is used to monitor and acquire temperature fluctuation data in real time during the mixing process. The temperature fluctuation data includes temperature change curves at various spatial locations of the mixing system. A temperature stability assessment model is established to determine whether the temperature fluctuation data exceeds a preset temperature threshold range. If the temperature deviation value in the temperature fluctuation data exceeds the preset temperature threshold range, a reverse compensation calculation is performed on the temperature fluctuation data based on the thermodynamic balance compensation principle to generate a temperature compensation signal, including: The feeding operation is started by the target feeding sequence, and the initial temperature signal under the mixed state is collected by the sensor to obtain the initial temperature signal of the spatial point; Kalman filtering is performed on the initial temperature signal of the spatial point to obtain the denoised temperature fluctuation value. The denoised temperature fluctuation values ​​are obtained, and the time series reconstruction method is used to splice the denoised temperature fluctuation values ​​to generate a temperature change line of spatial points. By inputting the temperature change line of the spatial point into a pre-established evaluation model for comparison, it is determined whether the temperature change line of the spatial point exceeds the preset threshold area. If the temperature change line of the spatial point exceeds the preset threshold area, then the portion of the temperature change line of the spatial point that exceeds the preset threshold area is extracted to determine the temperature deviation value. The temperature deviation value is obtained, and the temperature deviation value is subjected to inverse calculation based on the thermodynamic balance compensation principle to obtain the inverse compensation amount; The inverse compensation amount is used for digital-to-analog conversion to generate temperature compensation information.

6. The method for preparing a veterinary compound tilmicosin solution according to claim 1, characterized in that, Based on the energy adjustment requirements in the temperature compensation signal, a coupling relationship model between temperature adjustment and stirring intensity is established to obtain corresponding stirring speed data. This stirring speed data includes a stirrer speed variation sequence and a stirring power variation sequence. Linear smoothing processing based on the moving average filtering principle is applied to the abrupt changes and fluctuation noise in the stirring speed data to obtain the target stirring speed, including: The energy regulation amount contained in the temperature compensation signal is obtained, and a coupling relationship model between the temperature regulation amount and the stirring intensity is constructed based on the energy regulation amount. The initial stirring speed data is output through the coupling relationship model. The initial stirring speed data is obtained, and the rotation speed change sequence and power change sequence are extracted from the initial stirring speed data. First-order difference calculation is performed on the rotation speed change sequence and the power change sequence to obtain the speed difference sequence. According to the velocity difference sequence, if the absolute value of the difference in the velocity difference sequence is greater than a preset mutation threshold, the corresponding data point is determined to be a mutation point, and an abnormal velocity sequence containing the mutation point is extracted. For the abnormal velocity sequence, a moving average filtering algorithm is used to perform linear smoothing to eliminate the abrupt change points and fluctuation noise in the abnormal velocity sequence, resulting in a smooth velocity sequence. The smooth speed sequence is obtained, and the smoothness value corresponding to the smooth speed sequence is calculated. If the smoothness value is less than a preset smoothness threshold, the smooth speed sequence is determined to be the target stirring speed. Based on the target stirring speed, extract the timestamp information corresponding to the target stirring speed, generate a stirrer control command sequence based on the timestamp information, and output the stirrer control command sequence.

7. The method for preparing a veterinary compound tilmicosin solution according to claim 1, characterized in that, The mixed solvent is continuously stirred at the target stirring speed to obtain processed dissolution equilibrium data, which includes a solution transmittance change curve and a microscopic particle distribution image. Crystal precipitation is extracted from the dissolution equilibrium data using image recognition and particle size analysis algorithms. If the crystal precipitation is greater than a preset precipitation threshold, the component dissolution data is re-acquired, and the component solubility classification model parameters are updated iteratively until the crystal precipitation is less than or equal to the preset precipitation threshold, thus obtaining the target drug solution, including: The initial mixed solvent is continuously stirred according to the target stirring speed to obtain the dissolution equilibrium data after stirring. The dissolution equilibrium data includes the solution transmittance change curve and the initial micro-particle distribution image. The initial microparticle distribution image is processed by removing background noise using an image recognition algorithm to obtain the target microparticle distribution image. A particle size analysis algorithm is used to perform pixel area statistics and transmittance comparison on the target micro-particle distribution image and the solution transmittance change curve to extract the current crystal precipitation amount. If the current amount of crystal precipitation is greater than the preset precipitation threshold, the current component dissolution data is reacquired, and the component solubility classification model parameters are updated according to the current component dissolution data to obtain the updated component solubility classification model. The updated component solubility classification model is used to reclassify the current component solubility data and determine a new target stirring speed. According to the new target stirring speed, continue stirring of the initial mixed solvent until the amount of final extracted crystals is less than or equal to the preset precipitation threshold to obtain the target drug solution.

Images