Traditional Chinese medicine drying quality control method and system based on artificial intelligence

Abstract

The present application relates to the technical field of traditional Chinese medicine drying, and particularly relates to a traditional Chinese medicine drying quality control method and system based on artificial intelligence, which comprises determining the temperature of drying gas flowing into a drying tower, the concentration of a material liquid to be dried, the atomization pressure of the material liquid to be atomized, and the drying time; establishing a drying quality prediction model, inputting the temperature of the drying gas, the concentration of the material liquid to be dried, the atomization pressure, and the drying time into the drying quality prediction model to obtain a drying quality prediction value; determining a drying quality target value, and analyzing the drying quality prediction value and the drying quality target value to obtain a drying quality difference result; inputting the drying quality difference result into a quality control model to obtain drying adjustment parameters, and adjusting each drying parameter according to the drying adjustment parameters. The method and system can stabilize the drying quality of traditional Chinese medicine and effectively reduce the adverse rate of the whole batch of dried products.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traditional Chinese medicine drying, in particular to a traditional Chinese medicine drying quality control method and system based on artificial intelligence. BACKGROUND

[0002] Traditional Chinese medicine spray drying is an effective technology commonly used for preparing traditional Chinese medicine solid dispersion, which can effectively retain the effective components of traditional Chinese medicine by rapidly evaporating water in the high-temperature airflow, and finally obtain traditional Chinese medicine powder or granular material. This method is suitable for preparing traditional Chinese medicine granules, powder, oral liquid and other preparations.

[0003] The traditional spray drying process of traditional Chinese medicine usually relies on manual control, and relies on experience judgment and timing operation to adjust the drying parameters. Various influencing factors (such as airflow temperature, humidity change, spray pressure, liquid concentration, etc.) in the drying process have high complexity and nonlinearity, and it is difficult to accurately grasp the changes of various key quality indicators depending on experience, resulting in unstable drying quality problems; at the same time, the discovery of quality abnormalities has a lag, and the drying parameters cannot be adjusted in time and effectively, resulting in a high rate of poor quality of part of batches. SUMMARY

[0004] Therefore, the technical problem to be solved by the present application is to overcome the problems of unstable spray drying quality and high rate of poor quality of part of batches in the prior art, and to provide a traditional Chinese medicine drying quality control method and system based on artificial intelligence, which can stabilize the drying quality of traditional Chinese medicine and effectively improve the rate of poor quality of all batches of drying products.

[0005] In order to solve the above technical problems, the present application provides a traditional Chinese medicine drying quality control method based on artificial intelligence, which comprises,

[0006] Introducing a drying gas flow into the drying tower and measuring the temperature of the drying gas flow;

[0007] Measuring the concentration of the to-be-dried liquid; atomizing the to-be-dried liquid, and the atomized to-be-dried liquid enters the drying tower; measuring the atomization pressure of the to-be-dried liquid;

[0008] Drying the atomized to-be-dried liquid in the drying tower by the drying gas flow, and measuring the drying time;

[0009] Establishing a drying quality prediction model, inputting the temperature of the drying gas flow, the concentration of the to-be-dried liquid, the atomization pressure and the drying time into the drying quality prediction model, and obtaining a drying quality prediction value; wherein the indicators of the drying quality include moisture content, effective component content, particle size and bulk density;

[0010] determine a drying quality target value, and analyze the drying quality prediction value and the drying quality target value to obtain a drying quality difference result;

[0011] input the drying quality difference result into a quality control model to obtain a drying adjustment parameter, and adjust the temperature of the drying gas flow, the atomization pressure, the concentration of the to-be-dried liquid, and the drying time according to the drying adjustment parameter.

[0012] In an embodiment of the present application, a drying quality prediction model is established, including,

[0013] obtain historical data information of a drying process of a same model liquid as the to-be-dried liquid, and extract sample data information related to drying quality;

[0014] configure weights for the sample data information in the sample data window; wherein the sample data window is divided into a plurality of time points according to time sequence, and the sample data information at each time point is configured with weights according to formula (1):

[0015]

[0016] i represents the time point sequence number; f(i) represents the weight of the i-th time point; ti represents the time difference between the current time point and the data collection time point; λ represents the decay rate parameter, which affects the speed of weight decay over time;

[0017] deeply learn the sample data information according to the configured weights to obtain the drying quality prediction model.

[0018] In an embodiment of the present application, after configuring the weights according to formula (1), the configured weights are further normalized according to the following method:

[0019]

[0020] wherein f(i)' represents the normalized weight of the i-th time point; n represents the number of time points; j represents the time point sequence number; f(j) represents the weight of the j-th time point;

[0021] deeply learn the sample data information according to the normalized weights to obtain the drying quality prediction model.

[0022] In an embodiment of the present application, the drying quality prediction model includes an input layer, an LSTM layer, a full connection layer, and an output layer;

[0023] The sample data information of the sample data window is converted into a feature matrix capable of convolution operation in the input layer;

[0024] The feature matrix is subjected to feature extraction in the LSTM layer, and the LSTM layer can identify abnormal quality features of sample data;

[0025] The feature extraction result is subjected to mapping and transformation in the full connection layer;

[0026] The dry quality prediction value is output through the output layer;

[0027] The LSTM layer is provided with an activation function, and the feature extraction result is subjected to nonlinear transformation through the activation function.

[0028] In an embodiment of the present application, the control method further comprises,

[0029] The dry quality measured value is collected;

[0030] The dry quality measured value and the dry quality prediction value are subjected to difference calculation to obtain a prediction difference result;

[0031] The model parameters of the dry quality prediction model are adjusted according to the prediction difference result.

[0032] In an embodiment of the present application, the model parameters of the dry quality prediction model are adjusted according to the prediction difference result, which comprises

[0033] The prediction difference result is subjected to difference distribution analysis and error bias trend analysis to obtain a prediction difference analysis result;

[0034] The number of neurons of the LSTM layer is determined according to the activation degree of the LSTM layer;

[0035] The number of neurons of the LSTM layer is adjusted according to the prediction difference analysis result.

[0036] In an embodiment of the present application, the number of neurons of the LSTM layer is determined according to the activation degree of the LSTM layer, which comprises,

[0037] The activation function output value of each neuron is collected for the LSTM layer to obtain an activation value set;

[0038] At least one of the maximum value, the minimum value, the average value and the standard value is calculated for the activation value set;

[0039] The number of neurons of the LSTM layer is determined according to the calculation result.

[0040] In an embodiment of the present application, the control method further comprises,

[0041] The dry quality measured value is collected;

[0042] The dry mass measured value is analyzed with the dry mass target value to obtain a running difference result;

[0043] The dry adjustment parameter is adjusted according to the running difference result.

[0044] In an embodiment of the present application, adjusting the dry adjustment parameter according to the running difference result comprises,

[0045] The dry mass measured value is analyzed with the dry mass target value to obtain a running difference result;

[0046] The running difference rate is calculated in the following way:

[0047]

[0048] Wherein, a represents the running difference rate; M2 represents the dry mass measured value; M1 represents the dry mass target value;

[0049] If the running difference rate is less than 5%, the current dry adjustment parameter is kept;

[0050] If the running difference rate is greater than or equal to 5%, the running difference rate and the dry mass measured value are input into the quality control model to adjust the dry adjustment parameter.

[0051] The second aspect, to solve the above technical problems, the present application provides a kind of based on artificial intelligence's traditional Chinese medicine drying quality control system, for executing the based on artificial intelligence's traditional Chinese medicine drying quality control method described above.

[0052] The above technical solutions of the present application have the following beneficial effects compared with the prior art:

[0053] The based on artificial intelligence's traditional Chinese medicine drying quality control method and system described in the present application combine multiple key parameters such as the temperature of drying air flow, the concentration of material to be dried, atomization pressure and drying time, use artificial intelligence technology to build a drying quality prediction model, which can predict the quality change in the drying process in multiple dimensions; by analyzing the drying quality difference result, adjusting each key parameter, the quality control of drying process is more accurate and efficient; at the same time, using deep learning model to realize dynamic optimization and early warning of quality abnormality in traditional Chinese medicine drying process, solve the problems of unstable spray drying quality and high rejection rate of part of batches, stabilize the drying quality of traditional Chinese medicine, and effectively improve the drying quality of the whole batch Product BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to make the content of the present application more easily understood, the present application will be further described in detail below according to specific embodiments of the present application and in conjunction with the drawings, wherein,​

[0055] Figure 1 Flow chart of the method for controlling the drying quality of traditional Chinese medicine based on artificial intelligence in the preferred embodiment of the present application;

[0056] Figure 2 Flow chart of the method for establishing a drying quality prediction model in the preferred embodiment of the present application;

[0057] Figure 3 Framework diagram of the drying quality prediction model in the preferred embodiment of the present application; DETAILED DESCRIPTION

[0058] The present application will be further described below in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it, but the embodiments are not limiting to the present application.

[0059] Embodiment one

[0060] Reference Figure 1 As shown in the drawings, the embodiment of the present application discloses a method for controlling the drying quality of traditional Chinese medicine based on artificial intelligence, which comprises,

[0061] S100, introducing a drying gas flow into a drying tower and measuring the temperature of the drying gas flow;

[0062] S200, measuring the concentration of the to-be-dried liquid; atomizing the to-be-dried liquid, and introducing the atomized to-be-dried liquid into the drying tower; and measuring the atomization pressure of the to-be-dried liquid;

[0063] S300, drying the atomized to-be-dried liquid in the drying tower by the drying gas flow, and measuring the drying time;

[0064] In a specific application scenario, the process of traditional Chinese medicine spray drying comprises:

[0065] Spray drying is a process in which liquid material (referred to as "liquid") is sprayed into fine droplets (i.e. "atomization") through a nozzle, and then contacted with a high-temperature gas flow (usually hot air), referred to as "drying gas flow" below, which rapidly evaporates water under the action of the gas flow, and finally forms a powder; the entire process usually includes the following steps:

[0066] Generating a drying gas flow: introducing a drying gas flow (usually 180-220℃ air) into a drying tower through an air heater and an air conveyor, and measuring the temperature of the drying gas flow in real time during the process of introducing the drying gas flow.

[0067] Pretreatment: the to-be-dried liquid is subjected to filtration, concentration, concentration adjustment and other treatments to make it suitable for spray drying; the concentration of the to-be-dried liquid is measured after the pretreatment is completed.

[0068] Spray process: The pre-processed material liquid to be dried is atomized by a spray device (such as a spray tower or a spray disc) into fine droplets into the drying tower, and the atomization pressure is measured during this process.

[0069] Drying process: The fine droplets are in contact with the drying gas flow, and the water in the droplets is rapidly evaporated, leaving the solid components to form a powder; as the water in the droplets evaporates, the powder quickly coagulates into tiny particles, which are eventually collected at the bottom of the drying tower, and the drying time is measured during this process.

[0070] S400, a drying quality prediction model is established, the temperature of the drying gas flow, the concentration of the material liquid to be dried, the atomization pressure and the drying time are input into the drying quality prediction model, and a drying quality prediction value is obtained; wherein the indicators of the drying quality include moisture content, effective ingredient content, particle size and bulk density;

[0071] In specific application scenarios, the target of the drying quality prediction model is to predict the final drying quality (drying quality indicators include moisture content, effective ingredient content, particle size and bulk density) according to the key parameters (including the temperature of the drying gas flow, the concentration of the material liquid to be dried, the atomization pressure and the drying time) in the drying process. In order to establish an effective drying quality prediction model, first of all, the historical data related to the drying process need to be collected, and the sample data related to the drying quality are extracted from the historical data, including process parameter data, such as the drying gas flow temperature in each drying period, the concentration of the material liquid to be dried before each drying starts, the atomization pressure in each atomization process and the drying time of each drying process, etc.; and quality evaluation data corresponding to each drying period. These data can be collected in real time through online sensors and laboratory analysis equipment. In order to improve the accuracy of the model, the collected data should include multiple drying periods to cover different drying process conditions. A deep learning model (such as a long short-term memory network LSTM, a deep neural network, a convolutional neural network, etc.) is used to perform deep learning on the historical data to obtain a drying quality prediction model. After the drying quality prediction model is trained and verified to be effective, it is put into actual production use. The drying gas flow temperature, the concentration of the material liquid to be dried, the atomization pressure and the drying time in the current drying process are input into the drying quality prediction model to obtain the prediction value of each indicator of the drying quality.

[0072] S500, determine a drying quality target value, and analyze the drying quality prediction value and the drying quality target value to obtain a drying quality difference result;

[0073] In specific application scenarios, the setting of the drying quality target value needs to consider not only the quality standards of traditional Chinese medicine, but also factors such as production process, equipment capacity, and characteristics of medicinal materials; the target value generally refers to the ideal or standard value of a key quality indicator (such as moisture content, effective ingredient content, particle size, and bulk density) in the drying process, which will serve as the basis for drying quality control and be used for comparison with the actual predicted value, thereby providing guidance for the adjustment of the production process. Among them, the drying quality target value can be determined according to various methods, such as determining the target value through historical data, determining the target value in combination with pharmacopoeia and standardized requirements, determining the target value through experiments, etc.

[0074] The drying quality target value is a determined value, and the drying quality difference result can be obtained by analyzing the difference between the drying quality predicted value and the drying quality target value.

[0075] S600, input the drying quality difference result into a quality control model to obtain a drying adjustment parameter, and adjust the temperature of the drying gas flow, the atomization pressure, the concentration of the to-be-dried liquid, and the drying time according to the drying adjustment parameter.

[0076] In specific application scenarios, the drying quality difference result is input into a quality control model, which should be an adaptive control model established for each process of the drying process according to the characteristics and requirements of the drying process of the to-be-dried liquid. For example, a model can be established based on neural network model, genetic algorithm, etc. According to the model operation and analysis result, the best drying parameters are obtained, including drying gas flow temperature, liquid concentration, atomization pressure, drying time, etc. The best drying quality indicators are obtained after running according to these drying parameters. For example, if the moisture content is too high, it may be due to too low spray drying temperature or too short drying time, at which time the moisture content can be adjusted by adjusting the temperature, air speed, and drying time of the spray drying tower; if the effective ingredient content is too low, the extraction concentration of the raw material or the extraction process can be adjusted; if the particle size distribution is uneven, it may be due to unstable nozzle pressure or too large droplet size, and adjusting the spray pressure and droplet size can help the uniformity of particle size; if the bulk density is not suitable, the powder compaction method can be changed or the material vibration density can be increased to optimize it.

[0077] The artificial intelligence-based traditional Chinese medicine drying quality control method combines multiple key parameters such as the temperature of the drying gas flow, the concentration of the to-be-dried liquid, the atomization pressure, and the drying time, uses artificial intelligence technology to construct a drying quality prediction model, can predict the quality change in the drying process in multiple dimensions, and through analysis of the drying quality difference results, adjusts each key parameter, thereby ensuring that the quality control of the drying process is more accurate and efficient. At the same time, the deep learning model is used to realize dynamic optimization and early warning of quality abnormalities in the traditional Chinese medicine drying process, solve the problems of unstable spray drying quality and high bad rate of some batches, stabilize the drying quality of traditional Chinese medicine, and effectively improve the drying bad rate of the whole batch.

[0078] Further, as shown in FIG. 1, Figure 2 The drying quality prediction model is established, including:

[0079] The historical data information of the drying process of the to-be-dried liquid is obtained, and sample data information related to the drying quality is extracted;

[0080] In a specific application scenario, the historical data information of each link in the drying process of the same type of liquid as the to-be-dried liquid is obtained, such as temperature, humidity, spray pressure, drying gas flow rate, drying time, spray amount, etc.; quality index data such as particle size, moisture content, active ingredient content, bulk density; and external environment data such as control humidity, temperature, feeding conditions, etc.; the data related to the drying quality evaluation index (such as) is extracted from these historical data information, which is used as sample data for drying quality prediction model training.

[0081] A sample data window is constructed, and the sample data information in the sample data window is configured with a weight; wherein the sample data window is divided into a plurality of time points according to time sequence, and the sample data information at each time point is configured with a weight according to formula (1):

[0082]

[0083] i represents the time point serial number; f(i) represents the weight of the i th time point; t i represents the time difference between the current time point and the data collection time point; λ represents the decay rate parameter, which affects the speed of weight decay over time;

[0084] Deep learning is performed on the sample data information according to the configured weight, and the drying quality prediction model is obtained.

[0085] In a specific application scenario, a sample data window is constructed according to a time sequence, and the sample data window is designed to collect sample data in a period of time (for example, every hour or every 5 minutes), and the data in each time window contains drying process parameters and quality indicators in a period of time. Assuming that the data window is T hours, all sample data (such as drying air flow temperature, humidity, spray pressure, moisture content, etc.) in each T-hour window constitutes a data set. The extracted sample data information is arranged in time sequence for subsequent learning and analysis. According to specific requirements and actual conditions, the capacity and time length of the sample data window are set, the capacity can represent the number of entries of sample data information stored in the window, and the time length represents the time span of each data entry. According to the set time step, the sample data window is updated, that is, the data information in the sample data window is updated regularly, so that the data in the window maintains a certain timeliness with the actual situation.

[0086] The data in the sample data window is configured with weights in time sequence. Among them, in the drying process, the most recent operation parameters usually have a greater impact on the current quality, and the farther historical data has a smaller impact; based on this, the weights are configured according to a specific exponential decay method to emphasize the data closer to the current time, and the weights are configured larger; that is, the data at the most recent time is configured with a higher weight, and the data away from the current time is configured with a lower weight. In the scheme of the embodiment of the present application, by introducing the time sequence weight, the drying quality prediction model can more accurately capture the influence of the recent data, thereby improving the prediction accuracy of the drying quality change trend.

[0087] Further, after configuring the weights according to formula (1), the configured weights are normalized according to the following method, and the sample data information is subjected to deep learning according to the weights after the normalization operation to obtain the drying quality prediction model.

[0088]

[0089] Wherein, f(i)' represents the normalized weight of the i-th time point; n represents the number of time points; j represents the time point sequence number; f(j) represents the weight of the j-th time point; As the denominator of the normalization, it is ensured that the configured weights of each time point still maintain a relative proportion after normalization, and the sum of all weights is equal to 1.

[0090] In the scheme of the embodiment of the present application, the weight normalization operation can ensure that the sum of all weights is 1, so that the weight distribution is more reasonable, and the influence of model training effect caused by some weights being too large or too small is avoided.

[0091] Specifically, in order to realize high-precision prediction and control of the drying process of traditional Chinese medicine, the embodiment of the present application proposes a drying quality prediction model based on deep learning. The model is specially designed with a multi-layer structure, including an input layer, an LSTM layer, a fully connected layer, and an output layer, to fully utilize the characteristics of time series data and perform effective feature extraction and prediction. Referring to Figure 3 As shown in the figure, the drying quality prediction model includes an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives sample data information as input, usually represented as a multi-dimensional array. The sample data information in the sample data window is converted into a feature matrix that can perform convolution operations in the input layer. The feature matrix is subjected to feature extraction in the LSTM layer, which can identify abnormal quality features of the sample data. The feature extraction results are mapped and transformed in the fully connected layer. The drying quality prediction value is output through the output layer. The LSTM layer sets an activation function to perform nonlinear transformation on the feature extraction results.

[0092] In a specific application scenario, the data at each time point (such as drying air temperature, concentration of material to be dried, atomization pressure, drying time, moisture content, active ingredient content, bulk density, and particle size) is converted into a feature matrix in the input layer. These feature matrices not only contain the original data, but can also be further optimized through additional preprocessing steps (such as normalization). Feature extraction is performed in the LSTM layer to identify abnormal quality features in the sample data. An activation function (such as tanh, ReLU, etc.) is set to enhance the expression ability of the model through nonlinear transformation. For example, using tanh as the activation function can help the LSTM layer better handle the balance between negative and positive values, while ReLU can help speed up the training process. The fully connected layer maps and transforms the feature extraction results output by the LSTM layer to provide a compact and effective feature vector for the final prediction. The output layer directly gives the key prediction indicators of drying quality, such as the predicted values of moisture content, active ingredient content, particle size, and bulk density. By utilizing the powerful time series processing capability of the LSTM layer and combining the feature mapping of the fully connected layer, the accuracy and reliability of drying quality prediction are improved.

[0093] Further, the control method further includes: collecting a drying quality measured value; calculating the difference between the drying quality measured value and the drying quality predicted value to obtain a prediction difference result; and adjusting the model parameters of the drying quality prediction model according to the prediction difference result.

[0094] The adjusting the model parameters of the drying quality prediction model specifically includes: performing difference distribution analysis and error bias trend analysis on the prediction difference result to obtain a prediction difference analysis result; determining the number of neurons of the LSTM layer according to the activation degree of the LSTM layer; and adjusting the number of neurons of the LSTM layer according to the prediction difference analysis result.

[0095] In a specific application scenario, the difference distribution analysis is used to evaluate the bias and error characteristics between the drying quality prediction value and the drying quality measured value. For each time step, the error value is obtained by subtracting the two values, and the difference distribution is obtained by statistically analyzing the error values of all time steps. According to the difference distribution, it is analyzed whether the error is concentrated in certain specific ranges or presents a large fluctuation. The error bias trend analysis is used to identify the law of the prediction error changing with time, whether it presents a gradually increasing or gradually decreasing trend, or whether it presents a periodic fluctuation. Such a trend may indicate that the model fails to fully capture the law of certain time series data, resulting in a large prediction error in a specific time period. The activation degree of the LSTM layer can be measured by calculating the output value of the activation function (such as the tanh or sigmoid function output). If the output of most neurons is close to 0 or 1, it indicates that the activation degree is too low or too high, which may cause gradient disappearance or gradient explosion. The number of neurons of the LSTM layer determines the expression ability and fitting ability of the network. According to the prediction difference analysis result, the number of neurons of the LSTM layer is dynamically adjusted: if the error is large and the activation degree of the LSTM layer is low, the number of neurons of the LSTM layer is increased to improve the complexity of the model, so that it can learn more features. If the prediction error presents an unstable trend (such as a large fluctuation in error), increasing the number of neurons can help the network better learn the patterns and laws of the input data. If the error is small and stable, and the activation degree is high, the current number of neurons is maintained to avoid overfitting of the model.

[0096] Specifically, the determining the number of neurons of the LSTM layer according to the activation degree of the LSTM layer includes: collecting the output value of the activation function of each neuron of the LSTM layer to obtain an activation value set; calculating at least one of the maximum value, the minimum value, the average value and the standard value of the activation value set; and determining the number of neurons of the LSTM layer according to the calculation result. The method of determining the number of neurons according to the activation degree can make the model more flexible and adapt to the needs of different data and tasks. By dynamically adjusting the number of neurons, the performance and generalization ability of the model can be improved, and problems such as overfitting or underfitting can be avoided.

[0097] In other embodiments of the present application, the control method further includes collecting the drying quality measured value; analyzing the drying quality measured value and the drying quality target value to obtain a running difference result; and adjusting the drying adjustment parameter according to the running difference result.

[0098] The drying adjustment parameters should refer to the factors that can be adjusted in the important spray drying process, mainly including drying gas flow temperature: affecting the moisture evaporation rate; atomization pressure: affecting the atomized particle size, and then affecting the drying efficiency and product particle size; concentration of the to-be-dried liquid: affecting the evaporation degree of moisture and the preservation of effective components in the drying process; drying time: affecting the thoroughness of powder drying and moisture control.

[0099] Adjustment of the drying gas flow temperature: if the error between the actual measured value and the target value of the drying quality is small and within the allowable range, the current temperature is maintained unchanged, and the monitoring is continued. If the moisture content is too low, it indicates that the drying temperature is too high, causing excessive loss of traditional Chinese medicine effective components, and the gas flow temperature needs to be reduced. The temperature reduction helps to slow down the evaporation of moisture and maintain the stability of the effective components. If the moisture content is too high, it indicates that the drying is insufficient, and the gas flow temperature needs to be increased to accelerate the evaporation of moisture and shorten the drying time.

[0100] Adjustment of the atomization pressure: if the actual measured particle size is too large and the target particle size is small, the atomization pressure is increased to make the to-be-dried liquid atomized more finely, and the drying efficiency is improved. If the particle size is too small and too fine, the atomization pressure is reduced to avoid excessive atomization leading to powder adhesion or poor flowability.

[0101] Adjustment of the concentration of traditional Chinese medicine extraction: if the moisture content is high, but the effective component remains good, it indicates that the concentration of the to-be-dried liquid is too high, causing the water to evaporate too slowly. The liquid concentration is reduced to make the water evaporate more easily, thereby accelerating the drying process. If the effective component content is low, the concentration is adjusted to better preserve the traditional Chinese medicine components during the drying process and avoid excessive dilution.

[0102] Adjustment of the drying time: if the moisture content exceeds the standard, it indicates that the drying time may not be long enough. The drying time is increased to further remove moisture and ensure complete drying. If the moisture content is too low and the effective component loss is large, it indicates that the drying time is too long, and the drying time is shortened to avoid excessive drying leading to the loss of effective components.

[0103] Further, the drying adjustment parameters are adjusted according to the operation difference result, specifically including: calculating the difference between the actual measured value and the target value of the drying quality; calculating the operation difference rate according to the following mode:

[0104]

[0105] Wherein, a represents the operation difference rate; M2 represents the actual measured value of the drying quality; M1 represents the target value of the drying quality;

[0106] If the operation difference rate is less than 5%, the current drying adjustment parameter is maintained; if the operation difference rate is greater than or equal to 5%, the operation difference rate and the drying quality measured value are input into the quality control model to adjust the drying adjustment parameter. If the operation difference rate is less than 5%, it means that the quality of the drying process has approached the target value, and excessive adjustment may cause instability of the production process; by setting that the adjustment of the drying adjustment parameter is only performed when the difference rate is greater than or equal to 5%, frequent small adjustments are avoided, thereby improving the production efficiency and reducing unnecessary interference and adjustment process. At the same time, fluctuations caused by excessive adjustment are avoided, and the quality stability is improved. The system can also be more focused on obvious quality abnormal conditions and ignore small fluctuations. This strategy helps the system focus on the link that really needs to be adjusted, and enhances the sensitivity and response speed to abnormal conditions.

[0107] The traditional Chinese medicine drying quality control method based on artificial intelligence disclosed in the embodiments of the present application overcomes the problems of unstable spray drying quality and high part batch failure rate in the prior art, stabilizes the drying quality of traditional Chinese medicine, and effectively improves the drying failure rate of the whole batch.

[0108] Embodiment two

[0109] The embodiments of the present application are based on the same inventive concept as the above-mentioned embodiments, and disclose a traditional Chinese medicine drying quality control system based on artificial intelligence, which is used to execute the above-mentioned traditional Chinese medicine drying quality control method based on artificial intelligence. The technical effects that can be achieved are as described in the above-mentioned embodiments, and will not be described here.

[0110] In summary, the traditional Chinese medicine drying quality control method and system based on artificial intelligence disclosed in the present application combine the temperature of the drying gas flow, the concentration of the to-be-dried liquid, the atomization pressure, and the drying time and other key parameters, use artificial intelligence technology to construct a drying quality prediction model, can multi-dimensionally predict the quality change in the drying process; through analysis of the drying quality difference result, each key parameter is adjusted, thereby ensuring that the quality control of the drying process is more accurate and efficient; at the same time, the deep learning model is used to realize dynamic optimization of the traditional Chinese medicine drying process and early warning of quality abnormalities, solve the problems of unstable spray drying quality and high part batch failure rate, stabilize the drying quality of traditional Chinese medicine, and effectively improve the drying failure rate of the whole batch.

[0111] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0112] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0113] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0114] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0115] Obviously, the above-described embodiments are only examples for clarity of description and are not limiting on the implementation. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the implementations are not required to be exhausted, and the obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. An artificial intelligence-based traditional Chinese medicine drying quality control method, characterized in that: The method comprises the following steps: a drying gas flow is introduced into a drying tower, and the temperature of the drying gas flow is measured; the concentration of a liquid to be dried is measured, the liquid to be dried is atomized, and the atomized liquid to be dried enters the drying tower, and the atomization pressure of the liquid to be dried is measured; the atomized liquid to be dried is dried by the drying gas flow in the drying tower, and the drying time is measured; a drying quality prediction model is established, the temperature of the drying gas flow, the concentration of the liquid to be dried, the atomization pressure, and the drying time are input into the drying quality prediction model, and a drying quality prediction value is obtained; wherein the indicators of the drying quality include moisture content, effective ingredient content, particle size, and bulk density; a drying quality target value is determined, and the drying quality prediction value and the drying quality target value are analyzed to obtain a drying quality difference result; the drying quality difference result is input into a quality control model to obtain a drying adjustment parameter, and the temperature of the drying gas flow, the atomization pressure, the concentration of the liquid to be dried, and the drying time are adjusted according to the drying adjustment parameter; the drying quality prediction model is established, including: obtaining historical data information of a drying process of a same model liquid as the liquid to be dried, and extracting sample data information related to drying quality; a sample data window is constructed, and the sample data information in the sample data window is configured with a weight; wherein the sample data window is divided into a plurality of time points according to time sequence, and the sample data information at each time point is configured with a weight according to formula (1); the sample data information is subjected to deep learning according to the configured weight, and the drying quality prediction model is obtained; ; Equation (1) i denotes the time point index; wi denotes the weight at the i-th time point; ti denotes the time difference between the current time point and the data collection time point; denotes the decay rate parameter, which influences the speed of the weight decay over time; the drying quality prediction model comprises an input layer, an LSTM layer, a full connection layer, and an output layer; the sample data information of the sample data window is converted into a feature matrix capable of convolution operation in the input layer; the feature matrix is subjected to feature extraction in the LSTM layer, and the LSTM layer can identify abnormal quality features of sample data; the feature extraction result is subjected to mapping and transformation in the full connection layer; the drying quality prediction value is output through the output layer; wherein the LSTM layer is provided with an activation function, and the feature extraction result is subjected to nonlinear transformation through the activation function; The control method further comprises: collecting a drying quality measured value; performing difference calculation on the drying quality measured value and the drying quality prediction value to obtain a prediction difference result; performing difference distribution analysis and error bias trend analysis on the prediction difference result to obtain a prediction difference analysis result; determining the number of neurons of the LSTM layer according to the activation degree of the LSTM layer; and adjusting the number of neurons of the LSTM layer according to the prediction difference analysis result.

2. The artificial intelligence-based traditional Chinese medicine drying quality control method according to claim 1, characterized in that: After configuring the weight according to formula (1), the configured weight is further normalized according to the following mode: ； wherein, represents the normalized weight of the i-th time point; n represents the number of time points; j represents the time point sequence number of division; represents the weight of the j-th time point; the sample data information is subjected to deep learning according to the weight after normalization to obtain the drying quality prediction model.

3. The artificial intelligence-based traditional Chinese medicine drying quality control method according to claim 1, characterized in that: the number of neurons of the LSTM layer is determined according to the activation degree of the LSTM layer, The method comprises the following steps: For the LSTM layer, collect the activation function output value of each neuron to obtain an activation value set; For the activation value set, at least one of the maximum value, minimum value, average value and standard value is calculated; According to the calculation result, the number of neurons of the LSTM layer is determined.

4. The artificial intelligence-based traditional Chinese medicine drying quality control method according to claim 1, characterized in that: The control method further comprises, Collecting a dry mass measured value; Analyzing the dry mass measured value and the dry mass target value to obtain a running difference result; According to the running difference result, the dry adjustment parameter is adjusted.

5. The artificial intelligence-based traditional Chinese medicine drying quality control method according to claim 4, characterized in that: According to the running difference result, the dry adjustment parameter is adjusted, comprising, Difference calculation is performed on the dry mass measured value and the dry mass target value; The running difference rate is calculated in the following manner: ; wherein, represents the running difference rate; M2 represents the measured value of the dry mass; M1 represents the target value of the dry mass; If the running difference rate is less than 5%, the current dry adjustment parameter is maintained; If the running difference rate is greater than or equal to 5%, the running difference rate and the dry mass measured value are input into the quality control model to adjust the dry adjustment parameter.

6. An artificial intelligence-based traditional Chinese medicine drying quality control system, characterized in that: An artificial intelligence-based traditional Chinese medicine drying quality control method for performing any one of claims 1-5.

Images