A method for optimizing fluidized bed granulation process of herbal medicine based on correlation analysis

Abstract

The present invention relates to a method for optimizing a botanical fluidized bed granulation process based on association analysis, and belongs to the field of botanical fluidized bed granulation and technical optimization. The method comprises the following steps: S1, collecting process parameter time series data, particle yield rate and production time data from a historical database; S2, converting the process parameter time series data into a series of statistics according to predefined rules; S3, discretizing the statistics converted from the process parameter time series data, the particle yield rate and the production time data; S4, establishing association rules between discretized statistics, discretized particle yield rate and production time data; S5, screening strong association rules in association rules; S6 restoring the discretized statistics in the verified strong association rules one by one into continuous statistical control intervals. The present invention can improve the batch-to-batch consistency of particle yield rate and production time in the botanical fluidized bed granulation process.

Description

Technical Field

[0001] The invention belongs to the field of botanical medicine fluidized bed granulation and technology optimization, and relates to a botanical medicine fluidized bed granulation process optimization method based on association analysis. Background Art

[0002] Fluidized bed granulation integrates the steps of mixing, granulation and drying in the same equipment. First, the rising airflow is used to keep the raw material powder in a boiling state, and then the mist binder is sprayed into the powder to form granules, and then the moisture of the granules is dried to obtain dry granules. The use of fluidized bed granulation to prepare herbal medicine granules has the advantages of simple process, time saving and low labor intensity. The obtained granules have good uniformity, fluidity and compression molding properties.

[0003] Due to the weak scientific research foundation and insufficient understanding of the granulation process, the fluidized bed granulation of botanicals generally adopts an experience-based production model. On-site operators visually observe the fluidization state through the observation window and subjectively adjust the process parameters to ensure that the granule moisture and particle size indicators are within the set range. However, different operators have different operating habits, resulting in large differences in granule yield and production time between batches.

[0004] Establishing a mathematical model between process parameters and product quality, and then optimizing the control range of process parameters based on the model can achieve lean improvement of fluidized bed granulation process more scientifically. The establishment of traditional mathematical models such as multivariate linear regression models requires three types of parameters: critical material attributes (CMA), critical process parameters (CPP) and critical quality attributes (CQA), which correspond to the input herbal raw material quality attributes, process operation parameters and output particle quality attributes respectively. However, the chemical composition of herbal raw materials is complex and cannot be fully characterized, resulting in a large number of missing CMA attribute items.

[0005] Association analysis methods represented by the Apriori algorithm look for frequently occurring data patterns in data sets. In the case of incomplete fluidized bed granulation data, the association analysis method does not need to obtain CMA attribute items, but directly establishes the association relationship between CPP and CQA, and then adjusts the CPP range to reduce the batch-to-batch difference of CQA. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis. The technical problem to be solved by the present invention is to provide a method for optimizing the fluidized bed granulation process of botanical medicine, which can deeply understand the data rules implicit in the botanical medicine granulation process, improve the operational standardization of on-site operators, and improve the batch consistency of particle yield and production time in the fluidized bed granulation process of botanical medicine, which is conducive to the lean improvement of botanical medicine production workshops.

[0007] The present invention provides a method for optimizing a botanical medicine fluidized bed granulation process based on association analysis, comprising the following steps:

[0008] Step S1: collecting process parameter time series data, particle yield and production time data from the historical database;

[0009] Step S2: converting the process parameter time series data into a series of statistics according to predefined rules;

[0010] Step S3: discretizing the statistics, particle yield rate and production duration data converted from the process parameter time series data to obtain discretized statistics, discretized particle yield rate and production duration data respectively;

[0011] Step S4: establishing association rules between discretized statistics, discretized particle yield and production time data;

[0012] Step S5: Filter strong association rules among the association rules according to preset rules;

[0013] Step S6: Verify whether the selected strong association rules have generalization ability, extract the discretized statistics in the verified strong association rules, and restore them one by one to continuous statistical control intervals.

[0014] Furthermore, in step S1, the process parameter time series data includes inlet air temperature, air volume and spray gun pressure; the process parameter time series data and particle yield and production time data are aligned in the batch direction to generate batch records; the batch records are divided into a data mining set and a test set; the data mining set is used to mine the implicit association rules in the data, and the test set is used to test whether the screened association rules are generalizable.

[0015] Further, in the step S2, the extraction of the time series curve statistics is performed in the process parameter time series curve generated by the process parameter time series data; the extraction method is as follows: the granulation time, drying time and cooling time of the granulation, drying and cooling stages of a botanical fluidized bed granulation process are extracted as common variables; two temperature extreme point statistics are extracted from the inlet temperature curve, which are used as the maximum inlet temperatures of the granulation stage and the drying stage respectively; the "air volume front section slope", "air volume front section intercept" and "air volume rear section mean" statistics are extracted from the air volume curve, which are used as the air volume increase speed in the granulation stage, the initial value of the air volume at time zero and the average value of the air volume in the drying / cooling stage respectively; the "spray gun pressure standard deviation" statistic is extracted from the spray gun pressure curve as the standard deviation of the spray gun pressure in the granulation stage; and whether the operator has taken certain adjustment measures in production is described in advance as a Boolean statistic, and these Boolean statistics are called technical adjustment variables.

[0016] Furthermore, in step S3, the fuzzy C-means clustering method is used to adaptively discretize the time series curve statistics; the time series curve statistics are discretized into multiple letter variables, and the types of the letter variables are determined according to the number of sub-intervals into which the time series curve statistic interval is divided; for example, a certain statistic is divided into three sub-intervals of high, medium and low, and the letter variables can be represented as H, M and L.

[0017] Furthermore, in step S3, the particle yield rate and production time indicators are discretized according to the set process control upper and lower limits; the particle yield rate and production time indicators are discretized into multiple letter variables, which are indicator data H exceeding the upper control limit, indicator data M within the control range, and indicator data L below the lower control limit.

[0018] Further, in step S4, the Apriori algorithm is used to establish association rules between the discretized statistics and the discretized particle yield and production time indicators; the Apriori algorithm starts from a single item set that meets the minimum support and confidence, gradually increases the number of items to calculate the item set that meets the threshold, and stops calculating when the threshold is not met, and outputs the association rules and their corresponding support and confidence;

[0019] The Apriori algorithm is used to mine implicit association rules that meet specific requirements in a data set. The Apriori algorithm includes item sets, association rules, support, and confidence. The data set D = {i 1 ,i 2 ,…,i N The variables in} are called items, and a set of multiple items is called an item set; association rules are expressions of the form A→B, where A and B are both item sets, and A∪B=φ, A is the rule antecedent, and B is the rule consequent, reflecting the rule that when A appears, B appears together; association rules are evaluated using support and confidence;

[0020] Support(A,B) indicates the proportion of association rules that contain both rule antecedent A and rule consequent B in the entire database, reflecting the frequency of simultaneous occurrence of A and B:

[0021] Support(A,B)=p(A∪B)

[0022] Confidence(A,B) represents the ratio of the number of association rules that contain both rule antecedent A and rule consequent B to the number of association rules that contain A, reflecting the conditional probability of B appearing in the rule that contains A:

[0023]

[0024] In addition to the above evaluation indicators, the Apriori association rule can also add several interest indicators as needed.

[0025] Furthermore, in step S5, an interest index is introduced into the association rule, and the interest index includes the lifting degree, the particle yield rate P Pk and production time P pk ; Lift(A,B), which indicates the change in the probability of the rule consequence B when the rule antecedent A appears:

[0026]

[0027] Further, in step S5, the particle yield P pk and production time P pk , where P pk is the process performance index, P pk It also describes the degree of dispersion of data distribution and the degree of deviation of the mean of the measured value from the tolerance center. It is used to describe the process capability of the unit process. The calculation formula is as follows:

[0028]

[0029] Among them, T U and T L are the upper and lower control limits, s is the overall standard deviation of the sample space, is the measured mean, Indicates taking the smaller of the two values. pk The evaluation criteria are as follows:

[0030] Table 1P pk Evaluation criteria (σ is a measure of process capability in Lean Six Sigma)

[0031] <![CDATA[P pk Scope]]> level evaluate <![CDATA[1.67≤P pk ]]> 5σ Process capacity is too sufficient <![CDATA[1.33≤P pk <1.67]]> 4σ Adequate process capacity <![CDATA[1.00≤P pk <1.33]]> 3σ Process capability is acceptable <![CDATA[0.67≤P pk <1.00]]> 2σ Insufficient process capability <![CDATA[P pk <0.67]]> σ Serious lack of process capability

[0032] Furthermore, in step S5, the Pareto multi-objective optimization method is used to further screen the strong association rules in the association rules. The Pareto multi-objective optimization method is used to improve the degree and particle yield rate P pk and production time P pk The optimality is the goal. The multi-objective optimization problem consists of multiple objective functions and multiple related inequalities and equality constraints. The formula is as follows:

[0033]

[0034] Where x = {x 1 ,x 2 ,…,x p} T is the design variable in space Ω, f k (x) represents the objective function, g i (x) is a constraint function. The purpose of multi-objective optimization is to find the decision variable x that minimizes the components of the objective function.

[0035] Assume that the feasible solution set X f is a solution set consisting of design vectors x that can satisfy all constraint functions, and the feasible solution set X f For any solution x, there does not exist x′∈X f , so that F(x′)=(f 1 (x′),f 2 (x′),...,f k (x′)) is better than F(x)=(f 1 (x),f 2 (x),...,f k (x)), then x is called X f The Pareto optimal solution on the Pareto optimal solution set P * is the set of all Pareto optimal solutions, expressed as:

[0036] P * ={x∈X f |There is no x′∈X f , so that f(x′) is better than f(x)}

[0037] Furthermore, in step S6, if the lift of an association rule in both the data mining set and the test set is greater than 1, and the particle yield rate P pk and production time P pk If the results are improved compared with the original ones, the selected association rules are considered to be effective.

[0038] Compared with the prior art, the present invention has the following beneficial effects: the fluidized bed granulation of botanical medicine generally adopts an experience-based production model. Due to the lack of a standard operating procedure, the particle yield and production time of the granulation process vary greatly from batch to batch. Traditional mathematical models such as multivariate linear regression models require complete data, that is, CMA, CPP and CQA are required to be complete, but botanical raw materials are often unable to be fully characterized, and a large number of CMA attribute items are missing;

[0039] Compared with the traditional mathematical model, the association analysis algorithm used in the present invention does not need to obtain CMA attribute items, and can directly establish the association relationship between CPP and CQA, which helps to deeply understand the data rules hidden in the granulation process of herbal medicines;

[0040] The statistical control interval of process parameters obtained by using the association analysis algorithm is conducive to improving the operational standardization of on-site operators, improving the batch consistency of particle yield and production time in the fluidized bed granulation process of herbal medicine, and facilitating lean improvement of herbal medicine production workshops. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a method for optimizing a botanical medicine fluidized bed granulation process based on correlation analysis of the present invention;

[0042] Figure 2 (a) is a time series curve of the inlet air temperature in the fluidized bed granulation process of granules; Figure 2 (b) is the time series curve of the air volume in the fluidized bed granulation process of granules; Figure 2 (c) is a time series curve of the spray gun pressure in the fluidized bed granulation process of granules; the numbers 1 and 2 respectively indicate that the bed collapse prevention 1 and bed collapse prevention 2 operations were performed;

[0043] Figure 3 (a) and (b) are the process capability status diagrams of granule yield and production time in the data mining set respectively; Figure 3 (c) and (d) are the process capability status diagrams of the granule yield and production time of the granules on the test set respectively;

[0044] Figure 4 (a) is the process capability diagram of the particle yield of granules under the optimal production conditions; Figure 4 (b) in the figure is the process capability diagram of the production time of granules under the optimal production conditions. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0048] In order to better understand the purpose, structure and function of the present invention, the following is a detailed description of a method for optimizing a botanical fluidized bed granulation process based on correlation analysis in conjunction with the accompanying drawings.

[0049] Embodiment 1:

[0050] like Figure 1 As shown, the present invention provides a method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis, comprising the following steps:

[0051] Step S1: collecting process parameter time series data, particle yield and production time data from the historical database;

[0052] Step S2: converting the process parameter time series data into a series of statistics according to predefined rules;

[0053] Step S3: discretizing the statistics, particle yield rate and production duration data converted from the process parameter time series data to obtain discretized statistics, discretized particle yield rate and production duration data respectively;

[0054] Step S4: establishing association rules between discretized statistics, discretized particle yield and production time data;

[0055] Step S5: Filter strong association rules among the association rules according to preset rules;

[0056] Step S6: Verify whether the selected strong association rules have generalization ability, extract the discretized statistics in the verified strong association rules, and restore them one by one to continuous statistical control intervals.

[0057] Embodiment 2:

[0058] like Figure 1 As shown, the present invention provides a method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis, comprising the following steps:

[0059] Step S1: collecting process parameter time series data, particle yield and production time data from the historical database;

[0060] Step S2: converting the process parameter time series data into a series of statistics according to predefined rules;

[0061] Step S3: discretizing the statistics, particle yield rate and production duration data converted from the process parameter time series data to obtain discretized statistics, discretized particle yield rate and production duration data respectively;

[0062] Step S4: establishing association rules between discretized statistics, discretized particle yield and production time data;

[0063] Step S5: Filter strong association rules among the association rules according to preset rules;

[0064] Step S6: Verify whether the selected strong association rules have generalization ability, extract the discretized statistics in the verified strong association rules, and restore them one by one to continuous statistical control intervals.

[0065] The difference between this embodiment and embodiment 1 is that:

[0066] In step S1, the process parameter time series data includes air inlet temperature, air volume and spray gun pressure; the process parameter time series data and particle yield rate and production time data are aligned in the batch direction to generate batch records; the batch records are divided into a data mining set and a test set; the data mining set is used to mine the implicit association rules in the data, and the test set is used to test whether the selected association rules are generalizable;

[0067] In step S2, manually defined rules are used to extract statistics of the time series curve from the process parameter time series curve generated by the process parameter time series data; the statistical extraction method is as follows: the granulation time, drying time and cooling time of the granulation, drying and cooling stages of a typical botanical fluidized bed granulation process are collectively referred to as common variables; two statistics, namely "temperature extreme point 1" and "temperature extreme point 2", are extracted from the inlet temperature curve, which are the highest inlet temperatures in the granulation stage and the drying stage respectively; three statistics, namely "air volume front section slope", "air volume front section intercept" and "air volume rear section mean", are extracted from the air volume curve, which are the air volume increase speed in the granulation stage, the initial value of the air volume at time zero and the average value of the air volume in the drying / cooling stage respectively; the "spray gun pressure standard deviation" statistic is extracted from the spray gun pressure curve, which is the standard deviation of the spray gun pressure in the granulation stage; a number of Boolean statistics are defined to describe whether the operator has taken certain adjustment measures in production, and these Boolean statistics are called technical adjustment variables;

[0068] In step S3, the statistics of the time series curve are adaptively discretized using the fuzzy C-means clustering method; the fuzzy C-means clustering method is a soft clustering method, which uses membership to represent the distance between the data point and all cluster centers, and determines the classification of the data point according to the size of the membership. The key parameters include: the number of categories, the cluster centers and the data point membership matrix; under the premise of a given number of categories, the fuzzy C-means clustering method obtains the membership matrix and cluster centers of the data points by minimizing the objective function;

[0069] The objective function is as follows:

[0070]

[0071] Among them, m is the fuzzy index. The larger the m, the greater the fuzziness of the clustering. n is the number of samples, C is the number of cluster centers, and c is the number of cluster centers. j represents the jth cluster center, and its dimension is the same as the feature dimension; x i represents the i-th sample, u ij Represents sample x i For cluster center c j The membership degree of ;||*|| represents the Euclidean norm;

[0072] The membership calculation formula is as follows:

[0073]

[0074] Single sample x i The sum of the membership of all categories is 1, and the category with the largest membership belongs to that category; the cluster center calculation formula is as follows:

[0075]

[0076] The fuzzy C-means clustering algorithm continuously iterates u ij and c j , until J m Reach local optimum or global optimum;

[0077] max ij {|u ij (t+1) -u ij (t) |}<ε

[0078] Among them, t is the number of iterations, ε is the error threshold, and max ij {|u ij (t+1) -u ij (t) |} means when the number of iteration steps is t, i*j |u ij (t+1) -u ij (t) |The maximum value among the values.

[0079] In step S3, the statistics of the time series curve are discretized into a number of letter variables. The type of letter variables depends on how many sub-intervals the statistical interval is divided into. For example, if a certain statistic is divided into three sub-intervals of high, medium and low, the letter variables can be represented as H, M and L.

[0080] In step S3, the particle yield rate and production time indicators are discretized according to the set process control upper and lower limits; the particle yield rate and production time indicators are discretized into a number of letter variables, and the indicator data exceeding the upper control limit is marked as H, the indicator data within the control range is marked as M, and the indicator data below the lower control limit is marked as L.

[0081] Embodiment 3:

[0082] like Figure 1 As shown, the present invention provides a method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis, comprising the following steps:

[0083] Step S1: collecting process parameter time series data, particle yield and production time data from the historical database;

[0084] Step S2: converting the process parameter time series data into a series of statistics according to predefined rules;

[0085] Step S3: discretizing the statistics, particle yield rate and production duration data converted from the process parameter time series data to obtain discretized statistics, discretized particle yield rate and production duration data respectively;

[0086] Step S4: establishing association rules between discretized statistics, discretized particle yield and production time data;

[0087] Step S5: Filter strong association rules among the association rules according to preset rules;

[0088] Step S6: Verify whether the selected strong association rules have generalization ability, extract the discretized statistics in the verified strong association rules, and restore them one by one to continuous statistical control intervals.

[0089] The difference between this embodiment and the above embodiments is that:

[0090] In step S4, the Apriori algorithm is used to establish association rules between the discretized statistics and the discretized particle yield and production time indicators; the Apriori algorithm starts from a single item set that meets the minimum support and confidence, gradually increases the number of items to calculate the item set that meets the threshold, and stops calculating when the threshold is not met, and outputs the association rules and their corresponding support and confidence;

[0091] The Apriori algorithm is used to mine implicit association rules that meet specific requirements in a data set. The Apriori algorithm includes four basic concepts: item set, association rule, support, and confidence. The data set D = {i 1 ,i 2 ,…,i N The variables in} are called items, and a set containing multiple items is called an item set; association rules are expressions of the form A→B, where A and B are both item sets, and A∪B=φ, A is the rule antecedent, and B is the rule consequent, reflecting the rule that B appears together when A appears; association rules are evaluated using support and confidence; support Support(A,B) represents the proportion of association rules that contain both the rule antecedent A and the rule consequent B in the entire database, reflecting the frequency of simultaneous occurrence of A and B:

[0092] Support(A,B)=p(A∪B)

[0093] Confidence(A,B) represents the ratio of the number of association rules that contain both rule antecedent A and rule consequent B to the number of association rules that contain A, reflecting the conditional probability of B appearing in the rule that contains A:

[0094]

[0095] In addition to the above evaluation indicators, the Apriori association rule can also add several interest indicators as needed.

[0096] In step S5, the lift Lift(A,B) represents the change in the probability of the rule consequence B when the rule antecedent A appears:

[0097]

[0098] In step S5, the interest index particle collection rate P pk and production time P pk , where P pk is the process performance index, P pk It also describes the degree of dispersion of data distribution and the degree of deviation of the mean of the measured value from the tolerance center. It is used to describe the process capability of the unit process. The calculation formula is as follows:

[0099]

[0100] Among them, T U and T L are the upper and lower control limits, s is the overall standard deviation of the sample space, is the measured mean, Indicates taking the smaller value of two values; P pk The evaluation criteria are as follows:

[0101] Table 1P pk Evaluation criteria (σ is a measure of process capability in Lean Six Sigma)

[0102] <![CDATA[P pk Scope]]> level evaluate <![CDATA[1.67≤P pk ]]> 5σ Excessive process capacity <![CDATA[1.33≤P pk <1.67]]> 4σ Adequate process capacity <![CDATA[1.00≤P pk <1.33]]> 3σ Process capability is acceptable <![CDATA[0.67≤P pk <1.00]]> 2σ Insufficient process capability <![CDATA[P pk <0.67]]> σ Serious lack of process capability

[0103] In step S5, the Pareto multi-objective optimization method is used to further screen the strong association rules with practical significance in the association rules. The Pareto multi-objective optimization method is used to improve the degree and particle yield rate P pk and production time P pk The optimality is the goal. The multi-objective optimization problem consists of multiple objective functions and multiple related inequalities and equality constraints. The formula is as follows:

[0104]

[0105] Where x = {x 1 ,x 2 ,…,x p} T is the design variable in space Ω, f k (x) represents the objective function, g i (x) is a constraint function. The purpose of multi-objective optimization is to find the decision variable x that minimizes the components of the objective function.

[0106] Assume that the feasible solution set X f is a solution set consisting of design vectors x that can satisfy all constraint functions, and the feasible solution set X fFor any solution x, there does not exist x′∈X f , so that F(x′)=(f 1 (x′),f 2 (x′),...,f k (x′)) is better than F(x)=(f 1 (x),f 2 (x),...,f k (x)), then x is called X f The Pareto optimal solution on the Pareto optimal solution set P * is the set of all Pareto optimal solutions, expressed as:

[0107] P * ={x∈X f |There is no x′∈X f , so that f(x′) is better than f(x)}

[0108] In step S6, if the lift of an association rule in both the data mining set and the test set is greater than 1, and the particle yield rate P pk and production time P pk If the results are improved compared with the original results, the selected association rules are considered effective.

[0109] Embodiment 4:

[0110] like Figures 2 to 4 As shown, in this embodiment, the implementation process of the present invention is described in detail in combination with specific examples. The time series data of air inlet temperature, air volume and spray gun pressure, the particle yield rate after batch production and the production time data in the embodiment are provided by a pharmaceutical company. All algorithms are compiled using Python (v3.8.5; Python Software Foundation, 2020).

[0111] 1. Data selection

[0112] The time series data of granule inlet air temperature, air volume and spray gun pressure of 4 fluidized beds of the same model from February to August 2020 were exported from the Supervisory Control And Data Acquisition (SCADA) database. The granule yield and production time after batch production of 4 fluidized beds of the same model from February to August 2020 were exported from the Manufacturing Execution System (MES). After removing the data records containing missing values ​​and outliers, the process parameter time series data and the granule yield and production time data were aligned in the batch direction, and a total of 217 complete batch records were obtained. Among them, 140 records from February to May were used as the data mining set to extract association rules, and 77 records from June to August were used as the test set to verify the rationality of the association rules.

[0113] 2. Statistics extraction

[0114] A series of statistics are extracted from the timing curve of process parameters to facilitate subsequent correlation analysis. The granulation time, drying time and cooling time of the fluidized granulation process are defined as common variables. A total of 6 statistics are obtained, including "temperature extreme point 1", "temperature extreme point 2", "air volume front section slope", "air volume front section intercept", "air volume rear section mean", and "spray gun pressure standard deviation". In addition, operators will make some technical adjustments to prevent the occurrence of bed collapse. The so-called bed collapse refers to the collapse of the bed due to poor fluidization. "Bed collapse prevention 1" means that technical adjustments are taken during the granulation stage to prevent bed collapse, and "bed collapse prevention 2" means that technical adjustments are taken during the drying or cooling stage to prevent bed collapse. The timing curves of inlet temperature, air volume and spray gun pressure are shown in the figure. Figure 2 As shown, the statistics extracted from the timing curves are shown in Table 2.

[0115] Table 2 Statistics of granule preparation

[0116] project Statistics Data Types Process indicators Particle yield, production time Continuous Values Public variables Granulation time, drying time, cooling time Continuous Values Technology Adjustment Variables Bed collapse prevention 1, bed collapse prevention 2 Boolean Inlet air temperature Temperature extreme point 1, Temperature extreme point 2 Continuous Values Air volume Slope of the front section of air volume, intercept of the front section of air volume, mean of the rear section of air volume Continuous Values Gun pressure Gun pressure standard deviation Continuous Values

[0117] 3. Data discretization

[0118] According to the set control standards, the process indicators (granule yield, production time) are discretized into three levels of high, medium and low, which are [H0, M0, L0] and [H1, M1, L1] respectively. The common variables (granulation time, drying time, cooling time) are clustered into three levels of high, medium and low using the fuzzy C-means clustering method, which are [H2, M2, L2] to [H4, M4, L4] respectively. Depending on whether it occurs, the technical adjustment variables (bed collapse prevention 1, bed collapse prevention 2) are set to yes and no, which are [Y5, N5] and [Y6, N6] respectively. The statistics (temperature extreme point 1, temperature extreme point 2, air volume front slope, air volume front intercept, air volume back mean, spray gun pressure standard deviation) are clustered into three levels of high, medium and low using the fuzzy C-means clustering method, which are [H7, M7, L7] to [H12, M12, L12] respectively.

[0119] 4 Association rule mining

[0120] After the data was discretized, the Apriori association analysis algorithm was used to mine association rules on the 140 records in the data mining set, with the minimum support set to 0.1 and the minimum confidence set to 0.7. The rule consequence of the association rule was set to M0∪M1, so that the particle yield and production time were within the control range, and 947 valid association rules were obtained.

[0121] Top 5 association rules screening

[0122] Add the degree of improvement and particle yield rate P to the effective association rules pk and production time P pk Three interest indexes are used to select the association rules that meet the Pareto optimality of the three indexes at the same time. In order to ensure the basic process capability of the fluidized bed, only the particle yield rate P is retained. pk and production time P pk All the solutions are greater than 0.67, and finally 15 Pareto optimal association rules are obtained. The results are shown in Table 3.

[0123] Table 3 Pareto optimal association rules on granule data mining set

[0124]

[0125] The Pareto optimal association rules in Table 3 have a lift range of 1.210 to 1.296, and the particle yield rate P pk The range is 0.677 to 0.839, and the production time P pk The range is 0.878 to 1.252, where the particle yield P pk The most difficult to meet. Select the particle yield P pkAfter analyzing the top 6 association rules, we found that these rules have a relationship of inclusion and being included. In fact, these association rules are subsets of the association rules M2∪N5∪N6∪H7∪M9→M0∪M1 or M2∪N5∪H7∪M9∪H11→M0∪M1 with a rule antecedent of 5. Their lift is 1.296, which means that the yield rate and production time of batches that meet the rules can be 100% within the control range. It is worth noting that the common superset M2∪N5∪N6∪H7∪M9→M0∪M1 and M2∪N5∪H7∪M9∪H11→M0∪M1 does not appear in Table 3 because its support is 0.086, which is lower than the minimum support. However, this rule performs better than its two subsets in other indicators, with a confidence of 1.000, a lift of 1.296, and a yield rate of P. pk is 1.031, production time P pk It is 1.097.

[0126] 6. Verification, interpretation and application effect of strong association rules

[0127] 1) Verification of strong association rules. After screening out strong association rules, their generalization performance is further verified on the test set. The improvement of the strong association rule M2∪N5∪N6∪H7∪M9∪H11→M0∪M1 on the test set is 1.328, and the particle yield rate P pk is 0.872, production time P pk Compared with the unoptimized test set results, the proportion of batches with both particle yield and production time within the control range increased from 75.3% to 100%, and the particle yield P pk and production time P pk The improvement was 0.589 and 0.273 respectively. After further verification of other strong association rules, it was found that M2∪N5∪N6∪H7∪M9→M0∪M1 and M2∪N5∪N6∪M9→M0∪M1 also have strong generalization ability, but they are subsets of the association rule M2∪N5∪N6∪H7∪M9∪H11→M0∪M1, and compared with the parent set, they failed to further improve the process capability of the fluidized bed. Therefore, the parent set M2∪N5∪N6∪H7∪M9∪H11→M0∪M1 is still the final selected association rule.

[0128] 2) Strong association rule interpretation. The verified strong association rules are reduced to statistical control intervals. Specifically, when producing granules, the granulation time should be controlled between 5420 and 6700 seconds, the temperature extreme point 1 should be controlled at 120°C or above, and the slope of the front section of the air volume should be controlled between 1842 and 2739 m / s. 3 / h 2 The average air volume in the rear section should be controlled between 3931 and 5385 m3 / h. In addition, both bed collapse prevention 1 and bed collapse prevention 2 are necessary, which means that it is necessary to increase the water loss rate of drug particles during granulation to reduce the risk of bed collapse, including but not limited to using a more efficient dehumidifier to reduce the humidity of the incoming air, premixing dextrin and extract to make the drug particles in the fluidization process more loose and porous, and using a spray gun with more uniform spray.

[0129] 3) The effect of strong association rule application. The process capabilities of particle yield and production time on the data mining set and test set are as follows: Figure 3 As shown. Whether in the data mining set or the test set, the particle yield rate P pk and production time P pk All of them are less than 0.67, indicating that the process capacity of the fluidized bed granule production is seriously insufficient. The process capacity of the granule yield and production time is as follows: Figure 4 After optimization, the particle yield rate P in the data mining set is pk is 1.031, production time P pk is 1.097, and the particle yield rate P on the test set is pk is 0.872, production time P pk The results show that the process capability of particle yield and production time is greatly improved by using optimized production conditions, and the improved process capability is close to the 3σ level.

[0130] The above-mentioned embodiments only express several implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the attached claims.

Claims

1. A method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis, characterized in that: The following steps are involved: Step S1: collecting process parameter time series data, particle yield and production time data from the historical database; Step S2: converting the process parameter time series data into a series of statistics according to predefined rules; Step S3: discretizing the statistics, particle yield rate and production duration data converted from the process parameter time series data to obtain discretized statistics, discretized particle yield rate and production duration data respectively; Step S4: establishing association rules between discretized statistics, discretized particle yield and production time data; Step S5: Filter strong association rules among the association rules according to preset rules; Step S6: verify whether the selected strong association rules have generalization ability, extract the discretized statistics in the verified strong association rules, and restore them one by one to continuous statistical control intervals; Wherein, in the step S2, the extraction of the time series curve statistics is performed in the process parameter time series curve generated by the process parameter time series data; the extraction method is as follows: extracting the granulation time, drying time and cooling time of the granulation, drying and cooling stages of a botanical fluidized bed granulation process as common variables; extracting two temperature extreme point statistics from the inlet temperature curve, respectively as the maximum inlet temperature of the granulation stage and the drying stage; extracting the "air volume front section slope", "air volume front section intercept" and "air volume rear section mean" statistics from the air volume curve, respectively as the air volume increase speed in the granulation stage, the initial value of the air volume at time zero and the average value of the air volume in the drying / cooling stage; extracting the "spray gun pressure standard deviation" statistic from the spray gun pressure curve as the standard deviation of the spray gun pressure in the granulation stage; extracting the description of whether the operator has taken certain adjustment measures in production as a Boolean statistic.

2. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 1, characterized in that: In step S1, the process parameter time series data includes inlet air temperature, air volume and spray gun pressure; the process parameter time series data and particle yield and production time data are aligned in the batch direction to generate batch records; the batch records are divided into a data mining set and a test set; the data mining set is used to mine the implicit association rules in the data, and the test set is used to test whether the screened association rules are generalizable.

3. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 2, characterized in that: In step S3, the fuzzy C-means clustering method is used to adaptively discretize the time series curve statistics; the time series curve statistics are discretized into a plurality of letter variables, and the types of the letter variables are determined according to the number of sub-intervals into which the time series curve statistics interval is divided.

4. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 3, characterized in that: In step S3, the particle yield rate and production time indicators are discretized according to the set process control upper and lower limits; the particle yield rate and production time indicators are discretized into multiple letter variables, which are indicator data exceeding the upper control limit, indicator data within the control range, and indicator data below the lower control limit.

5. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 2, characterized in that: In step S4, the Apriori algorithm is used to establish association rules between the discretized statistics and the discretized particle yield and production time indicators; the Apriori algorithm starts from a single item set that meets the minimum support and confidence, gradually increases the number of items to calculate the item set that meets the threshold, and stops calculating when the threshold is not met, and outputs the association rules and their corresponding support and confidence; The Apriori algorithm includes item sets, association rules, support and confidence. Association rules are evaluated using support and confidence. Support(A, B) indicates the proportion of association rules that contain both rule antecedent A and rule consequent B in the entire database, reflecting the frequency of simultaneous occurrence of A and B: Support(A, B) = p(A∪B) Confidence(A, B) represents the ratio of the number of association rules that contain both rule antecedent A and rule consequent B to the number of association rules that contain A, reflecting the conditional probability of B appearing in the rule that contains A:

6. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 5, characterized in that: In step S5, the interest index is introduced into the association rule, and the interest index includes the lifting degree, the particle yield rate P pk and production time P pk ; Lift(A, B) indicates the change in the probability of the rule consequence B when the rule antecedent A appears; 7. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 6, characterized in that: In step S5, the particle yield P pk and production time P pk , where P pk is the process performance index, P pk It also describes the degree of dispersion of data distribution and the degree of deviation of the mean of the measured value from the tolerance center. It is used to describe the process capability of the unit process. The calculation formula is as follows: Among them, T U and T L are the upper and lower control limits, s is the overall standard deviation of the sample space, is the measured mean, Indicates taking the smaller of two values.

8. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 6, characterized in that: In step S5, the Pareto multi-objective optimization method is used to further screen the strong association rules in the association rules. The Pareto multi-objective optimization method is used to improve the degree and particle yield rate P pk and production time P pk The optimality is the goal. The multi-objective optimization problem consists of multiple objective functions and multiple related inequalities and equality constraints. The formula is as follows: Where x = {x1, x2, ..., x p } T is the design variable in space Ω, f k (x) represents the objective function, g i (x) is a constraint function. The purpose of multi-objective optimization is to find the decision variable x that minimizes each component of the objective function. Assume that the feasible solution set X f is a solution set consisting of design vectors x that can satisfy all constraint functions, and the feasible solution set X f For any solution x, there does not exist x′∈X f , so that F(x′)=(f1(x′),f2(x′),...,f k (x′)) is better than F(x)=(f1(x),f2(x),...,f k (x)), then x is called X f The Pareto optimal solution on the Pareto optimal solution set P* is the set of all Pareto optimal solutions, expressed as: P * ={x∈X f |There is no x′∈X f , so that f(x′) is better than f(x)).

9. The method for optimizing the fluidized bed granulation process of botanical medicine based on association analysis according to claim 6, characterized in that: In step S6, if the lifting degree of an association rule in both the data mining set and the test set is greater than 1, and the particle yield rate P pk and production time P pk If the results are improved compared with the original ones, the selected association rules are considered to be effective.

Images