Formula optimization system of plukenetia volubilis tea based on Bayesian algorithm and one-dimensional CNN iteration

Abstract

The invention discloses a formula optimization system for plukenetia volubilis tea based on a Bayesian algorithm and one-dimensional CNN iteration, and the system comprises the steps: obtaining the existing historical formula data of the plukenetia volubilis tea, and importing the historical formula data into a formula database; determining a plurality of specified historical formula data with the highest similarity degree in a formula database according to the nutritional indexes of the tea products with different purposes; based on a multi-objective Bayesian optimization algorithm, training to obtain a probability agent model; according to the method, a recommended formula is output through a probability agent model, a formula which does not pass tea purposes is adjusted according to an evaluation result of the recommended formula, finally, a one-dimensional CNN is introduced for iteration, and the operation is continuously repeated, so that the optimization system for various formulas of plukenetia volubilis linneo tea can be obtained.

Description

TECHNICAL FIELD

[0001] The application relates to the field of formula optimization, in particular to a formula optimization system for star oil vine tea based on a Bayesian algorithm and one-dimensional CNN iteration. BACKGROUND

[0002] Star oil vine is a perennial woody vine plant, which is composed of roots, stems, leaves and fruits, and has good nutritional value. The star oil vine fruit is a new type of oil crop. It is rich in high protein, high oil and high vitamin E, and has the effects of food therapy and auxiliary treatment. This year's research found that the stems before lignification also have rich nutrients, with lower content of similar nutrients than star oil vine fruits. The leaves, especially the leaves picked at a certain period, have a unique fragrance, which can be comparable to many tea fragrances.

[0003] The above three are mixed in a certain proportion and fermented to obtain a red tea with both nutrition and tea aroma, becoming a new nutritional tea for Chinese consumers.

[0004] In order to meet the needs of consumers, the three components of star oil vine tea need to be optimized. The current formula is only an empirical result and has not been scientifically verified. In particular, the adjustment of various formulas according to the needs of different consumers is a systematic optimization process.

[0005] To solve the above problems, the application provides a formula optimization system for star oil vine tea based on a Bayesian algorithm and one-dimensional CNN iteration, which comprises: A formula optimization system for star oil vine tea based on a Bayesian algorithm and one-dimensional CNN system iteration comprises: An intelligent formula database is set up, and historical formula data of different uses of star oil vine tea are imported into the database. The historical formula data includes: formula weight values, nutritional content values and efficacy values of vine fruits, vine stems and vine leaves after drinking; According to different user needs, the nutritional content values of the to-be-optimized formula are determined, and an expected value range is set. Then, according to the nutritional content values of various tea products, historical formula data with high similarity is specified in the formula database; Based on the multi-objective Bayesian algorithm, a Bayesian probability agent model is trained according to the nutritional content values of various tea products, the expected range and the specified historical formula data; The probability agent model outputs a recommended formula table for different user needs; According to the above formula table and the intelligent database, a one-dimensional CNN iteration model is introduced for iterative optimization to obtain continuously improved recommended formulas of star oil vine tea suitable for different needs.

[0006] The intelligent database is provided with a self-iteration function of the one-dimensional CNN; And according to the tea nutrition values of different requirements, a plurality of specified historical formula data with the highest similarity degree are determined in the formula intelligent database, specifically comprising: Through the Euclidean formula, The Euclidean distance between each formula nutrition value and the tea product nutrition value is calculated, wherein d is the Euclidean distance, Respectively represent the performance value of the formula nutrition value and the tea product nutrition value in the i-th dimension, and n is the dimension number of the performance parameter; According to the Euclidean distance, the similarity degree between each formula nutrition value and the tea product nutrition value is obtained; According to the similarity degree, a plurality of specified historical formula data with the highest similarity degree are determined.

[0007] Based on the multi-objective Bayesian optimization algorithm, the tea product nutrition value, the expected range and the plurality of specified historical formula data are trained to obtain a probability agent model, specifically comprising: A multi-objective Bayesian optimization algorithm is determined, and the product nutrition value, the expected range and the number of recommended formulas are specified for the multi-objective Bayesian optimization algorithm; The plurality of specified historical formula data are taken as input, and the recommended formula is taken as output to construct a probability agent model; The point with the maximum value of the collection function is selected as the recommended formula, and the probability agent model is trained by multiple iterations.

[0008] Then the probability agent iteration model is subjected to one-dimensional CNN iteration model, specifically comprising: Composed of a convolution layer, an activation function layer, a pooling layer and a full connection layer.

[0009] A one-dimensional convolution operation with a step of 2 is performed twice with a convolution kernel with a width of 3 The above-mentioned CNN iteration model can be iterated for a certain time according to specific requirements, and the intelligent database can be continuously improved.

[0010] The formulas of the vine tea for different purposes evolve respectively, and can be slightly changed and adjusted according to user feedback. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 The flowchart of the formula design method of the star oil vine tea based on the Bayesian algorithm and one-dimensional CNN iteration optimization of the present application; DETAILED DESCRIPTION

[0012] The technical scheme provided by the embodiments of the present application is described in detail below with reference to the drawings.

[0013] As Figure 1 shown, the embodiment of the present application provides a system for optimizing the formula of star oil tea based on Bayesian algorithm and one-dimensional CNN, which comprises: T1, an intelligent formula database is set up, and the personalized historical formula data of star oil tea for different purposes is imported into the database, wherein the historical formula data includes the formula weight values of fruit, stem and leaf, such as 1:1:1, the nutritional content values, and the efficacy values after drinking; T2, the nutritional content values of the to-be-optimized formula are determined according to different user needs, and the expected value range is set, and then the historical formula data with higher similarity is specified in the formula database according to the nutritional content values of various tea products; T3, based on the multi-objective Bayesian algorithm, the Bayesian probability agent model is trained according to the nutritional content values of various tea products, the expected range and the specified historical formula data; T4, the intelligent database is provided with a one-dimensional CNN self-iteration function; and according to the nutritional values of tea for different needs, the specified historical formula data with the highest similarity is determined in the formula intelligent database, which specifically includes: Through the Euclidean formula, the Euclidean distance between the nutritional value of each formula and the nutritional value of tea products is calculated, T5, based on the multi-objective Bayesian optimization algorithm, the probability agent model is trained according to the nutritional value of tea products, the expected range and the specified historical formula data T6, then the probability agent iteration model is performed through the one-dimensional CNN iteration model, which specifically includes: which is composed of convolution layer, activation function layer, pooling layer and full connection layer.

[0014] The one-dimensional convolution operation with a step of 2 is performed twice in succession using a convolution kernel with a width of 3 In this embodiment, we demonstrate the optimization and iteration process as follows: The nutritional value of tea products is 1, and according to its expected range, it reaches 10. The nutritional value of the product is 2, and according to its expected range, it reaches 0. The Euclidean distance between (10, 0,...) and (1.3, 40.80, 70,...) and (9.18, 5.90, 60.) is calculated, and it is assumed that the two recommended formulas obtained after screening are formula 18 and formula 21, then the three formula data are used as the input data of the multi-objective Bayesian optimization algorithm.

[0015] The formula 18 and the formula 21 are input to the multi-target Bayesian optimization algorithm, a probability agent model is constructed to approximate the real formula performance distribution, a collection function is constructed, and multi-round iteration optimization is carried out, the screened two formula data (the formula 18 and the formula 21) are taken as the first round iteration data of the Bayesian optimization algorithm, a probability agent model is constructed by Gaussian process regression on the data, the collection function values of each point in the whole space are calculated, two points with large collection function values are selected and output as two recommended formulas to be verified.

[0016] . Then, iteration operation is carried out The convolution kernel of the one-dimensional CNN model is selected as 3, the convolution step number is 2, the optimized formula and the efficacy value are input, and then iteration can be carried out on the formula with different efficacies.

[0017] There are many embodiments of the present application, and the selection and optimization of the star oil tea formula through the Bayesian algorithm optimization and one CNN iteration are all within the scope of the present application.

Claims

1. A star oil vine tea formula optimization system based on Bayesian algorithm and one-dimensional CNN iteration, comprising: setting up an intelligent formula database, importing existing personalized historical formula data of star oil vine tea for different purposes into the database, the historical formula data including formula weight values, nutritional content values, and efficacy values after drinking of vine fruits, vine stems, and vine leaves; determining the nutritional content of the formula to be optimized according to different user needs and setting an expected value range, and then according to the nutritional content values of various tea products, specifying historical formula data with higher similarity in the formula database; training a Bayesian probability agent model based on a multi-objective Bayesian algorithm according to the nutritional content values of various tea products, the expected range, and the specified historical formula data; outputting a recommended formula table for different user needs through the probability agent model; introducing a one-dimensional CNN iteration model for iterative optimization according to the formula table and the intelligent database to obtain continuously evolving and improved recommended formulas of star oil vine tea suitable for different needs.

2. The method of claim 1, wherein, The intelligent database has a one-dimensional CNN self-iteration function; 3. The method of claim 2, wherein determining a number of specified historical formula data with the highest similarity in the formula intelligent database according to the nutritional values of tea for different needs, specifically including: The Euclidean distance between the nutritional value of each formula and the nutritional value of the tea product is calculated by the Euclidean formula, The Euclidean distance between the nutritional value of each formula and the nutritional value of the tea product is calculated by the Euclidean formula, respectively represent the performance value of the nutritional value of the formula and the nutritional value of the tea product in the i-th dimension, and n is the number of dimensions of the performance parameter. obtaining the similarity between the nutritional values of each formula and the nutritional values of tea products according to the Euclidean distance; determining a number of specified historical formula data with the highest similarity according to the similarity. 3.The method of claim 1, wherein based on a multi-objective Bayesian optimization algorithm, the probability agent model is trained according to the nutritional values of tea products, the expected range, and the number of specified historical formula data, specifically including: determining a multi-objective Bayesian optimization algorithm, and specifying the nutritional values of products, the expected range, and the number of recommended formulas for the multi-objective Bayesian optimization algorithm; constructing a probability agent model by taking the number of specified historical formula data as input and taking recommended formulas as output; selecting the point with the maximum value of the collection function as the recommended formula, and performing multi-round iterative training on the probability agent model.

4. The method of claim 3, wherein, The probability agent iteration model is performed through a one-dimensional CNN iteration model, specifically including: composed of a convolutional layer, an activation function layer, a pooling layer, and a fully connected layer.

5. And using a convolution kernel with a width of 3 to perform one-dimensional convolution operation with a step of 2 for two consecutive times.

6. The method of claim 5, wherein, The above CNN iteration model can input iteratively for a certain period of time according to specific needs, and continuously improve the intelligent database.

7. The method of claim 1, wherein, The formulas of star oil vine tea for different purposes evolve respectively, and can make subtle changes and adjustments to the formulas according to user feedback.

Images