Tea fermentation pile process control method based on gradient boosting tree

Abstract

The invention relates to the technical field of tea pile fermentation technology, and in particular to a tea fermentation pile fermentation technology control method based on a gradient boosting tree, comprising the following steps: S1) collecting data: collecting a batch of production data of known high-quality black tea in the pile fermentation process for processing; and dividing the database into a training set and a test set; through the predictive ability of a machine learning model, the stability of the fermentation degree and the consistency of the quality can be effectively maintained, thereby significantly improving the stability of the fermentation degree and the consistency of the quality of the black tea under the changes in the pile fermentation process parameters caused by many factors in actual production.

Description

Technical Field

[0001] The invention relates to the technical field of tea pile fermentation technology, and in particular to a tea fermentation pile fermentation technology control method based on a gradient boosting tree. Background Art

[0002] The pile-fermentation process is a key step in the fermentation process of dark tea, including Pu'er tea, and is a key process for forming the color and aroma of tea. In various tea planting and production areas, this process is still completely completed by manual labor. Generally, the rolled tea leaves are piled directly on the ground or in a clean fermentation tank. Experienced production masters can distinguish and control the degree and effect of fermentation based on the temperature and humidity of the tea pile, the color of the tea leaves, and the smell produced by fermentation.

[0003] This process is inevitably affected by subjective factors, and is prone to uneven fermentation, insufficient fermentation, or excessive fermentation. In addition, the fermentation process is a process of chemical transformation. Many key factors within the tea pile, such as relative humidity, temperature, time, etc., are difficult to accurately control based on manual experience, so it is also difficult to ensure the consistency of the quality of black tea. On the other hand, the time span of the pile fermentation process of black tea is relatively long. The current uncontrollable or sudden factors include extreme weather, electrical equipment failures, etc., which cause changes in the process parameters of a certain pile fermentation process, early turning of the pile, and delayed turning of the pile; the overall fermentation degree and quality cannot be effectively and effectively controlled, which has become a key issue for brands in ensuring the consistency of black tea quality.

[0004] Therefore, there is an urgent need to develop an intelligent control method for tea fermentation using a machine learning model, so that under the changes of many different pile-fermentation process parameters during production, the final fermentation degree and effect can be close to consistent due to the prediction of the machine learning model. Summary of the invention

[0005] Technical problems to be solved: In order to overcome the disadvantage that the fermentation degree cannot be accurately controlled due to the changes in many different pile-fermentation process parameters in production, resulting in different quality of black tea.

[0006] The technical solution is: a tea fermentation pile process control method based on a gradient boosting tree, comprising the following steps:

[0007] S1) Data collection: Collect a batch of production data of known high-quality tea leaves during the pile fermentation process for processing; and divide the database into a training set and a test set;

[0008] Furthermore, the training set accounts for 70-80% of the database, and the test set accounts for 20-30% of the database.

[0009] S2) Data preprocessing: A gradient boosting tree regression model of fermentation process parameters and fermentation degree was established; the fermentation process parameters were taken as independent variables X, and the fermentation degree was taken as dependent variable Y; the number of fermentation times was taken as one of the leaf nodes and divided into J mutually non-overlapping regions, recorded as R1, R2, ..., R J , J> the number of times of heaping, and the constants determined in each area are recorded as C1, C2, ..., C J ;

[0010] Furthermore, the fermentation process parameters include temperature, relative humidity and fermentation time.

[0011] Furthermore, in the data preprocessing of step S2), the dependent variable (fermentation degree) of the production data of the known high-quality tea is quantitatively coded as 80-100, and the quantitative code is a quantitative value matched by the manual scoring system (10-100) according to the fermentation degree; since the fermentation degree of the known high-quality tea is first-class or special-class quality, the dependent variable score of the collected production data of the high-quality tea is higher.

[0012] S3) Model training: Use machine learning software to import the data of the training set into the gradient boosting tree regression model, set parameters, and the machine learning software automatically starts analysis and outputs results;

[0013] S4) Model evaluation: Use the test set data to evaluate the performance of the trained model, including the following parameters:

[0014] MSE: The expected value of the square of the difference between the predicted value and the actual value; the smaller the value, the higher the accuracy of the model;

[0015] RMSE: It is the square root of MSE. The smaller the value, the higher the model accuracy.

[0016] MAE: The average value of absolute error, which can reflect the actual situation of the prediction value error; the smaller the value, the higher the model accuracy;

[0017] MAPE: It is a variation of MAE and is a percentage value. The smaller the value, the higher the model accuracy.

[0018] R 2 : The closer the R2 value is to 1, the more accurate the model is.

[0019] S5) Model application: The model is applied to the control of the pile precipitation process. The process is as follows:

[0020] S501) Preparation for pile fermentation: automatic conveying and spraying humidification is adopted, and rolling stirring is performed during the humidification process; the tea leaves are conveyed to a temperature- and humidity-controllable fermentation bed, and the temperature and relative humidity in the pile fermentation process are maintained within a constant range through a controller; according to the trained model, a target value of the degree of tea fermentation is set;

[0021] Furthermore, the target value is set to 85-95.

[0022] S502) Primary fermentation: setting various process parameters, including temperature, relative humidity, and fermentation time, within the range of independent variables in the trained model; and fully turning the compost to disintegrate the blocks after the fermentation time is over;

[0023] S503) a model prediction and operation: input each process parameter set in step S502 as a new independent variable, the independent variable is a determined value, and the machine learning software automatically outputs the predicted value of the dependent variable; compare the predicted value with the target value, if the predicted value is less than the target value, enter the next pile fermentation;

[0024] S504): Repeat the above steps S502 and S503, and input the continuous data of the previous process parameters as new independent variables in the next model prediction and calculation; the machine learning software automatically outputs the predicted value of the dependent variable; compare the predicted value with the target value, if the predicted value is less than the target value, enter the next pile fermentation; if the predicted value is greater than or equal to the target value, it means that the target value of the tea fermentation degree has been achieved, and the pile fermentation process is terminated.

[0025] Furthermore, the machine learning software is SPSSPRO software; it can also be a Python machine learning library, such as scikit-learn or XGBoost.

[0026] Furthermore, in step S503, the process parameters are set as follows:

[0027] First fermentation: temperature is 40-50℃, relative humidity is 85-90%, and fermentation time is 24-36h;

[0028] Secondary fermentation: temperature is 50-55℃, relative humidity is 85-90%, and fermentation time is 96-120h;

[0029] Three times of fermentation: temperature is 50-55℃, relative humidity is 85-90%, and fermentation time is 96-120h;

[0030] Fourth pile fermentation: temperature is 45-50℃, relative humidity is 70-80%, and fermentation time is 60-72h;

[0031] Fifth pile fermentation: temperature is 40-45℃, relative humidity is 60-70%, and fermentation time is 48-60h.

[0032] Furthermore, it also includes n times of pile fermentation: the temperature is 40-45° C., the relative humidity is 50-60%, and the fermentation time is 24-48 hours.

[0033] Furthermore, in step S501, the controller is used to maintain various process parameters in the fermentation bed within a constant range, wherein the temperature is controlled within ±1.5 of the set value, and the relative humidity is controlled within ±2.5 of the set value.

[0034] The beneficial effects of the present invention are as follows: 1. An intelligent control method designed based on a gradient boosting tree and combined with a tea fermentation pile process. The control method of the present invention can generate clear decision rules that are easy to understand and explain. It can display the characteristics and conditions of each decision node, as well as the final prediction results of each boosting tree; this makes the gradient boosting regression algorithm more interpretable in process control.

[0035] 2. When the process parameters are passively or actively adjusted due to some uncontrollable or sudden factors, the predictive ability of the machine learning model can effectively maintain the stability of the fermentation degree and the consistency of quality, thereby significantly improving the stability of the fermentation degree and the consistency of quality of tea leaves under the changes of the pile fermentation process parameters caused by many factors in actual production. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Flow chart of the GBDT regression model application of the present invention.

[0037] Figure 2 It is a schematic diagram of the algorithm implementation principle of the GBDT regression model of the present invention.

[0038] Figure 3 This is a diagram of the SPSSPRO software interface of the GBDT regression model in the test example of the present invention.

[0039] Figure 4 This is a model evaluation result table of the trained GBDT regression model in the test example of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0041] Example 1

[0042] like Figure 1 As shown, the tea fermentation pile process control method based on the gradient boosting tree includes the following steps:

[0043] S1) Data collection: Collect a batch of production data of known high-quality tea leaves during the pile fermentation process for processing; and divide the database into a training set and a test set;

[0044] As a more preferred implementation manner, the training set accounts for 70-80% of the database, and the test set accounts for 20-30% of the database.

[0045] S2) Data preprocessing: A gradient boosting tree regression model of fermentation process parameters and fermentation degree was established; the fermentation process parameters were taken as independent variables X, and the fermentation degree was taken as dependent variable Y; the number of fermentation times was taken as one of the leaf nodes and divided into J mutually non-overlapping regions, recorded as R1, R2, ..., R J , J> the number of times of heaping, and the constants determined in each area are recorded as C1, C2, ..., C J ;

[0046] As a more preferred implementation mode, the pile fermentation process parameters include temperature, relative humidity and fermentation time.

[0047] As a more preferred implementation mode, in the step S1) of collecting data, the dependent variable (fermentation degree) of the production data of the known high-quality tea is quantitatively coded as 80-100, and the quantitative coding is matched with a quantitative value by a manual scoring system (10-100) according to the fermentation degree; since the fermentation degree of the known high-quality tea is first-class or special-class quality, the dependent variable score of the collected production data of the high-quality tea is higher.

[0048] S3) Model training: Use machine learning software to import the data of the training set into the gradient boosting tree regression model, set parameters, and the machine learning software automatically starts analysis and outputs results;

[0049] S4) Model evaluation: Use the test set data to evaluate the performance of the trained model, including the following parameters: MSE: the expected value of the square of the difference between the predicted value and the actual value; the smaller the value, the higher the model accuracy;

[0050] RMSE: It is the square root of MSE. The smaller the value, the higher the model accuracy.

[0051] MAE: The average value of absolute error, which can reflect the actual situation of the prediction value error; the smaller the value, the higher the model accuracy;

[0052] MAPE: It is a variation of MAE and is a percentage value. The smaller the value, the higher the model accuracy.

[0053] R 2 : Compare the predicted value with the case of using only the mean. The closer the result is to 1, the higher the accuracy of the model.

[0054] S5) Model application: The model is applied to the control of the pile precipitation process. The process is as follows:

[0055] S501) Preparation for pile fermentation: automatic conveying and spraying humidification is adopted, and rolling stirring is performed during the humidification process; the tea leaves are conveyed to a temperature- and humidity-controllable fermentation bed, and the temperature and relative humidity in the pile fermentation process are maintained within a constant range through a controller; according to the trained model, a target value of the degree of tea fermentation is set;

[0056] As a more preferred implementation manner, the target value is set to 85-95.

[0057] S502) Primary fermentation: setting various process parameters, including temperature, relative humidity, and fermentation time, within the range of independent variables in the trained model; and fully turning the compost to disintegrate the blocks after the fermentation time is over;

[0058] S503) a model prediction and operation: input each process parameter set in step S502 as a new independent variable, the independent variable is a determined value, and the machine learning software automatically outputs the predicted value of the dependent variable; compare the predicted value with the target value, if the predicted value is less than the target value, enter the next pile fermentation;

[0059] S504): Repeat the above steps S502 and S503, and input the continuous data of the previous process parameters as new independent variables in the next model prediction and calculation; the machine learning software automatically outputs the predicted value of the dependent variable; compare the predicted value with the target value, if the predicted value is less than the target value, enter the next pile fermentation; if the predicted value is greater than or equal to the target value, it means that the target value of the tea fermentation degree has been achieved, and the pile fermentation process is terminated.

[0060] Working principle: Figure 2 The model algorithm principle shown in the figure is that the gradient boosting tree (GBDT) regression model is an additive model, and its training process is serial, that is, it is trained one tree at a time. Each new regression tree is constructed on the basis of all the previously constructed regression trees. In each iteration, the GBDT model learns a new regression tree, and the goal of the tree is to fit the negative gradient direction of the current loss function. In each iteration, the model obtains a loss function by calculating the difference between the prediction result of the current model and the "true value" (the target value in the present invention). Then, the model constructs a new regression tree according to the negative gradient direction of this loss function. In this way, a new regression tree is generated in each iteration, which gradually reduces the loss function of the model and improves the accuracy of the prediction. Finally, the GBDT model adds the prediction results of all regression trees to obtain the final prediction result.

[0061] The present invention uses existing software technology to implement the application of the algorithm, without manual calculation and derivation, and only needs to define variables (independent variables, dependent variables, data quantification and other data preprocessing steps (see S2)); such as SPSSPRO software, its built-in implementation algorithm of the gradient boosting tree (GBDT) regression model is as follows Figure 2 As shown, the software is easy to use and can be used more accurately based on the pre-processed data, such as Figure 3 As shown, the output result is directly and automatically calculated. Therefore, the present invention does not deduce the implementation process of the algorithm principle in detail.

[0062] The present invention applies the trained GBDT regression model to the control of the fermentation pile process of black tea or Pu'er tea. In the pile process step, the target value of the fermentation degree of black tea or Pu'er tea is set, and the process parameters set manually are used as inputs. The predicted value of the dependent variable is obtained through model calculation and prediction, and then compared with the target value; if the predicted value is less than the target value, the next pile is repeated; if the predicted value is greater than or equal to the target value, it means that the target value of the fermentation degree of black tea or Pu'er tea is completed, and the pile process is terminated. By introducing a machine learning model for intelligent control, the interference of subjective factors can be reduced, and a more accurate and stable fermentation degree can be achieved. This will help improve the consistency of the quality of black tea or Pu'er tea and ensure that each batch of products has similar quality characteristics.

[0063] At the same time, the algorithm of the gradient boosting tree regression model can effectively process mixed data, including continuous and discrete features. In the fermentation process of dark tea or Pu'er tea, the process parameters may be continuous (such as the continuous data of the first few process parameters in step S504 (temperature, relative humidity, fermentation time)) or discrete (such as fermentation degree), and the decision tree can flexibly process these different types of features and provide accurate predictions for process control.

[0064] As a more preferred implementation mode, the machine learning software is SPSSPRO software; it can also be a Python machine learning library, such as scikit-learn or XGBoost.

[0065] As a more preferred implementation mode, the process parameters in step S503 are set as follows:

[0066] First fermentation: temperature is 40-50℃, relative humidity is 85-90%, and fermentation time is 24-36h;

[0067] Secondary fermentation: temperature is 50-55℃, relative humidity is 85-90%, and fermentation time is 96-120h;

[0068] Three times of fermentation: temperature is 50-55℃, relative humidity is 85-90%, and fermentation time is 96-120h;

[0069] Fourth pile fermentation: temperature is 45-50℃, relative humidity is 70-80%, and fermentation time is 60-72h;

[0070] Fifth pile fermentation: temperature is 40-45℃, relative humidity is 60-70%, and fermentation time is 48-60h.

[0071] As a more preferred embodiment, it also includes n times of pile fermentation: the temperature is 40-45°C, the relative humidity is 50-60%, and the fermentation time is 24-48h.

[0072] As a more preferred implementation mode, in step S501, the process parameters in the fermentation bed are maintained within a constant range by a controller, wherein the temperature is controlled within ±1.5 of the set value, and the relative humidity is controlled within ±2.5 of the set value.

[0073] Test case: Gradient boosted tree (GBDT) regression model test

[0074] Test steps: Figure 3 As shown, use SPSSPRO software to create a new data analysis; select [Machine Learning Regression - Gradient Boosted Tree (GBDT) Regression]; import the corresponding [Gradient Boosted Tree (GBDT) Regression] data as required and set the parameters [Among them, the loss function selects LAD minimum absolute error, the node splitting evaluation criterion selects Friedman MSE (mean square error of improvement score), and the learning rate is 0.05]; click [Start Analysis], and SPSSPRO software automatically completes the linear regression analysis results. Output as shown Figure 3 Model evaluation results are shown.

[0075] Experimental analysis: From Figure 4 From the data analysis of the model evaluation results table, we can see that MSE, RMSE, MAE, MAPE and R 2 The indicators such as process parameters, fermentation degree, etc. have been quantified and parameter settings have been adjusted (the loss function is the minimum absolute error of LAD, the node splitting evaluation criterion is the mean square error of the improved score, and the learning rate is 0.05), and a reliable and stable model has been obtained. 2 The R 2 It is 0.932, which clearly shows that the model fitting effect is excellent.

[0076] Test results: It can be concluded from the test case that the established gradient boosting tree (GBDT) regression model can be used for automatic prediction. New independent variables or continuous independent variables (process parameters) are input, and the corresponding fermentation degree (dependent variable) is predicted by the model.

[0077] In summary, the control method based on the gradient boosting tree of the present invention is of great significance to the fermentation and piling process of dark tea or Pu'er tea. It not only improves the interpretability and stability of process control, but also ensures the consistency of dark tea quality. This will bring significant benefits to dark tea production, improve production efficiency and product quality, meet the brand's requirements for quality consistency in the market, and help promote the development of the dark tea industry.

[0078] The above-mentioned embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person skilled in the art, several modifications, improvements and substitutions 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 patent shall be subject to the attached claims.

Claims

1. A tea fermentation pile process control method based on a gradient boosting tree, characterized in that: The steps include: S1) Data collection: Collect a batch of production data of known high-quality tea leaves during the pile fermentation process for processing; and divide the database into a training set and a test set; S2) Data preprocessing: A gradient boosting tree regression model of fermentation process parameters and fermentation degree was established; the fermentation process parameters were taken as independent variables X, and the fermentation degree was taken as dependent variable Y; the number of fermentation times was taken as one of the leaf nodes and divided into J mutually non-overlapping regions, denoted as R1, R2, ..., R J , J> the number of times of heaping, and the constants determined in each area are recorded as C1, C2, ..., C J ; The pile fermentation process parameters include temperature, relative humidity and fermentation time; S3) Model training: Use machine learning software to import the data of the training set into the gradient boosting tree regression model, set parameters, and the machine learning software automatically starts analysis and outputs results; S4) Model evaluation: Use the test set data to evaluate the performance of the trained model, including the following parameters: MSE: The expected value of the square of the difference between the predicted value and the actual value; the smaller the value, the higher the model accuracy; RMSE: RMSE is the square root of MSE. The smaller the value, the higher the model accuracy. MAE: The average value of absolute error, which can reflect the actual situation of the prediction value error; the smaller the value, the higher the model accuracy; MAPE: MAPE is a variation of MAE. It is a percentage value. The smaller the value, the higher the model accuracy. R²: compares the predicted value to the case of using only the mean. The closer the R² value is to 1, the more accurate the model is. S5) Model application: The model is applied to the control of the pile precipitation process. The process is as follows: S501) Preparation for pile fermentation: transporting tea leaves to a temperature- and humidity-controlled fermentation bed, maintaining the temperature and relative humidity in the pile fermentation process within a constant range through a controller; setting a target value for the degree of tea fermentation according to the trained model; S502) Primary fermentation: setting various process parameters, including temperature, relative humidity, and fermentation time, within the range of independent variables in the trained model; and fully turning the compost to disintegrate the blocks after the fermentation time is over; S503) a model prediction and operation: input each process parameter set in step S502 as a new independent variable, the independent variable is a determined value, and the machine learning software automatically outputs the predicted value of the dependent variable; compare the predicted value with the target value, if the predicted value is less than the target value, enter the next pile fermentation; S504): repeat the above steps S502 and S503, and input the continuous data of each process parameter of the previous times as a new independent variable in the next model prediction and calculation; The machine learning software automatically outputs the predicted value of the dependent variable; the predicted value is compared with the target value. If the predicted value is less than the target value, the next pile fermentation process is started; if the predicted value is greater than or equal to the target value, it means that the target value of the tea fermentation degree has been achieved, and the pile fermentation process is ended.

2. The tea fermentation pile process control method based on gradient boosting tree according to claim 1, characterized in that: The training set accounts for 70-80% of the database, and the test set accounts for 20-30% of the database.

3. The tea fermentation pile process control method based on gradient boosting tree according to claim 1 is characterized in that: The machine learning software is SPSSPRO software, Python's scikit-learn or XGBoost.

4. The tea fermentation pile process control method based on gradient boosting tree according to claim 1, characterized in that: In the data preprocessing of S2), the dependent variable of the known high-quality tea production data is quantitatively coded as 80-100, and the quantitative code is a quantitative value matched by a manual scoring system according to the fermentation degree.

5. The tea fermentation pile process control method based on gradient boosting tree according to claim 1, characterized in that: The target value is set to 85-95.

6. The tea fermentation pile process control method based on gradient boosting tree according to any one of claims 1 to 5, characterized in that: In step S503, the process parameters are set as follows: First pile fermentation: temperature is 40-50℃, relative humidity is 85-90%, and fermentation time is 24-36h; Secondary fermentation: temperature is 50-55℃, relative humidity is 85-90%, and fermentation time is 96-120h; Three times of fermentation: temperature is 50-55℃, relative humidity is 85-90%, and fermentation time is 96-120h; Fourth pile fermentation: temperature is 45-50℃, relative humidity is 70-80%, and fermentation time is 60-72h; Fifth pile fermentation: temperature is 40-45℃, relative humidity is 60-70%, and fermentation time is 48-60h.

7. The tea fermentation pile process control method based on gradient boosting tree according to claim 1, characterized in that: In step S501, the controller is used to maintain the process parameters in the fermentation bed within a constant range, wherein the temperature is controlled within ±1.5 of the set value, and the relative humidity is controlled within ±2.5 of the set value.

Images