Coffee deep processing process parameter optimization method, medium and system

Abstract

The invention provides a coffee deep processing process parameter optimization method, medium and system, and belongs to the technical field of coffee deep processing, and the method comprises the following steps: firstly, collecting historical coffee processing data, including process parameters of each process and corresponding finished product quality indexes; secondly, analyzing the correlation degree between each process parameter and the quality index by adopting a Bayesian network and a mutual information algorithm, and establishing a preliminary parameter-quality prediction model; and based on the prediction model, generating a group of candidate process parameter schemes by using a data augmentation and optimal transmission theory. Next, small-batch experiments are carried out on the parameter schemes, and corresponding actual quality index data are obtained; and finally, according to experimental data, selecting an optimal process parameter as a target parameter to be output. According to the method, machine learning, an optimization algorithm, small-batch experimental verification and other technical means are fused, and the technical problem that balance optimization of multiple quality factors is difficult to consider in existing coffee deep processing technology optimization is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of coffee deep processing, and in particular, relates to a method, medium and system for optimizing process parameters of coffee deep processing. Background Art

[0002] The traditional coffee deep processing method mainly relies on the experience of the barista, and adjusts the process parameters of each process through repeated attempts, such as roasting temperature and time, extraction temperature and pressure, in order to obtain high-quality coffee products that meet consumer needs. However, this experience-based adjustment method has many defects. First, the formulation of optimal process parameters requires a lot of experiments and tastings, and the process optimization cycle is long and the cost is high. Secondly, even if the current optimal parameter combination is found, it is difficult to maintain the optimal state for a long time with changes in raw materials, environment and other factors. Furthermore, different consumer groups have different preferences for coffee quality, and a set of fixed process parameters is difficult to take into account different needs. In other words, the existing coffee deep processing process optimization has the technical problem of being difficult to take into account the balanced optimization of multiple quality factors. Summary of the invention

[0003] In view of this, the present invention provides a method, medium and system for optimizing coffee deep processing process parameters, which can solve the technical problem of the existing coffee deep processing process optimization that it is difficult to achieve balanced optimization of multiple quality factors.

[0004] The present invention is achieved in that:

[0005] A first aspect of the present invention provides a method for optimizing coffee deep processing process parameters, comprising the following steps:

[0006] S10, obtaining process parameters of multiple steps in multiple groups of historical coffee deep processing processes and their corresponding quality indicators of finished coffee products;

[0007] S20, using Bayesian network and mutual information algorithm to calculate the correlation between each process parameter and each quality index;

[0008] S30, based on the calculated correlation, construct a multi-objective prediction model for coffee quality, and use a random forest algorithm to establish a mapping relationship between parameters and quality indicators, which is recorded as a preliminary mapping;

[0009] S40, based on the preliminary mapping, using a data augmentation method according to the process parameters and their corresponding quality indicators of the finished coffee product to obtain augmented data;

[0010] S50, applying the optimal transmission theory to the augmented data to obtain a candidate process parameter group;

[0011] S60, clustering the candidate process array to obtain a plurality of cluster centers, obtaining the process parameters corresponding to each cluster center to perform small batch experiments to obtain quality index data of finished coffee products as experimental data;

[0012] S70, selecting the process parameters corresponding to the best experimental data as the optimized target process parameters and outputting them.

[0013] On the basis of the above technical solution, the coffee deep processing process parameter optimization method of the present invention can also be improved as follows:

[0014] Wherein, the process parameters include baking process parameters, extraction process parameters, and post-processing process parameters.

[0015] Furthermore, the baking process parameters include baking temperature, baking time, baking machine speed, feed amount, and hot air speed.

[0016] Furthermore, the extraction process parameters include extraction temperature, extraction pressure, extraction time, water-coffee ratio, and grinding particle size.

[0017] Furthermore, the post-processing process parameters include cooling temperature, drying time, packaging gas composition, storage temperature, and storage humidity.

[0018] Furthermore, the quality indicators include acidity, bitterness, aroma concentration, body, and cleanliness.

[0019] Furthermore, the application of the optimal transmission theory to the augmented data is specifically: taking the parameter space generated by all process parameters in the augmented data as the source distribution, and taking the indicator space generated by all quality indicators in the augmented data as the target distribution; constructing a transmission cost function using Euclidean distance and parameter weights to reflect the relationship between process parameter adjustment and quality indicator improvement; solving the optimal transmission plan to obtain an adjustment path from current parameters to target quality; based on preset target quality indicators, screening adjustment paths that meet quality requirements and corresponding process parameter combinations, combining preliminary mapping relationships for feasibility assessment, and finally obtaining a candidate process array.

[0020] Furthermore, the candidate process array is specifically a combination of multiple sets of process parameters that meet the quality target and are within the feasible range of the process.

[0021] Specifically, the specific steps of step S10 are: obtaining multiple sets of process parameters of multiple processes in the deep processing of coffee and their corresponding quality indicators of the finished coffee products. First, it is necessary to collect and organize historical data. Assume that there are n sets of historical data, each set of data contains m process parameters and k quality indicators. These data can be represented in matrix form: process parameter matrix P and quality indicator matrix Q. The element p of the process parameter matrix P is ijRepresents the jth process parameter of the i-th group of data, element q of the quality index matrix Q ij represents the jth quality index of the i-th group of data. These historical data provide the basis for the subsequent correlation analysis between process parameters and quality indicators.

[0022] The specific steps of step S20 are: using Bayesian network and mutual information algorithm to calculate the correlation between each process parameter and each quality index. First, it is necessary to construct a Bayesian network. Assume that there are m process parameter nodes X1, X2, ..., X m and k quality indicator nodes Y1,Y2,…,Y k The structure of the Bayesian network can be determined by a structural learning algorithm (such as the K2 algorithm or the greedy search algorithm). For each pair of process parameters X i and quality index Y j , calculate the mutual information I(X i ; Y j ). To calculate mutual information, we need to estimate the joint probability distribution p(x, y) and the marginal probability distributions p(x) and p(y). Here we use the kernel density estimation method. Finally, we get a correlation matrix R, where r ij =I(X i ; Y j ) represents the correlation between the i-th process parameter and the j-th quality index. The correlation matrix reflects the correlation strength between each process parameter and the quality index, and provides a basis for the subsequent multi-objective prediction model construction.

[0023] The specific steps of step S30 are: based on the calculated correlation, a multi-objective prediction model for coffee quality is constructed, and a mapping relationship between parameters and quality indicators is established using a random forest algorithm, which is recorded as a preliminary mapping. A mapping relationship between process parameters and quality indicators is established using a random forest algorithm. For each quality indicator Y j , build a random forest model RF j Each random forest model RF j It consists of T decision trees, and the prediction result of each decision tree is The final prediction result is the average of all decision tree prediction results: The construction process of each decision tree includes: extracting n samples with replacement from the original sample to form a bootstrap sample set, randomly selecting m try features, find the best split point to maximize the purity of the child nodes after the split, and repeat until the stopping condition is reached. The splitting criteria of the decision tree can use the Gini index or information gain. The random forest model RF established through the above steps j This constitutes the process parameters to quality indicators Y j The initial mapping relationship.

[0024] The specific steps of step S40 are: based on the preliminary mapping, according to the process parameters and their corresponding quality indicators of the finished coffee product, the data augmentation method is used to obtain augmented data. In order to expand the original data set, the following data augmentation methods can be used: 1) Add Gaussian noise, that is, for each original sample (x i ,y i ), generate new samples (x i ′,y i '),in Linear interpolation, that is, for any two original samples (x i ,y i ) and (x j ,y j ), generate new samples (x new ,y new ), where x new =λx i +(1-λ)x j ,y new =λy i +(1-λ)y j ,λ∈[0,1]; 3) SMOTE algorithm, that is, for the minority class sample, find its k nearest neighbors, and then randomly select a nearest neighbor x j , generate new samples x new =x i +α(x j -x i ),α∈[0,1]. Through the above method, the original data set can be expanded to N samples to form an augmented data set D aug .

[0025] The specific steps of step S50 are: applying the optimal transmission theory to the augmented data to obtain a candidate process parameter set. First, define the source distribution μ (process parameter space) and the target distribution v (quality index space) as empirical distributions. Then define the transmission cost function where w i is the weight of the i-th parameter. The optimal transmission problem can be expressed as: min γ∈Π(μ,v) ∫ X×Y c(x,y)dγ(x,y). To solve this problem, we can use the Sinkhorn algorithm. The iterative steps of the algorithm are: 1) Initialize u (0) =v (0) =1; 2) For t=1,2,…, repeat u (t) =μ. / K(v (t-1) ),v (t) =v. / (K T u (t)) until convergence, where K = e -C / ∈ is the Gibbs kernel, ∈ is the regularization parameter. The final transmission plan is P ij =u i K ij v j According to the transmission plan, the adjustment path from the current parameters to the target quality can be obtained. The path that meets the preset quality target is selected, and the feasibility is evaluated in combination with the preliminary mapping relationship, and finally the candidate process parameter group is obtained.

[0026] The specific steps of step S60 are: clustering the candidate process array to obtain multiple cluster centers, obtaining the process parameters corresponding to each cluster center and performing small batch experiments to obtain the quality index data of the finished coffee product as experimental data. First, the K-means algorithm is used to cluster the candidate process parameter groups. The objective function of the K-means algorithm is Where K is the number of clusters, is the i-th sample belonging to the j-th cluster, μ j is the center of the jth cluster. The algorithm steps include: 1) randomly initialize K cluster centers; 2) repeat a) assign each sample to the nearest cluster center, b) update each cluster center until convergence. In order to determine the optimal number of clusters K, the silhouette coefficient can be used Select the K value that maximizes S as the optimal number of clusters. For each cluster center μ j , conduct small batch experiments and obtain experimental data E j =(e j1 ,e j2 ,…,e jk ).

[0027] The specific steps of step S70 are: select the process parameters corresponding to the optimal experimental data as the optimized target process parameters and output them. Through the previous steps, multiple candidate process parameter groups and their corresponding experimental data are obtained. Now it is necessary to select the optimal process parameters from them as the final result output. The candidate solutions can be judged according to the following criteria: 1) Whether all the quality indicators of the finished coffee product have reached the expected goals; 2) Whether the experimental values ​​of each quality indicator are relatively balanced to avoid the situation where some indicators are too high and other indicators are too low; 3) Whether the process parameters are within the feasible range and will not cause problems such as equipment failure or excessive cost. Taking the above factors into consideration, select the process parameters corresponding to the optimal experimental data as the final optimization result output.

[0028] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for optimizing coffee deep processing process parameters.

[0029] A third aspect of the present invention provides a coffee deep processing parameter optimization system, which includes the above-mentioned computer-readable storage medium.

[0030] Compared with the prior art, the coffee deep processing process parameter optimization method, medium and system provided by the present invention have the following beneficial effects:

[0031] 1. Improved the accuracy and robustness of parameter-quality relationship modeling. Traditional linear or simple nonlinear models are difficult to capture the complex parameter interactions in the actual production process. The present invention uses Bayesian networks and random forest algorithms to better explore this nonlinear relationship, thereby greatly improving prediction accuracy.

[0032] 2. The coordinated optimization of multiple quality factors is achieved. Many existing methods only optimize a single quality index, but fail to take into account the balanced improvement of multiple quality attributes such as acidity, bitterness, and aroma. Based on the optimal transmission theory, the present invention can integrate the optimization targets of various quality indicators into one framework and provide the optimal process parameters that meet multiple quality targets.

[0033] 3. The optimization results are highly operational. The present invention not only provides ideal quality index target values, but also obtains specific and feasible process parameter solutions through cluster analysis and small batch experiments, providing direct technical support for production practice. Compared with empirical adjustments, this method based on data analysis is more scientific, systematic and replicable.

[0034] In summary, the present invention solves the technical problem that it is difficult to achieve balanced optimization of multiple quality factors in the existing coffee deep processing technology optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A flow chart of the method provided by the present invention;

[0036] Figure 2 It is the parameter-quality correlation matrix diagram;

[0037] Figure 3 This is the K-means clustering result diagram. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0039] like Figure 1 FIG. 1 is a flow chart of a method for optimizing coffee deep processing process parameters provided by the present invention. The method comprises the following steps:

[0040] S10, obtaining process parameters of multiple steps in multiple groups of historical coffee deep processing processes and their corresponding quality indicators of finished coffee products;

[0041] S20, using Bayesian network and mutual information algorithm to calculate the correlation between each process parameter and each quality index;

[0042] S30, based on the calculated correlation, construct a multi-objective prediction model for coffee quality, and use a random forest algorithm to establish a mapping relationship between parameters and quality indicators, which is recorded as a preliminary mapping;

[0043] S40, based on the preliminary mapping, using a data augmentation method to obtain augmented data according to the process parameters and their corresponding quality indicators of the finished coffee product;

[0044] S50, applying the optimal transmission theory to the augmented data to obtain a candidate process parameter group;

[0045] S60, clustering the candidate process array to obtain multiple cluster centers, obtaining the process parameters corresponding to each cluster center, performing small batch experiments to obtain quality index data of finished coffee products as experimental data;

[0046] S70, selecting the process parameters corresponding to the best experimental data as the optimized target process parameters and outputting them.

[0047] The specific implementation methods of the above steps are described in detail below:

[0048] The specific implementation of step S10 is as follows:

[0049] First, we need to collect and organize historical data. Assume that there are n sets of historical data, each set of data contains m process parameters and k quality indicators. These data can be expressed in matrix form:

[0050] Process parameter matrix P:

[0051]

[0052] Among them, the element p ij Represents the jth process parameter of the i-th group of data.

[0053] Quality index matrix Q:

[0054]

[0055] Among them, the element q ij Represents the j-th quality indicator of the i-th group of data.

[0056] These historical data provide the basis for the subsequent correlation analysis between process parameters and quality indicators.

[0057] The specific implementation of step S20 is as follows:

[0058] First, we need to build a Bayesian network. Assume there are m process parameter nodes X1, X2, …, X m and k quality indicator nodes Y1,Y2,…,Y k The structure of a Bayesian network can be determined using a structure learning algorithm such as the K2 algorithm or a greedy search algorithm.

[0059] For each pair of process parameters X i and quality index Y j , calculate the mutual information I(X i ; Y j ):

[0060]

[0061] Where p(x,y) is X i and Y j The joint probability distribution of X is p(x) and p(y) are i and Y j The marginal probability distribution of .

[0062] To calculate these probability distributions, kernel density estimation methods can be used:

[0063]

[0064] Where K(·) is the kernel function (such as Gaussian kernel), h x and h y is the bandwidth parameter.

[0065] Finally, we get a correlation matrix R:

[0066]

[0067] Among them, r ij =i(X i ; Y j ) represents the correlation between the i-th process parameter and the j-th quality index. The correlation matrix reflects the correlation strength between each process parameter and the quality index, and provides a basis for the subsequent multi-objective prediction model construction.

[0068] The specific implementation of step S30 is as follows:

[0069] The random forest algorithm is used to establish the mapping relationship between process parameters and quality indicators. j , build a random forest model RF j Each random forest model RF jIt consists of T decision trees, and the prediction result of each decision tree is The final prediction result is the average of all decision tree prediction results:

[0070]

[0071] The construction process of each decision tree is as follows:

[0072] Randomly select n samples with replacement from n samples to form a Bootstrap sample set.

[0073] Randomly select m try Features (usually ).

[0074] Find the best split point among the selected features so that the purity of the child nodes after the split is improved the most.

[0075] Repeat steps 2 and 3 until a stopping condition (such as maximum depth or minimum number of leaf node samples) is reached.

[0076] The splitting criteria of the decision tree can use the Gini index or information gain. Taking the Gini index as an example, the Gini index of node t is defined as:

[0077]

[0078] Among them, p j is the proportion of samples belonging to category j in node t.

[0079] Information gain is defined as the impurity of the parent node minus the impurity of the weighted child nodes:

[0080]

[0081] Where N is the number of samples of the parent node, N i is the number of samples of the ith child node.

[0082] The random forest model RF established through the above steps j This constitutes the process parameters to quality indicators Y j The initial mapping relationship.

[0083] The specific implementation of step S40 is as follows:

[0084] In order to expand the original dataset, several data augmentation methods can be used:

[0085] Add Gaussian noise:

[0086] For each original sample (x i ,y i ), generate new samples (x i′,y i ′):

[0087]

[0088] Among them, σ x and σ y It is the noise intensity parameter and needs to be adjusted according to the actual situation.

[0089] Linear interpolation:

[0090] For any two original samples (x i ,y i ) and (x j ,y j ), can generate new samples (x new ,y new ):

[0091] x new =λx i +(1-λ)x j ;

[0092] y new =λy i +(1-λ)y j ;

[0093] Among them, λ∈[0,1] is the interpolation coefficient.

[0094] SMOTE (Synthetic Minority Over-sampling Technique):

[0095] For minority class samples, the SMOTE algorithm can be used to generate new samples. i , find its k nearest neighbors, and then randomly select a nearest neighbor x j , generate new samples x new :

[0096] x new =x i +α(x j -x i );

[0097] Among them, α∈[0,1] is a random number.

[0098] Through the above method, the original data set can be expanded to N samples to form an augmented data set D aug ={(x1,y1),(x2,y2),…,(x N ,y N)}. These newly generated data can further enrich the relationship between process parameters and quality indicators, laying the foundation for the subsequent application of optimal transmission theory.

[0099] The specific implementation of step S50 is as follows:

[0100] First, define the source distribution μ (process parameter space) and the target distribution ν (quality index space) as empirical distributions:

[0101]

[0102] Among them, δ x represents the Dirac delta function.

[0103] Then define the transmission cost function c(x,y):

[0104]

[0105] Among them, w i is the weight of the i-th parameter, which can be determined based on the previously calculated correlation.

[0106] The optimal transmission problem can be formulated as:

[0107] min γ∈Π(μ,v) ∫ X×Y c(x,y)dγ(x,y);

[0108] where Π(μ,v) is the set of all joint distributions that transfer μ to v.

[0109] To solve this problem, we can use the Sinkhorn algorithm. First, discretize the problem:

[0110]

[0111] st∑ i P ij =μ i ,∑ i P ij =v j ,P ij ≥0;

[0112] Among them, C ij =c(x i ,y j ) is the cost matrix, P ij It is a transmission plan.

[0113] The iterative steps of the Sinkhorn algorithm are as follows:

[0114] Initialize u (0) =v (0) =1.

[0115] For t=1,2,…, repeat the following steps until convergence:

[0116] u (t) =μ. / K(v (t-1) );

[0117] v (t) =v. / (K T u (t) );

[0118] Where K = e -C / ∈ is the Gibbs kernel, ∈ is the regularization parameter, and . / denotes element-wise division.

[0119] The final transmission plan is P ij =u i K ij v j According to the transmission plan, the best adjustment path from the current parameters to the target quality can be obtained. The path that meets the preset quality target is selected, and the feasibility is evaluated in combination with the preliminary mapping relationship, and finally the candidate process parameter group is obtained.

[0120] The specific implementation of step S60 is as follows:

[0121] First, the K-means algorithm is used to cluster the candidate process parameter groups and obtain multiple cluster centers. The objective function of the K-means algorithm is:

[0122]

[0123] Where K is the number of clusters, is the i-th sample belonging to the j-th cluster, μ j is the center of the jth cluster.

[0124] The steps of the K-means algorithm are as follows:

[0125] Randomly initialize K cluster centers μ1,μ2,…,μ K .

[0126] Repeat the following steps until convergence:

[0127] For each sample x i , assign it to the nearest cluster center:

[0128] c i =argmin j ||x i -μ j || 2 ;

[0129] Update each cluster center:

[0130]

[0131] Among them, C j is the set of all samples belonging to the jth cluster.

[0132] To determine the optimal number of clusters K, the Silhouette Coefficient can be used:

[0133]

[0134] Among them, a(i) is the average distance between sample i and other samples in the same cluster, and b(i) is the average distance between sample i and the samples in the nearest other cluster.

[0135] The overall silhouette coefficient is the average of all sample silhouette coefficients:

[0136]

[0137] The K value that maximizes S is selected as the optimal number of clusters.

[0138] For each cluster center μ j , conduct small batch experiments and obtain experimental data E j =(e j1 ,e j2 ,…,e jk ), where e ji represents the experimental value of the i-th quality index under the process parameters corresponding to the j-th cluster center. Step S70 compares the quality index performance of different experimental data and selects the process parameters corresponding to the best experimental data as the final optimization result output.

[0139] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for optimizing coffee deep processing process parameters.

[0140] A third aspect of the present invention provides a coffee deep processing parameter optimization system, which includes the above-mentioned computer-readable storage medium.

[0141] Specifically, the principle of the present invention is to make full use of historical production data, and through advanced data analysis and modeling techniques such as Bayesian networks, random forests and optimal transmission theory, build a multi-objective prediction model from process parameters to product quality, and find the best combination of process parameters based on this model.

[0142] Specifically, firstly, in step S10, multiple groups of historical process parameter data and corresponding quality index data in the coffee deep processing process are collected to form a process parameter matrix P and a quality index matrix Q. These historical data provide a basis for subsequent correlation analysis and modeling.

[0143] In step S20, the Bayesian network structure learning algorithm is used to construct the association network topology between process parameters and quality indicators. Then, the mutual information calculation is used to quantitatively evaluate the association between each pair of parameters and quality indicators, and the association matrix R is obtained. The purpose of this step is to explore the potential dependency between parameters and quality, laying the foundation for establishing a prediction model.

[0144] Based on the correlation information obtained in step S20, a multi-objective prediction model from process parameters to quality indicators is established in step S30 using a random forest algorithm. Random forest is an integrated learning method consisting of multiple decision trees that can effectively fit complex nonlinear relationships. By training the random forest model, a preliminary parameter-quality mapping relationship can be obtained.

[0145] In order to further enrich this mapping relationship, step S40 adopts the method of data augmentation, using Gaussian noise, linear interpolation, SMOTE and other technologies to expand the original data set and generate more samples. These newly added samples can enable the prediction model to better summarize the complex relationship between parameters and quality.

[0146] With the preliminary prediction model and augmented data, step S50 applies the optimal transmission theory to find the best adjustment path from the current parameters to the target quality. The optimal transmission problem is to find the optimal transmission plan between the source distribution (process parameter space) and the target distribution (quality index space) to minimize the total transmission cost. By solving this optimization problem, the best process parameter combination that meets each quality target can be obtained.

[0147] Finally, step S60 performs cluster analysis on these candidate parameter solutions, finds the key cluster centers, and conducts small batch experimental verification. The purpose of this step is to further confirm the final optimal process parameters and provide feasible optimization results for production practice.

[0148] To sum up, the core innovations of the present invention are: first, the use of advanced data analysis methods such as Bayesian networks and random forests can better explore complex parameter-quality relationships; second, the introduction of optimal transmission theory can achieve the coordinated optimization of multiple quality factors; third, through data augmentation and experimental verification, a final optimization result with strong operability is given.

[0149] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: The R&D team of a coffee company has been committed to improving the quality of its coffee products to meet the needs of different consumer groups. The production management department of the company has collected and sorted out the historical data of coffee deep processing in the past five years, including the process parameters of each process such as roasting, extraction, and post-processing, as well as the corresponding quality index data. By analyzing and modeling these data, the R&D team decided to adopt the coffee deep processing process parameter optimization method proposed in the present invention, in order to find a set of more optimal process parameter formulas.

[0150] First, in step S10, the R&D team compiled a total of 100 sets of data from the past five years. Each set of data contained five process parameters and four quality indicators, forming the process parameter matrix P and quality indicator matrix Q shown in Table 1.

[0151] Table 1 Overview of historical production data

[0152]

[0153] In step S20, the R&D team first constructed a Bayesian network structure containing 5 process parameter nodes and 4 quality index nodes. The K2 algorithm was used to learn the network structure and determine the dependencies between the nodes.

[0154] Then, the R&D team calculated the mutual information value between each pair of process parameters and quality indicators, obtained the correlation matrix R as shown in Table 2, and established the corresponding parameter-quality correlation matrix, as shown in Table 2. Figure 2 As shown, it can be seen that the baking temperature and time have a strong positive correlation with acidity and bitterness, the extraction temperature and pressure have a strong positive correlation with aroma concentration and body, while the correlation between packaging humidity and various quality indicators is relatively weak.

[0155] Table 2 Parameter-quality correlation matrix

[0156] acidity Bitterness Aroma concentration posture Baking temperature 0.78 0.71 0.46 0.39 Baking time 0.72 0.65 0.41 0.35 Extraction temperature 0.52 0.48 0.81 0.76 Extraction pressure 0.47 0.43 0.75 0.71 Packaging humidity 0.29 0.26 0.22 0.19

[0157] Based on the above analysis results, the R&D team constructed four random forest models in step S30, which are used to predict acidity, bitterness, aroma concentration and body respectively. Taking the prediction of acidity as an example, the specific construction process of the random forest model is as follows:

[0158] First, 100 samples are randomly selected with replacement from 100 samples to form a bootstrap sample set. Then, three features (baking temperature, baking time, extraction temperature) are randomly selected for splitting. Among the selected features, find the best split point that increases the purity of the child nodes after splitting the most. Repeat the above steps until the maximum tree depth of 5 or the minimum number of leaf node samples of 5 is reached.

[0159] The random forest model consists of 100 such decision trees, and the prediction result of each tree is The final prediction is the average of all tree predictions:

[0160]

[0161] Similarly, the R&D team built three random forest models for predicting bitterness, aroma concentration, and body. By training these models, a preliminary mapping relationship from process parameters to quality indicators was obtained.

[0162] In order to further enrich this mapping relationship, the R&D team used data augmentation technology in step S40. First, Gaussian noise was used to add noise with a mean of 0 and a standard deviation of 0.2 to the original 100 samples, generating 100 new samples. Then, 10 pairs were randomly selected from the original samples, and 10 new samples were generated using the linear interpolation method. Finally, for minority samples (such as samples with acidity below 4.0), the SMOTE algorithm was used to generate 20 new samples.

[0163] Combining the above three methods, the R&D team obtained 230 augmented samples, forming a new dataset D aug These new samples further enrich the relationship between parameters and quality, laying the foundation for subsequent optimal transmission optimization.

[0164] With the preliminary prediction model and augmented data, the R&D team applies the optimal transmission theory in step S50 to find the best adjustment path from the current parameters to the target quality. First, the process parameter space and quality index space are defined as the empirical distribution μ and v, respectively:

[0165]

[0166] Among them, x i is the process parameter of the i-th sample, y i is the quality index of the i-th sample.

[0167] Then, the transmission cost function is defined as:

[0168] c(x,y)=0.4(x1-y1) 2 +0.3(x2-y2) 2 +0.2(x3-y3) 2 +0.1(x4-y4) 2 +0.05(x5-y5) 2 ;

[0169] Among them, the weight of each parameter is determined according to the correlation matrix R obtained in step S20.

[0170] Finally, the R&D team used the Sinkhorn algorithm to solve the optimal transmission problem and obtained the best adjustment path from the current parameter combination to the target quality index. Based on this path, the R&D team screened out three sets of candidate parameter solutions that met the preset quality goals and were within the feasible range of the process, as shown in Table 3.

[0171] Table 3 Three groups of candidate process parameters

[0172]

[0173] In step S60, the R&D team performs K-means cluster analysis on the three groups of candidate parameters, and the number of clusters is set to 2. Figure 3 As shown, Figure 3 The Centroids in represent the cluster centers.

[0174] After iterative calculation, two cluster centers were finally obtained. The specific parameters are shown in Table 4.

[0175] Table 4 Process parameters of two cluster centers

[0176]

[0177] For these two cluster centers, the R&D team conducted small batch experiments and tested the corresponding quality indicator data. The results are shown in Table 5.

[0178] Table 5 Experimental data of quality indicators of two cluster centers

[0179] Cluster Center acidity Bitterness Aroma concentration posture 1 4.0 4.3 4.2 4.6 2 4.1 4.4 4.3 4.7

[0180] By comparing these two sets of experimental data, the R&D team determined that cluster center 2 is the best parameter solution, namely:

[0181] Baking temperature is 211°C, baking time is 20 minutes, extraction temperature is 96°C, extraction pressure is 1.4MPa, and packaging humidity is 58%.

[0182] This optimization result not only meets the overall target requirements of the R&D team for each quality index (acidity ≥ 4.0, bitterness ≥ 4.3, aroma concentration ≥ 4.2, body ≥ 4.5), but also has been verified in terms of process feasibility. Compared with the previous empirical adjustment method, this optimization method based on Bayesian network, random forest and optimal transmission theory has significant advantages in both theoretical analysis and practical application:

[0183] It can more accurately mine complex parameter-quality relationships and improve the accuracy and robustness of the prediction model.

[0184] The collaborative optimization of multiple quality indicators is achieved, and the optimal combination of process parameters that meets various quality goals is given.

[0185] The optimization results are highly operational and provide direct guidance for production practice.

[0186] This method has good versatility and scalability. It is not only suitable for the coffee industry, but can also be applied to the processing technology optimization of other agricultural products.

[0187] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing coffee deep processing process parameters, characterized in that: The following steps are involved: S10, obtaining process parameters of multiple steps in multiple groups of historical coffee deep processing processes and their corresponding quality indicators of finished coffee products; S20, using Bayesian network and mutual information algorithm to calculate the correlation between each process parameter and each quality index; S30, based on the calculated correlation, construct a multi-objective prediction model for coffee quality, and use a random forest algorithm to establish a mapping relationship between parameters and quality indicators, which is recorded as a preliminary mapping; S40, based on the preliminary mapping, using a data augmentation method according to the process parameters and their corresponding quality indicators of the finished coffee product to obtain augmented data; S50, applying the optimal transmission theory to the augmented data to obtain a candidate process parameter group; S60, clustering the candidate process array to obtain a plurality of cluster centers, obtaining the process parameters corresponding to each cluster center to perform small batch experiments to obtain quality index data of finished coffee products as experimental data; S70, selecting the process parameters corresponding to the best experimental data as the optimized target process parameters and outputting them.

2. The method for optimizing coffee deep processing process parameters according to claim 1, characterized in that: The process parameters include baking process parameters, extraction process parameters, and post-processing process parameters.

3. The method for optimizing coffee deep processing parameters according to claim 2, characterized in that: The baking process parameters include baking temperature, baking time, baking machine speed, feed amount, and hot air speed.

4. The method for optimizing coffee deep processing parameters according to claim 3, characterized in that: The extraction process parameters include extraction temperature, extraction pressure, extraction time, water-coffee ratio, and grinding particle size.

5. The method for optimizing coffee deep processing parameters according to claim 4, characterized in that: The post-processing process parameters include cooling temperature, drying time, packaging gas composition, storage temperature, and storage humidity.

6. The method for optimizing coffee deep processing parameters according to claim 5, characterized in that: The quality indicators include acidity, bitterness, aroma concentration, body, and cleanliness.

7. The method for optimizing coffee deep processing parameters according to claim 6, characterized in that: The specific application of the optimal transmission theory to the augmented data is: taking the parameter space generated by all process parameters in the augmented data as the source distribution, and taking the indicator space generated by all quality indicators in the augmented data as the target distribution; constructing a transmission cost function using Euclidean distance and parameter weights to reflect the relationship between process parameter adjustment and quality indicator improvement; solving the optimal transmission plan to obtain an adjustment path from current parameters to target quality; based on preset target quality indicators, screening adjustment paths that meet quality requirements and corresponding process parameter combinations, combining preliminary mapping relationships to conduct feasibility assessments, and finally obtaining a candidate process array.

8. The method for optimizing coffee deep processing parameters according to claim 7, characterized in that: The candidate process array is specifically a combination of multiple sets of process parameters that meet the quality target and are within the feasible range of the process.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the method for optimizing coffee deep processing process parameters according to any one of claims 1 to 8.

10. A coffee deep processing process parameter optimization system, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images