Production control method of infant freeze-dried food

Abstract

The invention belongs to the field of production control of freeze-dried food, and particularly relates to a production control method of infant freeze-dried food. The invention relates to a production control method of infant freeze-dried food. The production control method comprises the following steps: performing multi-dimensional feature normalization on raw material characteristics and production process parameters; a model is constructed by adopting a particle swarm optimization (PSO) optimization ridge regression algorithm, product quality indexes are predicted, and production process parameters are adjusted in real time; a large amount of production data is collected, potential rules and trends are mined by using a big data analysis technology, and accurate guidance is provided for production. According to the method, the potential relationship and trend between the process parameters and the product quality can be found, and accurate guidance is provided for optimizing the production process, so that the product quality and the production efficiency are improved, and the requirement of the market for high-quality infant food is met.

Description

Technical Field

[0001] The invention belongs to the field of production control of freeze-dried food, and in particular relates to a production control method of freeze-dried food for infants. Background Art

[0002] In the current infant food market, consumers have increasingly higher requirements for food quality and safety. Freeze-drying technology is widely used in the production of infant food because it can retain the nutrients and taste of food to the maximum extent. However, the production process of infant freeze-dried food is relatively complicated, involving multiple factors such as raw material characteristics and production process parameters. How to effectively control the production process and ensure the stability and safety of product quality has become an urgent problem to be solved in the industry.

[0003] At present, traditional production control methods often rely on experience and fixed process parameters, lack of analysis and adjustment of real-time data, and are difficult to meet the market demand for high-quality infant food. In addition, differences in raw material properties and uncertainties in the production process also pose challenges to product quality control. Therefore, a more scientific and accurate production control method is needed that can monitor and adjust production process parameters in real time and explore potential laws and trends to improve the quality and safety of infant freeze-dried food. Summary of the invention

[0004] In view of the technical problems existing in the above-mentioned traditional production process, the present invention proposes a solution which is simple in method, easy in operation and can effectively control the production process to ensure the stability and safety of product quality.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is a production control method of infant freeze-dried food, comprising the following steps:

[0006] S1. First, conduct a comprehensive test on the raw materials of infant freeze-dried food to obtain the characteristics of the raw materials, collect the parameters of the production process, and normalize the multi-dimensional features of the two;

[0007] S2. Secondly, the particle swarm optimization (PSO) ridge regression algorithm is used to build a model, and the normalized raw material characteristics and production process are used to predict product quality indicators;

[0008] S3, then the model analyzes and predicts product quality indicators in real time based on the data, and adjusts production process parameters in real time;

[0009] S4. Finally, collect a large amount of production data, use big data analysis technology to explore potential patterns and trends, and provide more accurate guidance for subsequent production.

[0010] In step S2, the particle swarm PSO is used to optimize the ridge regression algorithm model. The model is constructed by using ridge regression and the particle swarm algorithm PSO is used to optimize the parameters, and then the cross-validation technology is used to select the optimal regularization parameter λ. The specific implementation steps are:

[0011] S21. First, the definition of the ridge regression loss function consists of two parts: the residual square and the regularization term is the residual square, is the regularization term, y i is the actual target value of the ith sample, x ij is the value of the i-th sample on the j-th feature, β0 is the intercept term, β j is the regression coefficient of the jth feature, and λ is the regularization coefficient, which is used to control the complexity of the model;

[0012] S22. Secondly, use the particle swarm algorithm PSO to find the regression coefficient β with the minimum loss function j and regularization coefficient λ, the PSO update formula is in, represents the velocity of the i-th particle in the j-th dimension, Represents the position of the i-th particle in the j-th dimension, that is, the regression coefficient β j and λ, ω are inertia weights, which control the inertia of the particle. c1, c2 are learning factors. r1, r2 are random numbers between 0 and 1, which are used to increase the randomness of the search. represents the individual optimal position of the ith particle in the jth dimension, represents the global optimal position of the jth dimension in the population;

[0013] S23, then perform cross-validation to evaluate the performance of the model, divide the training data into k non-overlapping subsets, use k-1 of them for model training in turn, and use the remaining one to verify the model performance, repeat k times, select a different validation set each time, and then calculate the mean square error of each validation, and take the average of all k validation results as the overall performance of the model;

[0014] S24. Finally, the mean square error obtained through cross-validation corresponds to different regularization coefficient λ values, and the regularization coefficient λ value with the smallest mean square error is selected as the optimal regularization parameter.

[0015] Preferably, the specific implementation steps of performing multidimensional feature normalization in step S1 are:

[0016] S11. First calculate the mean and standard deviation of each feature, mean: Standard Deviation: Among them, xij represents the jth feature of the i-th sample, and N represents the number of samples;

[0017] S12. Next, based on the calculated mean and standard deviation, each feature is standardized. At the same time, the correlation between the features is considered to ensure that the normalization process of the data retains the intrinsic connection of the multidimensional features. The standardization formula is: where z ij It is the standardized data;

[0018] S13. Then, weights are assigned according to the importance of different features to ensure that the features that have a greater impact on the final product quality have higher weights during the normalization process. The weighted normalization formula is: i ' j =z ij ·ω j Among them, ω j is the weight of the jth feature;

[0019] S14. Finally, after the model is trained, the weights are continuously updated to ensure a positive improvement in the prediction capability.

[0020] Preferably, the collected raw material characteristics include nutrient content, fiber content, freshness and microbial content.

[0021] Preferably, the collected production process parameters include freezing rate, pre-freezing temperature, drying chamber pressure, heating plate temperature, drying time, material thickness and partition spacing.

[0022] Preferably, the product quality index is a score from 0 to 100, and the product is qualified only when the prediction result is greater than 85 points, otherwise it is unqualified.

[0023] Preferably, the big data analysis technology is used in step S4 to cluster, mine association rules and analyze trends of a large amount of historical production data, discover the potential relationship and trend between process parameters and product quality, and provide accurate guidance for optimizing the production process.

[0024] Compared with the prior art, the advantages and positive effects of the present invention are: multi-dimensional feature normalization of raw materials and production process parameters, taking into account the correlation and importance between the features, ensuring the validity of the data and the positive improvement of the prediction ability. The particle swarm PSO optimization ridge regression algorithm model is used to achieve real-time analysis and prediction quality scoring of production process parameters, which can quickly and accurately adjust production process parameters and further improve product quality. The use of big data analysis technology to explore potential laws and trends provides precise guidance for optimizing production processes, which helps to improve production efficiency and product quality. DETAILED DESCRIPTION

[0025] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0027] Embodiment, the production process of infant freeze-dried food is relatively complex, involving multiple factors such as raw material characteristics and production process parameters. The production control method often relies on experience and fixed process parameters, lacks analysis and adjustment of real-time data, and is difficult to meet the market demand for high-quality infant food. The present invention proposes a production control method for infant freeze-dried food.

[0028] First, the raw material characteristics and production processes of infant freeze-dried food are collected. Specifically, the raw materials of infant freeze-dried food are comprehensively tested to obtain raw material characteristics including nutrient content, fiber content, freshness, microbial content, etc., and production process parameters such as freezing rate, pre-freezing temperature, drying chamber pressure, heating plate temperature, drying time, material thickness, and partition spacing are collected.

[0029] Three groups of experimental data were collected for prediction. The first group had a nutrient content of 25 mg / 100 g, a fiber content of 3 g / 100 g, a freshness of 95%, a microbial content of 10 CFU / g, a freezing rate of 5 °C / h, a pre-freezing temperature of -20 °C, a drying chamber pressure of 10 Pa, a heating plate temperature of 50 °C, a drying time of 10 h, a material thickness of 5 mm, and a partition spacing of 10 cm; the second group had a nutrient content of 22 mg / 100 g, a fiber content of 2.5 g / 100 g, a freshness of 92%, a microbial content of 18 CFU / g, a freezing rate of 4℃ / h, pre-freezing temperature -18℃, drying chamber pressure 8Pa, heating plate temperature 48℃, drying time 9h, material thickness 4mm, partition spacing 10cm; the third group of nutrient content 24mg / 100g, fiber content 2.8g / 100g, freshness 94%, microbial content 9CFU / g, freezing rate 4.5℃ / h, pre-freezing temperature -19℃, drying chamber pressure 9Pa, heating plate temperature 49℃, drying time 9.5h, material thickness 8mm, partition spacing 5cm;

[0030] Considering the need to transform the above collected data into information language that can be used for machine learning, a multi-feature normalization method is adopted. First, a comprehensive test is conducted on the food raw materials to obtain the characteristics of the raw materials, and the parameters of the production process are collected, and the two are normalized into multi-dimensional features. The specific operation steps are to first calculate the mean and standard deviation of each feature, the mean: Standard Deviation: Among them, x ij represents the jth feature of the i-th sample, and N represents the number of samples. Secondly, based on the calculated mean and standard deviation, each feature is standardized. At the same time, the correlation between the features is considered to ensure that the normalization process of the data retains the inherent connection of the multidimensional features. The standardization formula is: where z ij is the standardized data; then weights are assigned according to the importance of different features to ensure that the features that have a greater impact on the final product quality have higher weights during the normalization process. The weighted normalization formula is: i ' j =z ij ·ω j Among them, ω j is the weight of the jth feature; finally, the weight is continuously updated after model training to ensure a positive improvement in predictive ability.

[0031] Then, considering that in some cases, ridge regression may not completely eliminate the impact of multicollinearity, the present invention optimizes the parameters through the particle swarm algorithm PSO, and then uses the cross-validation technique to select the optimal regularization parameter. Specifically, the first is the definition of the ridge regression loss function, which consists of two parts, namely the residual square and the regularization term in, is the residual square, is the regularization term, y i is the actual target value of the ith sample, x ij is the value of the i-th sample on the j-th feature, β0 is the intercept term, β j is the regression coefficient of the jth feature, λ is the regularization coefficient, which is used to control the complexity of the model; secondly, the particle swarm algorithm PSO is used to find the regression coefficient β with the minimum loss function j and regularization coefficient λ, the PSO update formula is in, represents the velocity of the i-th particle in the j-th dimension, Represents the position of the i-th particle in the j-th dimension, that is, the regression coefficient β j and λ, ω are inertia weights, which control the inertia of the particle. c1, c2 are learning factors. r1, r2 are random numbers between 0 and 1, which are used to increase the randomness of the search. represents the individual optimal position of the ith particle in the jth dimension, Represents the global optimal position of the jth dimension in the population; Next, perform cross-validation to evaluate the performance of the model, divide the training data into k non-overlapping subsets, use k-1 of them for model training in turn, and use the remaining one to verify the model performance, repeat k times, select a different validation set each time, and then calculate the mean square error of each validation, and take the average of all k validation results as the overall performance of the model; Finally, the mean square error obtained by cross-validation corresponds to different regularization coefficient λ values, and the regularization coefficient λ value with the smallest mean square error is selected as the optimal regularization parameter. After the optimization of the model, the quality index of the product is finally predicted. The product quality index is a score from 0 to 100. When the prediction result is greater than 85 points, it can be detected as a qualified product, otherwise it is unqualified.

[0032] After the above three groups of experimental samples were predicted by the model, the mean square error of each group was calculated. The experimental errors of the three results were 0.1524 for the first group, 0.1238 for the second group, and 0.3271 for the third group. The third group performed the worst. The reason for the analysis was that the material thickness increased and the partition spacing decreased. The final product quality index also showed this result. The product quality index of the first group was 88, which was qualified, the second group was 87, which was qualified, and the third group was 84, which was unqualified. When it was predicted that the product quality index of the third group of indicators was unqualified, the production parameters were adjusted. For the third group of experimental data, what needed to be improved was to reduce the material thickness and increase the partition spacing. Therefore, we kept other variables unchanged, adjusted the material thickness to 4.2mm, and increased the partition spacing to 10cm. The mean square error obtained by the prediction again was 0.1694, and the product quality index was predicted to be 88 points, which met the qualified standard.

[0033] Finally, in order to explore potential rules and trends, provide more accurate guidance for subsequent production, and solve the problem of difficulty in effectively using historical production data in traditional production control methods, a solution using big data analysis technology was adopted. Specifically, a large amount of historical production data was clustered, association rules mined, and trend analyzed through big data analysis technology. First, historical production data was clustered and similar data points were grouped together to better understand the distribution and characteristics of the data. Secondly, association rule mining was performed to find the potential relationship between process parameters and product quality. Finally, trend analysis was performed to discover the development trend of process parameters and product quality so as to adjust the production process in a timely manner. This operation can discover the potential relationship and trend between process parameters and product quality, provide accurate guidance for optimizing the production process, thereby improving product quality and production efficiency, and meeting the market demand for high-quality infant food.

[0034] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A production control method for infant freeze-dried food, characterized in that: The following steps are involved: S1. First, conduct a comprehensive test on the raw materials of infant freeze-dried food to obtain the characteristics of the raw materials, collect the parameters of the production process, and normalize the multi-dimensional features of the two; S2. Secondly, the particle swarm optimization (PSO) ridge regression algorithm is used to build a model, and the normalized raw material characteristics and production process are used to predict product quality indicators; S3, then the model performs real-time analysis and prediction of the product quality indicators based on the data, and adjusts the production process parameters in real time; S4. Finally, collect a large amount of production data and use big data analysis technology to explore potential rules and trends to provide more accurate guidance for subsequent production; In step S2, the particle swarm PSO is used to optimize the ridge regression algorithm model. The model is constructed by using ridge regression and the particle swarm algorithm PSO is used to optimize the parameters, and then the cross-validation technology is used to select the optimal regularization parameter λ. The specific implementation steps are: S21. First, the definition of the ridge regression loss function consists of two parts: the residual square and the regularization term in, is the residual square, is the regularization term, y i is the actual target value of the ith sample, x ij is the value of the i-th sample on the j-th feature, β0 is the intercept term, β j is the regression coefficient of the jth feature, and λ is the regularization coefficient, which is used to control the complexity of the model; S22. Secondly, use the particle swarm algorithm PSO to find the regression coefficient β with the minimum loss function j and regularization coefficient λ, the PSO update formula is in, represents the velocity of the i-th particle in the j-th dimension, Represents the position of the i-th particle in the j-th dimension, that is, the regression coefficient β j and λ, ω are inertia weights, which control the inertia of the particle. c1, c2 are learning factors. r1, r2 are random numbers between 0 and 1, which are used to increase the randomness of the search. represents the individual optimal position of the ith particle in the jth dimension, represents the global optimal position of the jth dimension in the population; S23, then perform cross-validation to evaluate the performance of the model, divide the training data into k non-overlapping subsets, use k-1 of them for model training in turn, and use the remaining one to verify the model performance, repeat k times, select a different validation set each time, and then calculate the mean square error of each validation, and take the average of all k validation results as the overall performance of the model; S24. Finally, the mean square error obtained through cross-validation corresponds to different regularization coefficient λ values, and the regularization coefficient λ value with the smallest mean square error is selected as the optimal regularization parameter.

2. The method for controlling the production of freeze-dried food for infants according to claim 1, characterized in that: The specific implementation steps of multi-dimensional feature normalization in step S1 are: S11. First calculate the mean and standard deviation of each feature, mean: Standard Deviation: Among them, x ij represents the jth feature of the i-th sample, and N represents the number of samples; S12. Next, based on the calculated mean and standard deviation, each feature is standardized. At the same time, the correlation between the features is considered to ensure that the normalization process of the data retains the inherent connection of the multidimensional features. The standardization formula is: where z ij It is the standardized data; S13. Then, weights are assigned according to the importance of different features to ensure that the features that have a greater impact on the final product quality have higher weights during the normalization process. The weighted normalization formula is: i ' j =z ij ·ω j Among them, ω j is the weight of the jth feature; S14. Finally, after the model is trained, the weights are continuously updated to ensure a positive improvement in the prediction capability.

3. The method for controlling the production of freeze-dried food for infants according to claim 1, characterized in that: The collected raw material characteristics include nutrient content, fiber content, freshness and microbial content.

4. The method for controlling the production of freeze-dried food for infants according to claim 1, characterized in that: The collected production process parameters include freezing rate, pre-freezing temperature, drying chamber pressure, heating plate temperature, drying time, material thickness and partition spacing.

5. The method for controlling the production of freeze-dried food for infants according to claim 1, characterized in that: The product quality index is a score from 0 to 100. A product is considered qualified only when the predicted result is greater than 85 points, otherwise it is considered unqualified.

6. The method for controlling the production of freeze-dried food for infants according to claim 1, characterized in that: The big data analysis technology used in step S4 is to cluster, mine association rules and analyze trends on a large amount of historical production data through big data analysis technology, discover the potential relationship and trend between process parameters and product quality, and provide accurate guidance for optimizing the production process.

Images