Raw material quality intelligent screening optimization system

Abstract

The invention relates to the technical field of quality control, in particular to a raw material quality intelligent screening optimization system, which comprises a preliminary feature screening module, a raw material detection database, a sample humidity, protein content and fat proportion data extraction module, and a mean value and a standard deviation of each item of data calculation module. According to the method, the humidity, the protein content and the fat proportion of the raw materials are finely screened and analyzed, the key features influencing the product quality are identified, the feature weight is dynamically adjusted, and real-time optimization is performed according to the performance and the error of the features in prediction, so that the reaction speed and the accuracy of the model are improved; abnormal data points in a data set are cleared, data quality is optimized, processing efficiency and prediction accuracy are further improved, features are adjusted and recombined through real-time production data, feature configuration is made to better meet actual production requirements, prediction stability and accuracy are enhanced through a multi-model fusion technology, and prediction efficiency is improved. Therefore, the overall production efficiency and the product quality are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of quality control, and in particular to an intelligent screening and optimization system for raw material quality. Background Art

[0002] The field of quality control technology is a broad technology that involves multiple industrial and commercial fields, focusing on ensuring that the quality of products and services meets or exceeds predetermined standards. In modern industrial production and service provision, quality control systems use automation and intelligent technologies such as sensor technology, data analysis, and machine learning algorithms to improve detection efficiency and accuracy. These technologies can monitor various parameters in the production process in real time, automatically detect product defects, unqualified raw materials, and other problems, and provide decision support to reduce scrap rates and improve production efficiency. The development of intelligent quality control systems not only optimizes production processes, but also helps companies save costs and remain competitive through predictive maintenance.

[0003] Among them, the intelligent screening and optimization system for pet food raw material quality refers to the use of intelligent technology to detect and optimize the quality of raw materials in pet food production. The main purpose of this system is to automate the raw material detection process through advanced algorithms and sensor technology to ensure that all raw materials used in the production of pet food meet safety and quality standards. This system can reduce human errors, speed up the screening process, and more accurately evaluate the quality of raw materials through continuous monitoring and real-time data analysis, thereby improving overall production efficiency and product quality. This is crucial for the pet food industry because high-quality raw materials directly affect the safety and nutritional value of the final product.

[0004] Existing technologies are deficient in data processing efficiency and model adaptability. Despite a high degree of automation, data utilization efficiency and deep mining capabilities are limited. When raw materials or the environment of a production line change, traditional models take a long time to adapt to these changes, affecting production efficiency. In addition, existing systems may not be accurate enough when processing outlier data, resulting in noise in the data set, which directly affects the accuracy and reliability of the prediction model. These problems may lead to increased production costs, unstable product quality, and damage to the company's market competitiveness and brand reputation. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an intelligent screening and optimization system for raw material quality.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: A raw material quality intelligent screening and optimization system comprises:

[0007] The preliminary feature screening module extracts the moisture, protein content, and fat ratio data of the samples from the raw material testing database, calculates the mean and standard deviation of each data, and screens the features with a coefficient of variation higher than the set threshold by calculating the coefficient of variation of multiple data to generate a screening feature set;

[0008] The dynamic weight adjustment module adjusts the weights of the features in the screening feature set, updates the weights according to the prediction errors of the features and their contributions to the results, optimizes the model response speed through weight adjustment, and generates optimized feature weights;

[0009] The abnormal data identification module uses the optimized feature weights to identify and separate outlier samples in the data set, remove outlier data points, and generate a screening data set;

[0010] The feature optimization and integration module uses the screening data set to re-evaluate and combine features, adjust feature criticality and combination methods according to real-time production data, and generate an integrated feature configuration;

[0011] The quality prediction integration module performs quality prediction based on the integrated feature configuration, adopts multi-model fusion technology, evaluates the prediction results of each model, dynamically adjusts the prediction weight according to the performance of each model, and generates quality control prediction values.

[0012] The screening feature set includes moisture, protein content, and fat ratio. The optimized feature weight specifically refers to weight adjustment and model response speed optimization. The screening data set includes the remaining data after removing outlier data points. The integrated feature configuration includes feature criticality and feature combination method. The quality control prediction value specifically includes multi-model fusion technology, model prediction result evaluation, and dynamic adjustment of prediction weight.

[0013] As a further solution of the present invention, the step of obtaining the screening feature set is specifically:

[0014] Extract moisture data, protein content data and fat ratio data from the raw material testing database to generate original data records;

[0015] The mean and standard deviation of each item in the original data record are calculated using the formula:

[0016]

[0017] and

[0018]

[0019] Calculate the mean record and standard deviation record;

[0020] Among them, x irepresents a single data point, n represents the total number of data points, μ represents the mean, and σ represents the standard deviation;

[0021] Based on the mean and standard deviation records, the coefficient of variation formula is used:

[0022]

[0023] Calculate the coefficient of variation for each data item and generate a record of the coefficient of variation;

[0024] Among them, α is the adjustment coefficient, which is used to refer to the square effect of the standard deviation, and CV stands for the coefficient of variation, which is a statistical indicator used to measure the degree of dispersion of data points;

[0025] Data with a coefficient of variation higher than a preset threshold value are screened from the coefficient of variation records, features with high variability are determined, and a screening feature set is generated.

[0026] As a further solution of the present invention, the step of obtaining the optimized feature weights is specifically as follows:

[0027] Extracting the prediction error and contribution of each feature to the model result from the screening feature set to generate error data and contribution data;

[0028] Based on the error data and contribution data, the formula is adopted:

[0029]

[0030] Calculate and generate updated weights;

[0031] Among them, w new,i represents the updated weight of the i-th feature, which is the new weight calculated after adjusting the prediction error and contribution of the feature. i represents the prediction error of the i-th feature, c i represents the contribution of the i-th feature, λ is a regularization parameter, γ is a proportional adjustment coefficient, max(c j ) Select the maximum value among all feature contributions;

[0032] Applying the updated weights to the adjustment of the model, performing simulation runs, evaluating the model response speed, and generating performance evaluation records;

[0033] According to the data in the performance evaluation record, the weights of multiple features in the model are adjusted, feature criticality analysis is performed, the contribution of multiple features to model performance is identified, and optimized feature weights are generated.

[0034] As a further solution of the present invention, the step of obtaining the screening data set is specifically:

[0035] Using the optimized feature weights, the weight of each sample in the data set is scored, and a sample weight score list is generated by calculating the weighted scores;

[0036] According to the sample weight score list, the threshold is set using the formula:

[0037]

[0038] Generate a list of standardized weight scores by referring to the impact of the distribution of overall weights on the weights of individual samples;

[0039] Among them, S i represents the standardized weight score of the i-th sample, w i represents the weight of the i-th sample, n is the total number of samples, w j Represents the weight of the jth sample;

[0040] Based on the standardized weight score list, by comparing the standardized weight score of each sample with a set score threshold, identifying outlier samples whose scores are lower than the threshold, and generating an outlier sample list;

[0041] Based on the outlier sample list, all identified outlier samples are removed from the data set, and other samples are retained to generate a screening data set.

[0042] As a further solution of the present invention, the step of acquiring the integrated feature configuration is specifically:

[0043] Extracting real-time production data representation and statistical attributes of each feature from the screening data set to generate feature analysis results;

[0044] Based on the feature analysis results, the contribution and criticality of each feature are re-evaluated using the formula:

[0045]

[0046] Generate feature criticality adjustment results;

[0047] Among them, R f Represents the re-evaluation score of the feature, which is used to measure the criticality of the feature after considering the variability, f i is the original contribution of the i-th feature, reflecting the direct influence of the feature on the production process, v i is the coefficient of variation of the characteristic, which measures the consistency or stability of the characteristic performance in differentiated production batches, and n is the total number of characteristics, which indicates the number of characteristics referenced in the analysis;

[0048] The data in the feature criticality adjustment result is used to optimize the feature combination, and feature selection, feature crossover, and feature expansion are performed according to real-time data. The data is screened and reorganized based on its relevance and redundancy to generate an integrated feature configuration.

[0049] As a further solution of the present invention, the step of obtaining the quality control prediction value is specifically as follows:

[0050] Based on the integrated feature configuration, multi-model fusion technology is applied to perform quality prediction, and preliminary model performance indicators are obtained by comparing the performance of multiple models on the training data set;

[0051] Based on the preliminary model performance indicators, the prediction results of each model are analyzed, the accuracy and error are calculated, and the contribution weight of each model is calculated according to the accuracy and error, using the formula:

[0052]

[0053] Generate model weight results;

[0054] Among them, W m Represents the weight of model m, which is used to determine the contribution of the model in prediction. m represents the prediction error of model m, λ is a positive adjustment coefficient used to adjust the influence of the error on the weight, n represents the total number of models, E i represents the prediction error of the ith model;

[0055] The model weight results are used to dynamically adjust the impact of multiple models on the predicted value, and the optimal model combination is used to predict product quality to generate quality control prediction values.

[0056] Compared with the prior art, the advantages and positive effects of the present invention are:

[0057] In the present invention, by carefully screening and analyzing the key quality indicators of raw materials such as moisture, protein content and fat ratio, the key features that affect product quality are accurately identified. The feature weights are dynamically adjusted, and real-time optimization is performed based on the performance and error of the features in the prediction, thereby improving the response speed and accuracy of the model. Abnormal data points in the data set are removed, data quality is optimized, and processing efficiency and prediction accuracy are further improved. Real-time production data is used to adjust and recombine features, so that the feature configuration is more in line with actual production needs, and the stability and accuracy of the prediction are enhanced through multi-model fusion technology, thereby improving overall production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a system flow chart of the present invention;

[0059] Figure 2 A flow chart of the steps for obtaining the screening feature set of the present invention;

[0060] Figure 3 A flow chart of the steps for obtaining the optimized feature weights of the present invention;

[0061] Figure 4 A flow chart of the steps for obtaining a screening data set for the present invention;

[0062] Figure 5 A flow chart of the acquisition steps for integrating feature configurations of the present invention;

[0063] Figure 6 Flow chart of the steps for obtaining the quality control prediction value of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0065] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0066] Embodiment 1

[0067] See also Figure 1 , a raw material quality intelligent screening and optimization system includes:

[0068] The preliminary feature screening module extracts the moisture, protein content, and fat ratio data of the samples from the raw material testing database, calculates the mean and standard deviation of each data, and screens the features with a coefficient of variation higher than the set threshold by calculating the coefficient of variation of multiple data to generate a screening feature set;

[0069] The dynamic weight adjustment module adjusts the weights of the features in the screening feature set, updates the weights according to the prediction error of the features and their contribution to the results, optimizes the model response speed through weight adjustment, and generates optimized feature weights;

[0070] The abnormal data identification module uses optimized feature weights to identify and separate outlier samples in the data set, remove outlier data points, and generate a filtered data set;

[0071] The feature optimization and integration module uses the screening data set to re-evaluate and combine features, adjusts feature criticality and combination methods based on real-time production data, and generates integrated feature configurations;

[0072] The quality prediction integration module performs quality prediction based on integrated feature configuration, adopts multi-model fusion technology, evaluates the prediction results of each model, dynamically adjusts the prediction weight according to the performance of each model, and generates quality control prediction values.

[0073] The screening feature set includes moisture, protein content, and fat ratio. Optimizing feature weights specifically refers to weight adjustment and model response speed optimization. The screening data set includes the remaining data after removing outlier data points. The integrated feature configuration includes feature criticality and feature combination methods. The quality control prediction value specifically includes multi-model fusion technology, model prediction result evaluation, and dynamic adjustment of prediction weights.

[0074] See also Figure 2 , the specific steps for obtaining the screening feature set are:

[0075] Extract moisture data, protein content data and fat ratio data from the raw material testing database to generate original data records;

[0076] The process of extracting moisture data, protein content data, and fat ratio data from the raw material testing database involves not only database query operations, but also preliminary screening of data quality to ensure the accuracy and reliability of the data, check the original data in the database, eliminate incomplete data or outliers, and verify the integrity and consistency of the data to ensure the accuracy of subsequent calculations. Through this series of data processing steps, the original data records are generated.

[0077] Calculate the mean and standard deviation of each data item in the original data record using the formula:

[0078]

[0079] and

[0080]

[0081] Calculate the mean record and standard deviation record;

[0082] Among them, x i represents a single data point, n represents the total number of data points, μ represents the mean, and σ represents the standard deviation;

[0083] Calculating the mean and standard deviation involves adding and averaging the values ​​of each data point to obtain the mean. Calculating the standard deviation requires adding the square of the difference between the value of each data point and the mean, and then dividing by the number of data points minus one to obtain the degree of dispersion of the data. These statistical operations ensure the assessment of data fluctuations and dispersion. Through these steps, the mean and standard deviation of each data type are calculated to generate mean and standard deviation records.

[0084] Based on the mean and standard deviation records, the coefficient of variation formula is used:

[0085]

[0086] Calculate the coefficient of variation for each data item and generate a record of the coefficient of variation;

[0087] Among them, α is the adjustment coefficient, which is used to refer to the square effect of the standard deviation, and CV stands for the coefficient of variation, which is a statistical indicator used to measure the degree of dispersion of data points;

[0088] formula:

[0089]

[0090] The benefit of the formula is that by referring to the influence of the standard deviation and its square on the mean, the sensitivity and descriptive accuracy of the coefficient of variation are enhanced, so that the formula can more effectively identify and distinguish data features with differentiated volatility, and is suitable for capturing abnormal or significantly variable features in the non-uniform distribution of data features.

[0091] Detailed explanation of the formula and the process of formula calculation and derivation:

[0092] Set the data point set to x 1 , x 2 , …, x n , the mean μ is calculated as The standard deviation σ is calculated as The initial value of the adjustment coefficient α is set to 0.1 (this initial value is obtained based on the fluctuation analysis of previous data, and this value can be adjusted according to actual data later). i =

[0093] 45, 55, 50, 60, 40, calculated mean μ = 50, standard deviation σ ≈ 7.07, then the coefficient of variation CV:

[0094]

[0095] The results show that the obtained coefficient of variation reflects the relative volatility and dispersion between data points, and the numerical results can effectively guide the subsequent data analysis process and identify data features with high variability.

[0096] Data with coefficient of variation higher than a preset threshold are screened from the coefficient of variation records to determine features with high variability and generate a screening feature set.

[0097] Screening data with a coefficient of variation higher than a preset threshold includes comparing each coefficient of variation value with the threshold. The threshold is set based on the fluctuation of the data and the actual application requirements. The 75% quantile of the coefficient of variation is taken as the threshold. In this way, those data features that statistically show a high degree of volatility or instability are identified and selected, thereby determining features with high variability and generating a screening feature set.

[0098] See also Figure 3 , the specific steps for obtaining the optimized feature weights are:

[0099] Extract the prediction error and contribution of each feature to the model result from the screened feature set to generate error data and contribution data;

[0100] Error data includes the error value of each feature in the model prediction, and contribution data includes the contribution ratio of each feature to the model results. These data are obtained through statistical analysis and data mining techniques. The prediction error is calculated by monitoring the difference between the model output and the actual value. The contribution of multiple features is evaluated by analyzing the impact of multi-feature weights on changes in model output. Data analysts will use data to adjust model parameters, optimize the overall performance of the model, and ensure that the weight distribution of features reasonably reflects their true impact on the model results.

[0101] Based on the error data and contribution data, the formula is adopted:

[0102]

[0103] Calculate and generate updated weights;

[0104] Among them, w new,i represents the updated weight of the i-th feature, which is the new weight calculated after adjusting the prediction error and contribution of the feature. i represents the prediction error of the i-th feature, c i represents the contribution of the i-th feature, λ is a regularization parameter, γ is a proportional adjustment coefficient, max(c j ) Select the maximum value among all feature contributions;

[0105] formula:

[0106]

[0107] The benefit of the formula is that by introducing the regularization parameter λ and the normalized contribution γ, the model's sensitivity to features with small feature errors but large contributions is enhanced, while the weights of features with large errors are reduced, thereby improving the accuracy of predictions while maintaining model stability.

[0108] Detailed explanation of the formula and the process of formula calculation and derivation:

[0109] Set λ=0.05,γ=1.2, and set the prediction error e of feature i i = 0.03, the contribution of feature i c i =0.4, the maximum contribution of all features max(c j )=0.5, then:

[0110] 1. Calculate the error adjustment item:

[0111]

[0112] 2. Calculate contribution adjustment items:

[0113]

[0114] Then multiply by γ to get

[0115] 0.8×1.2=0.96

[0116] 3. Calculation

[0117] w new,i =12.5+0.96=13.46

[0118] The results show that through the above calculation, the new weight of feature i is 13.46, which reflects the criticality of feature i to the model under given error and contribution. The weight will be used in the model to adjust its contribution to the results and improve the prediction accuracy.

[0119] Apply the updated weights to the model adjustments, run simulations, evaluate the model response speed, and generate performance evaluation records;

[0120] Through dynamic simulation and performance monitoring technology, during the simulation operation, the response time and processing speed of the model under the new weights are recorded, and the impact of the new weights on the model efficiency is analyzed. The performance record data is obtained through real-time data capture and post-data analysis, ensuring the accuracy of the results and the feasibility of the operation. The key in this process is to compare the performance data before and after the model adjustment, and analyze how the new weights affect the response speed and accuracy of the model.

[0121] According to the data in the performance evaluation record, the weights of multiple features in the model are adjusted, feature criticality analysis is performed, the contribution of multiple features to model performance is identified, and optimized feature weights are generated.

[0122] According to the data in the performance evaluation record, it is necessary to first adjust the weights of multiple features of the model, analyze the contribution of each feature to the model output, identify the most critical features, and make weighted adjustments on this basis. First, conduct a preliminary evaluation of all features, collect and organize data from the performance evaluation records to ensure the accuracy and completeness of the data. On this basis, use information gain and chi-square tests to quantify the contribution of each feature. In this process, the criticality of the feature can be determined by calculating the impact on the prediction accuracy of the model. If the correlation between a feature and the prediction target is high, its weight should be relatively increased, and if the correlation is low, its weight should be relatively reduced. Next, use the weighted average method to combine the contribution of multiple features with the corresponding weights, readjust the feature input of the model, and obtain the optimized feature weights. According to the impact of differentiated features on model performance, the adjusted weights can also be verified and optimized through methods such as cross-validation to ensure that the model has improved in multiple performance indicators.

[0123] See also Figure 4 , the specific steps for obtaining the screening data set are:

[0124] By optimizing the feature weights, the weight of each sample in the data set is scored, and a sample weight score list is generated by calculating the weighted scores;

[0125] In the process of scoring samples using optimized feature weights, we first need to obtain the feature data of each sample in the data set. These data are collected by various sensors or log files, and then weighted processing is performed on the multiple features of each sample, where the weight value is based on the optimized feature weight determined by data analysis in the previous step. The key to this step is how to apply these optimized feature weights to the sample features. The weight score of each sample is obtained by multiplying its feature value with the corresponding weight value and summing them up. This process ensures that the weight score of each sample can truly reflect its criticality in the data set. The generated sample weight score list is formed by arranging the weighted scores of each sample.

[0126] According to the sample weight score list, set the threshold using the formula:

[0127]

[0128] Generate a list of standardized weight scores by referring to the impact of the distribution of overall weights on the weights of individual samples;

[0129] Among them, S irepresents the standardized weight score of the i-th sample, w i represents the weight of the i-th sample, n is the total number of samples, w j Represents the weight of the jth sample;

[0130] formula:

[0131]

[0132] The benefit of the formula is that by introducing square root and absolute value operations, this method refers to the collective influence of all sample weights, so that even when individual weight values ​​are extremely high or low, it can effectively perform normalization, thereby ensuring the fairness and stability of the calculation.

[0133] Detailed explanation of the formula and the process of formula calculation and derivation:

[0134] Suppose there is a small data set with a weight distribution of: w = [0.2, 0.3, 0.5]. The steps to calculate the standardized weight score of each sample are as follows:

[0135] 1. Calculate the square root of the sum of the absolute values ​​of the weights:

[0136]

[0137] 2. Calculate the standardized weight score for each sample:

[0138]

[0139] These calculations show the weight of each sample relative to the population.

[0140] The results show that the weight score can effectively reflect the relative criticality of each sample in the entire dataset, where samples with high weight scores are key data points, while samples with low scores are outliers or less critical monitoring values.

[0141] Based on the standardized weight score list, by comparing the standardized weight score of each sample with the set score threshold, outlier samples with scores lower than the threshold are identified to generate an outlier sample list;

[0142] Based on the list of standardized weight scores, a threshold is set to determine which samples are considered outliers. This threshold is set according to the characteristics and distribution of the data set. A common method is to set the threshold to the lower 10% or lower 25% of the score to screen outliers that have a negative impact on model performance. This screening process involves comparing the score of each sample with the threshold and marking those samples below the threshold as outliers. The generated list of outlier samples provides a basis for data cleaning and preprocessing, ensuring the quality of the data and the accuracy of the model.

[0143] Based on the list of outlier samples, all identified outlier samples are removed from the data set, and other samples are retained to generate a screened data set.

[0144] After removing the identified outlier samples, the remaining samples constitute the cleaned data set, which will be used for iterative analysis and model training. The cleaning process not only refers to the weight score of the sample, but also involves other data quality indicators, including completeness, consistency and relevance to the current analysis task. This step ensures that each sample in the data set contributes to the modeling and decision-making process, so that the generated screening data set has high reliability and representativeness, laying a solid foundation for subsequent data analysis and machine learning model training.

[0145] See also Figure 5 , the steps for obtaining the integrated feature configuration are as follows:

[0146] Extract real-time production data performance and statistical attributes of each feature from the screened data set to generate feature analysis results;

[0147] When extracting features from the screened data set, the real-time performance and statistical properties of multiple features are analyzed. This process involves the classification and preprocessing of batch data, where data preprocessing includes missing value processing, outlier identification and replacement. These operations ensure the accuracy and applicability of the data and provide basic data support for generating reliable feature analysis results. Through these detailed data processing steps, the feature analysis results obtained can accurately reflect the performance of each feature in actual production and its statistical characteristics.

[0148] Based on the feature analysis results, re-evaluate the contribution and criticality of each feature using the formula:

[0149]

[0150] Generate feature criticality adjustment results;

[0151] Among them, R f Represents the re-evaluation score of the feature, which is used to measure the criticality of the feature after considering the variability, f i is the original contribution of the i-th feature, reflecting the direct influence of the feature on the production process, v i is the coefficient of variation of the characteristic, which measures the consistency or stability of the characteristic performance in differentiated production batches, and n is the total number of characteristics, which indicates the number of characteristics referenced in the analysis;

[0152] formula:

[0153]

[0154] The benefit of the formula is that it re-evaluates the criticality of the feature by combining the contribution and coefficient of variation of the feature, so that the weight of the feature can be adjusted more reasonably by referring not only to the contribution of the feature but also to its stability.

[0155] Detailed explanation of the formula and the process of formula calculation and derivation:

[0156] There are three features, and their contribution is f 1 =0.5, f 2 =0.3, f 3 =0.2, and the coefficients of variation are v 1 =0.1, v 2 =0.3, v 3 =0.6, then:

[0157]

[0158] The results show that after comprehensively considering the contribution and coefficient of variation of the features, a more scientific feature criticality score is obtained, which helps to more accurately select features and assign weights when facing actual production data.

[0159] Use the data in the feature criticality adjustment results to optimize feature combinations, perform feature selection, feature crossover, and feature expansion based on real-time data, screen and reorganize data based on relevance and redundancy, and generate integrated feature configurations.

[0160] In the process of feature criticality adjustment, the data set is first analyzed in multiple dimensions to determine the criticality of each feature in the target task, and feature selection is performed in combination with real-time data, which involves extracting representative and information-rich features from the data set. First, feature screening is performed, and redundant and irrelevant features are eliminated using statistical analysis, information gain, and chi-square test methods. Then, the remaining features are cross-combined to generate current features, and the expression ability of features is improved through corresponding feature engineering techniques. The key to feature cross-combination and feature expansion is to capture the matching combination mode, and to combine two independent features into one current feature using second-order cross features, thereby capturing more complex nonlinear relationships. In this process, the data standardization and normalization techniques used ensure that features of differentiated scales can be reasonably compared within the same framework. For redundancy processing, correlation analysis can be used to eliminate highly correlated features, which can avoid the generation of multicollinearity problems and thus improve the prediction ability of the model. By cyclically adjusting and optimizing these feature combinations, a feature set that can effectively reflect the inherent laws in the data is constructed.

[0161] See also Figure 6 , the specific steps for obtaining the quality control prediction value are:

[0162] Based on the integrated feature configuration, multi-model fusion technology is used for quality prediction. By comparing the performance of multiple models on the training data set, preliminary model performance indicators are obtained;

[0163] In the process of quality prediction based on multi-model fusion technology with integrated feature configuration, the performance of multiple models on the training data set is comprehensively evaluated. This step relies on a data-driven approach. By collecting and organizing quality data, the performance indicators of each model, including accuracy and response time, are calculated and sorted to determine which models perform better and are suitable for inclusion in the fusion model. This method can maximize the use of existing data resources and improve the reliability and accuracy of the prediction model without incurring additional costs.

[0164] Based on the preliminary model performance indicators, the prediction results of each model are analyzed, the accuracy and error are calculated, and the contribution weight of each model is calculated according to the accuracy and error, using the formula:

[0165]

[0166] Generate model weight results;

[0167] Among them, W m Represents the weight of model m, which is used to determine the contribution of the model in prediction. m represents the prediction error of model m, λ is a positive adjustment coefficient used to adjust the influence of the error on the weight, n represents the total number of models, E i represents the prediction error of the ith model;

[0168] formula:

[0169]

[0170] The benefit of the formula is that by introducing the exponential decay factor λ and the model error E m , effectively adjusting the weight distribution of multiple models during prediction, so that models with smaller errors have higher weights, thereby improving the accuracy of the overall prediction.

[0171] Detailed explanation of the formula and the process of formula calculation and derivation:

[0172] The prediction errors of the three models are set to 0.1, 0.2, and 0.3 respectively, and λ is set to 1. The weights of multiple models are calculated according to the formula:

[0173]

[0174] The results show that model 1, due to its smallest error, obtains the highest weight and thus dominates the prediction results, which reflects the strategy of optimizing model output by dynamically adjusting weights.

[0175] Using the model weight results, dynamically adjust the impact of multiple models on the predicted value, use the optimal model combination to predict product quality, and generate quality control prediction values.

[0176] The process of dynamically adjusting the weights of multiple models to optimize quality predictions includes monitoring real-time production data, analyzing changes in quality control parameters at multiple data points, comparing these real-time data with previous data patterns, and using the calculated weight results to update the prediction model to ensure that the model output always reflects the current production status and quality trends. In addition, the system can automatically identify models that need to be retrained and prompt model retraining or parameter adjustment to maintain the efficient operation and high accuracy of the entire prediction system.

[0177] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A raw material quality intelligent screening and optimization system, characterized in that: The system comprises: The preliminary feature screening module extracts the moisture, protein content, and fat ratio data of the samples from the raw material testing database, calculates the mean and standard deviation of each data, and screens the features with a coefficient of variation higher than the set threshold by calculating the coefficient of variation of multiple data to generate a screening feature set; The dynamic weight adjustment module adjusts the weights of the features in the screening feature set, updates the weights according to the prediction errors of the features and their contributions to the results, optimizes the model response speed through weight adjustment, and generates optimized feature weights; The abnormal data identification module uses the optimized feature weights to identify and separate outlier samples in the data set, remove outlier data points, and generate a screening data set; The feature optimization and integration module uses the screening data set to re-evaluate and combine features, adjust feature criticality and combination methods according to real-time production data, and generate an integrated feature configuration; The quality prediction integration module performs quality prediction based on the integrated feature configuration, adopts multi-model fusion technology, evaluates the prediction results of each model, dynamically adjusts the prediction weight according to the performance of each model, and generates quality control prediction values.

2. The intelligent screening and optimization system for raw material quality according to claim 1 is characterized in that: The screening feature set includes moisture, protein content, and fat ratio. The optimized feature weight specifically refers to weight adjustment and model response speed optimization. The screening data set includes the remaining data after removing outlier data points. The integrated feature configuration includes feature criticality and feature combination method. The quality control prediction value specifically includes multi-model fusion technology, model prediction result evaluation, and dynamic adjustment of prediction weight.

3. The intelligent screening and optimization system for raw material quality according to claim 2 is characterized in that: The steps of obtaining the screening feature set are specifically as follows: Extract moisture data, protein content data and fat ratio data from the raw material testing database to generate original data records; The mean and standard deviation of each item in the original data record are calculated using the formula: and Calculate the mean record and standard deviation record; Among them, x i represents a single data point, n represents the total number of data points, μ represents the mean, and σ represents the standard deviation; Based on the mean and standard deviation records, the coefficient of variation formula is used: Calculate the coefficient of variation for each data item and generate a record of the coefficient of variation; Among them, α is the adjustment coefficient, which is used to refer to the square effect of the standard deviation, and CV stands for the coefficient of variation, which is a statistical indicator used to measure the degree of dispersion of data points; Data with a coefficient of variation higher than a preset threshold value are screened from the coefficient of variation records, features with high variability are determined, and a screening feature set is generated.

4. The raw material quality intelligent screening and optimization system according to claim 3 is characterized in that: The steps for obtaining the optimized feature weights are specifically as follows: Extracting the prediction error and contribution of each feature to the model result from the screening feature set to generate error data and contribution data; Based on the error data and contribution data, the formula is adopted: Calculate and generate updated weights; Among them, w new,i represents the updated weight of the i-th feature, which is the new weight calculated after adjusting the prediction error and contribution of the feature. i represents the prediction error of the i-th feature, c i represents the contribution of the i-th feature, λ is a regularization parameter, γ is a proportional adjustment coefficient, max(c j ) Select the maximum value among all feature contributions; Applying the updated weights to the adjustment of the model, performing simulation runs, evaluating the model response speed, and generating performance evaluation records; According to the data in the performance evaluation record, the weights of multiple features in the model are adjusted, feature criticality analysis is performed, the contribution of multiple features to model performance is identified, and optimized feature weights are generated.

5. The intelligent screening and optimization system for raw material quality according to claim 4 is characterized in that: The steps of obtaining the screening data set are specifically as follows: Using the optimized feature weights, the weight of each sample in the data set is scored, and a sample weight score list is generated by calculating the weighted scores; According to the sample weight score list, the threshold is set using the formula: Generate a list of standardized weight scores by referring to the impact of the distribution of overall weights on the weights of individual samples; Among them, S i represents the standardized weight score of the i-th sample, w i represents the weight of the i-th sample, n is the total number of samples, w j Represents the weight of the jth sample; Based on the standardized weight score list, by comparing the standardized weight score of each sample with a set score threshold, identifying outlier samples whose scores are lower than the threshold, and generating an outlier sample list; Based on the outlier sample list, all identified outlier samples are removed from the data set, and other samples are retained to generate a screening data set.

6. The intelligent screening and optimization system for raw material quality according to claim 5, characterized in that: The steps of obtaining the integrated feature configuration are specifically as follows: Extracting real-time production data representation and statistical attributes of each feature from the screening data set to generate feature analysis results; Based on the feature analysis results, the contribution and criticality of each feature are re-evaluated using the formula: Generate feature criticality adjustment results; Among them, R f Represents the re-evaluation score of the feature, which is used to measure the criticality of the feature after considering the variability, f i is the original contribution of the i-th feature, reflecting the direct influence of the feature on the production process, v i is the coefficient of variation of the characteristic, which measures the consistency or stability of the characteristic performance in differentiated production batches, and n is the total number of characteristics, which indicates the number of characteristics referenced in the analysis; The data in the feature criticality adjustment result is used to optimize the feature combination, and feature selection, feature crossover, and feature expansion are performed according to real-time data. The data is screened and reorganized according to its relevance and redundancy to generate an integrated feature configuration.

7. The intelligent screening and optimization system for raw material quality according to claim 6, characterized in that: The steps for obtaining the quality control prediction value are specifically as follows: Based on the integrated feature configuration, multi-model fusion technology is applied to perform quality prediction, and preliminary model performance indicators are obtained by comparing the performance of multiple models on the training data set; Based on the preliminary model performance indicators, the prediction results of each model are analyzed, the accuracy and error are calculated, and the contribution weight of each model is calculated according to the accuracy and error, using the formula: Generate model weight results; Among them, W m Represents the weight of model m, which is used to determine the contribution of the model in prediction. m represents the prediction error of model m, λ is a positive adjustment coefficient used to adjust the influence of the error on the weight, n represents the total number of models, E i represents the prediction error of the ith model; The model weight results are used to dynamically adjust the impact of multiple models on the predicted value, and the optimal model combination is used to predict product quality to generate quality control prediction values.

Images