Feed online production intelligent monitoring method and system based on data analysis

Abstract

The invention discloses a feed online production intelligent monitoring method and system based on data analysis, and relates to the technical field of intelligent monitoring, original data RAW is collected and preprocessed to obtain an operation feature vector VPS, and a data foundation is laid for accurate analysis. Based on the operation feature vector VPS and a historical database HIS, a predicted quality feature vector VPQ containing specific indexes such as starch gelatinization degree Ysg is generated in real time, a comprehensive feed quality index SIFQ is calculated, a decision signal Sig is generated, multi-dimensional overall evaluation of product quality is achieved, and a clear basis is provided for regulation and control. And when the decision signal Sig indicates that the quality does not reach the standard, an adjustment instruction AIR is generated, and scientific and efficient process adjustment is ensured. And adjusting equipment according to an adjustment instruction AIR, continuously monitoring and updating a historical database HIS, and based on this, carrying out iterative optimization on an inherent coefficient set SCA, endowing the system with self-learning and continuous improvement capabilities, and ensuring long-term adaptability of the model and stable production optimization.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and in particular to a method and system for intelligent monitoring of online feed production based on data analysis. Background Art

[0002] Intelligent monitoring technology has expanded from the industrial sector to agriculture and the critical feed processing industry. As the foundation of animal husbandry, the quality of feed directly affects animal husbandry efficiency and food safety. Its production process is complex and variable, and each link has a significant impact on the quality of the finished product. Therefore, real-time, dynamic online monitoring of the entire production process has become an inevitable trend. However, traditional monitoring often finds it difficult to effectively utilize the massive, multi-dimensional data generated during the process for in-depth analysis and intelligent prediction. The industry urgently needs a technical solution that can transform data into insights and achieve intelligent online management and control. For example, in the production of different feeds such as broilers, aquaculture, or sows, ensuring accurate nutritional composition, excellent physical form, and the effectiveness of key nutrients poses specific application requirements and challenges for "online intelligent monitoring of feed production based on data analysis."

[0003] Although existing feed production processes have partially adopted automated equipment and certain online monitoring methods, such as monitoring basic process parameters like temperature, pressure, and motor current, several significant shortcomings and deficiencies remain when it comes to "specific things in specific areas"—that is, achieving "intelligent" online monitoring of feed intrinsic quality and complex process conditions. First, traditional online monitoring focuses on single, isolated process parameters and lacks the ability to provide real-time, dynamic predictions of key feed quality indicators influenced by multiple parameters. Quality assessment often relies on post-production laboratory inspections, a lag that makes it difficult to promptly identify and correct potential quality risks in the production process. Second, there is a lack of an effective mechanism that integrates multi-dimensional process parameters and predicted quality information to provide a comprehensive and dynamic assessment of the overall quality level of current feed production. Operators often struggle to comprehensively assess the overall process quality and potential issues based solely on discrete parameter readings. Third, when deviations in production parameters are detected or predicted quality is poor, existing adjustment methods often rely on operator experience or rigid control logic, making it difficult to quickly, accurately, and optimally adapt to raw material fluctuations, environmental changes, and multi-objective quality requirements. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a method and system for intelligent monitoring of online feed production based on data analysis, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a feed online production intelligent monitoring method based on data analysis, comprising the following steps:

[0006] S1. Collect raw data RAW from the production line through various sensors, perform data preprocessing on the raw data RAW and generate a running feature vector VPS;

[0007] S2. Based on the running feature vector VPS, combined with the historical database HIS, a key quality indicator prediction algorithm is constructed and a prediction quality feature vector VPQ is generated;

[0008] S3. Based on the predicted quality feature vector VPQ and the operating feature vector VPS, a comprehensive feed quality index algorithm is constructed to calculate the feed quality index SIFQ and compare it with the preset feed quality threshold SIFQ_th, and a decision signal Sig is generated according to the comparison result;

[0009] S4, execute the optimization instruction based on the decision signal Sig, calculate the quality index deviation column vector VQE between the predicted quality feature vector VPQ and the preset target quality feature vector VPQ_tgt, and build the influence matrix correction response model based on the historical database HIS to generate the adjustment instruction AIR;

[0010] S5. Adjust the equipment operating status based on the adjustment instruction AIR, continuously monitor and generate the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, upload and update the historical database HIS, and perform iterative optimization based on the historical database HIS.

[0011] Preferably, S1 includes S11;

[0012] S11. Collect raw data RAW from the production line through various sensors, clean the raw data RAW, remove outliers and check data integrity, fill in the data with the average value in the historical database HIS, time-align the cleaned raw data RAW, and generate a structured operational feature vector VPS;

[0013] The raw data RAW includes the material inlet moisture RMc, material crushing particle size distribution index RMs, conditioner steam jacket temperature CTp, conditioner moisture addition amount CMw, material average retention time CRt in the conditioner, pelletizer main motor current load PLd, pelletizer ring membrane outer surface temperature PDt and production line real-time production speed FLr;

[0014] The various sensors include near-infrared spectroscopy moisture analyzer, laser diffraction particle size analyzer, temperature detector, electromagnetic flowmeter, precision current transformer, non-contact infrared thermometer and electronic belt scale.

[0015] Preferably, S2 includes S21 and S22;

[0016] S21. Based on the running characteristic vector VPS and in combination with the historical database HIS, a key quality indicator prediction algorithm is constructed. The key quality indicator prediction algorithm includes a starch gelatinization prediction algorithm, a particle quality index prediction algorithm, and a lysine effective retention rate prediction algorithm. The starch gelatinization degree Ysg, the particle quality index Ypdi, and the lysine effective retention rate Ylaa are calculated. The inherent coefficient set SCA of the key quality indicator prediction algorithm is obtained through nonlinear regression analysis based on the historical database HIS. The obtained quality indicator values are aggregated to generate a predicted quality characteristic vector VPQ.

[0017] Among them, the intrinsic coefficient set SCA of the key quality indicator prediction algorithm is as follows:

[0018] Starch gelatinization degree prediction algorithm coefficients: tempering effect dominant coefficient Asg, raw material moisture negative effect coefficient Bsg, basic gelatinization degree Csg, excessive tempering temperature penalty factor Dsg and tempering intensity comprehensive response factor ks1;

[0019] Pellet quality index prediction algorithm coefficients: the dominant coefficient Apdi of the contribution of starch gelatinization degree Ysg to the pellet quality index Ypdi, the granulation pressure contribution coefficient Bdpi, the dominant coefficient Cdpi of the negative effect of raw material particle size, the basic particle size Ddpi, the penalty factor Edpi for excessive granulation pressure, the saturation adjustment factor kp1 of starch gelatinization degree Ysg, and the sensitivity factor kp2 for the negative effect of raw material particle size;

[0020] The coefficients of the lysine effective retention rate prediction algorithm are: lysine content benchmark Lint, the dominant factor of lysine degradation rate in the conditioning stage kl1, the dominant factor of lysine degradation rate in the granulation stage kl2 and the regulating factor of production rate on lysine retention kl3.

[0021] Preferably, S22, based on the current running characteristic vector VPS, combined with the starch gelatinization prediction algorithm coefficient in the inherent coefficient set SCA of the key quality indicator prediction algorithm, construct a starch gelatinization prediction algorithm to calculate the starch gelatinization degree Ysg;

[0022] Among them, the starch gelatinization degree prediction algorithm expression is as follows:

[0023] ;

[0024] ;

[0025] Where Psg represents the penalty term for the steam jacket temperature of the conditioner, CTp_opt represents a reference upper limit value of the preset steam jacket temperature CTp of the conditioner, max(a, b) represents the maximum value of a and b, and arctan represents the tangent function;

[0026] Based on the current running feature vector VPS and the particle quality index prediction algorithm coefficients in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a particle quality index prediction algorithm is constructed to calculate the particle quality index Ypdi;

[0027] Among them, the particle quality index prediction algorithm expression is as follows:

[0028] ;

[0029] ;

[0030] Where, Ppdi represents the current load penalty term of the pellet mill, PLd_opt represents a reference upper limit value of the preset pellet mill main motor current load PLd, ln represents the logarithmic function, and exp represents the natural exponential function;

[0031] Based on the current running feature vector VPS and the lysine effective retention rate prediction algorithm coefficients in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a lysine effective retention rate prediction algorithm is constructed to calculate the lysine effective retention rate Ylaa;

[0032] The lysine effective retention rate prediction algorithm expression is as follows:

[0033] ;

[0034] Where, 1+kl3*FLr represents the adjustment term of the production line's real-time production speed FLr on the production environment.

[0035] Preferably, S3 includes S31 and S32;

[0036] S31. Constructing a regional effective function term f(Y, L, U) based on the predicted quality feature vector VPQ, constructing a deviation cost penalty term Pdev based on the operating feature vector VPS, constructing a comprehensive feed quality index algorithm by combining the regional effective function term f(Y, L, U) and the deviation cost penalty term Pdev, calculating a feed quality index SIFQ, and comparing the feed quality index SIFQ with a preset feed quality threshold SIFQ_th;

[0037] If the feed quality index SIFQ ≥ the preset feed quality threshold SIFQ_th, the current feed comprehensive quality is deemed to meet the standard and a decision signal Sig is generated and assigned a value of False;

[0038] If the feed quality index SIFQ is less than the preset feed quality threshold SIFQ_th, it is determined that the current feed comprehensive quality does not meet the standard and a decision signal Sig is generated and assigned a value of True.

[0039] Preferably, S32, wherein the regional effective function term f (Y, L, U) is expressed as follows:

[0040] ;

[0041] Wherein, VPQ_sum represents the total number of elements of the prediction quality feature vector VPQ, i represents the sequence number of the i-th element in the prediction quality feature vector VPQ, f_sum(Y, L, U) represents the sum of the regional effective functions of the prediction quality feature vector VPQ, Yi represents the i-th element in the prediction quality feature vector VPQ, Li represents the preset lower limit value of the i-th element in the prediction quality feature vector VPQ, Ui represents the preset upper limit value of the i-th element in the prediction quality feature vector VPQ, C1_i represents the preset lower limit penalty coefficient of the i-th element in the prediction quality feature vector VPQ, and C2_i represents the preset upper limit penalty coefficient of the i-th element in the prediction quality feature vector VPQ;

[0042] Among them, the deviation cost penalty term Pdev is expressed as follows:

[0043] ;

[0044] Wherein, VPS_sum represents the total number of elements of the running feature vector VPS, j represents the sequence number of the j-th element in the running feature vector VPS, Pdev_sum represents the sum of the deviation cost penalties of the running feature vector VPS, Xj represents the j-th element in the running feature vector VPS, Xj_U represents the preset upper limit of the j-th element in the running feature vector VPS, Xj_L represents the preset lower limit of the j-th element in the running feature vector VPS, w1_j represents the preset upper limit penalty coefficient of the j-th element in the running feature vector VPS, and w2_j represents the preset lower limit penalty coefficient of the j-th element in the running feature vector VPS;

[0045] Among them, the algorithm expression of comprehensive feed quality index is as follows:

[0046] ;

[0047] Where δ represents the influencing factor of the real-time production rate FLr of the production line, and FLRbase represents the preset expected production rate.

[0048] Preferably, S4 includes S41 and S42;

[0049] S41, executing the optimization instruction based on the decision signal Sig:

[0050] If the decision signal Sig is False, it is determined that the current comprehensive quality of the feed meets the standard and there is no need to generate the adjustment instruction AIR;

[0051] If the decision signal Sig is Ture, it is determined that the current comprehensive quality of the feed does not meet the standard and an adjustment instruction AIR needs to be generated;

[0052] The adjustment instruction AIR is generated as follows:

[0053] By calculating the difference between the predicted quality characteristic vector VPQ and the preset target quality characteristic vector VPQ_tgt, the quality index deviation column vector VQE is obtained. According to the statistical relationship between the changes in the current running characteristic vector VPS in the historical database HIS and the changes in the predicted quality characteristic vector VPQ, an influence matrix correction response model is constructed to obtain the influence matrix Minf. Combined with the deviation importance weight diagonal matrix W_e, the original process parameter adjustment vector Cmd is calculated and the adjustment instruction AIR is generated.

[0054] Preferably, S42, wherein the influence matrix correction response model expression is as follows:

[0055] ;

[0056] Where ∂ represents the symbol of partial derivative;

[0057] Among them, the deviation importance weight diagonal matrix W_e is expressed as follows:

[0058] ;

[0059] Where we_sg represents the deviation adjustment weight of starch gelatinization degree Ysg set by the staff, we_pdi represents the deviation adjustment weight of particle quality index Ypdi set by the staff, and we_laa represents the deviation adjustment weight of lysine effective retention rate Ylaa set by the staff;

[0060] The calculation formula of the original process parameter adjustment vector is as follows:

[0061] .

[0062] Preferably, S5 includes S51;

[0063] S51, based on the adjustment instruction AIR, automatically adjust the production line operating parameters through the PLC interface, continuously monitor and generate the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, and upload and update the historical database HIS;

[0064] If the actual feed quality index AFQI ≥ the preset feed quality threshold SIFQ_th, a yes instruction is output;

[0065] If the actual feed quality index AFQI is less than the preset feed quality threshold SIFQ_th, a no instruction is output;

[0066] If the no instruction ratio is ≤98%, no iterative optimization is required;

[0067] If the proportion of no instructions is greater than 98%, the adjusted operating characteristic vector VPS_t1, actual quality characteristic vector VLQ and actual feed quality index AFQI of the production batch that outputs the no instruction need to be marked as unqualified data, and the inherent coefficient set SCA of the key quality indicator prediction algorithm should be adjusted based on the adjusted operating characteristic vector VPS_t1, actual quality characteristic vector VLQ and actual feed quality index AFQI of the production batch that outputs the yes instruction.

[0068] The feed online production intelligent monitoring system based on data analysis includes a production data acquisition and preprocessing module, an online quality prediction module, an evaluation and decision-making module, an adaptive process optimization module, and an iterative optimization module;

[0069] The production data acquisition and preprocessing module collects raw data RAW from the production line through various sensors, performs data preprocessing on the raw data RAW and generates a running feature vector VPS;

[0070] The online quality prediction module builds a key quality indicator prediction algorithm based on the running feature vector VPS and combines it with the historical database HIS to generate the predicted quality feature vector VPQ;

[0071] The evaluation and decision module constructs a comprehensive feed quality index algorithm based on the predicted quality feature vector VPQ and the operating feature vector VPS, calculates the feed quality index SIFQ and compares it with the preset feed quality threshold SIFQ_th, and generates a decision signal Sig according to the comparison result;

[0072] The adaptive process optimization module executes the optimization instruction based on the decision signal Sig, calculates the quality index deviation column vector VQE between the predicted quality feature vector VPQ and the preset target quality feature vector VPQ_tgt, and generates the adjustment instruction AIR by constructing the influence matrix correction response model based on the historical database HIS;

[0073] The iterative optimization module adjusts the equipment operating status based on the adjustment instruction AIR, continuously monitors and generates the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, and uploads and updates the historical database HIS, and performs iterative optimization based on the historical database HIS.

[0074] The present invention provides a method and system for intelligent monitoring of online feed production based on data analysis, which has the following beneficial effects:

[0075] (1) Through multi-dimensional data fusion and intelligent algorithm construction, real-time monitoring, quality prediction and adaptive optimization of the entire feed production process are achieved. This method integrates production line sensor data and historical databases, breaking through the limitations of isolated analysis of single parameters in traditional monitoring, and can dynamically capture the coupling relationship between process parameters and quality indicators. Through a closed-loop feedback mechanism, the system significantly improves the stability and controllability of the feed production process, reduces the risk of quality fluctuations, and reduces the need for manual intervention. Compared with the lagging means of relying on laboratory sampling, this invention realizes the transformation from "post-detection" to "online prediction and immediate correction", providing feed companies with an efficient and accurate intelligent management and control solution, helping to improve both production efficiency and product quality.

[0076] (2) Using sensors such as near-infrared spectroscopy moisture analyzer and laser diffraction particle size analyzer, raw data RAW, such as material inlet moisture RMc, conditioner steam jacket temperature CTp, and pelletizer main motor current load PLd, are collected in real time. After data cleaning and time alignment, an operating feature vector VPS is generated. Based on this, a key quality indicator prediction algorithm is constructed in combination with the historical database HIS to dynamically calculate starch gelatinization degree Ysg, pellet quality index Ypdi, and lysine effective retention rate Ylaa, and aggregate them into a predicted quality feature vector VPQ. Among them, Ysg quantifies the effect of the conditioning process on starch conversion through parameters such as the conditioning effect dominant coefficient Asg and the penalty factor Dsg. Ypdi accurately evaluates the pellet forming quality by integrating the granulation pressure contribution coefficient Bpdi and the raw material particle size negative effect Cdpi. Ylaa reflects the retention efficiency of key nutrients through the degradation rate factor kl1 and the production rate adjustment factor kl3. Based on the predicted quality feature vector VPQ and the operational feature vector VPS, a comprehensive feed quality index (SIFQ) algorithm was constructed. By dynamically weighting the regional effectiveness function f(Y, L, U) and the deviation cost penalty term Pdev, the SIFQ feed quality index was quantified and compared with the preset feed quality threshold SIFQ_th, generating a decision signal Sig. This algorithm not only achieves the coordinated evaluation of multiple quality indicators but also, by incorporating the influencing factor δ of the production line's real-time production rate FLr, balances production efficiency and quality, providing a scientific basis for subsequent optimization.

[0077] (3) By judging the decision signal Sig, when the decision signal Sig determines that the quality does not meet the standard, based on the deviation column vector VQE between the predicted quality feature vector VPQ and the target vector VPQ_tgt, combined with the influence matrix Minf and the deviation weight matrix We, an adjustment instruction AIR is generated to accurately correct key parameters such as the tempering temperature CTp and the retention time CRt. After the adjustment, the system continuously monitors the generated adjustment operation feature vector VPS_t1 and the actual quality feature vector VLQ, and updates the historical database HIS. The iterative optimization module dynamically optimizes the inherent coefficient set SCA of the key quality indicator prediction algorithm by analyzing the compliance rate between the actual feed quality index AFQI and the preset feed quality threshold SIFQ_th, ensuring that the model continues to adapt to raw material fluctuations and process changes. This closed-loop mechanism not only avoids the blindness of traditional experience adjustment, but also significantly improves the long-term stability and generalization performance of the system through data-driven self-learning capabilities, providing reliable technical support for the intelligent upgrade of feed production. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 This is a schematic diagram of the steps of the method for intelligent monitoring of online feed production based on data analysis of the present invention;

[0079] Figure 2 This is a schematic diagram of a block diagram of an intelligent monitoring system for online feed production based on data analysis according to the present invention;

[0080] Figure 3 This is a flow chart of the evaluation and decision-making module. DETAILED DESCRIPTION

[0081] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0082] Example 1

[0083] The present invention provides an intelligent monitoring method for online feed production based on data analysis. Figure 1 , including the following steps:

[0084] S1. Collect raw data RAW from the production line through various sensors, perform data preprocessing on the raw data RAW and generate a running feature vector VPS;

[0085] S2. Based on the running feature vector VPS, combined with the historical database HIS, a key quality indicator prediction algorithm is constructed and a prediction quality feature vector VPQ is generated;

[0086] S3. Based on the predicted quality feature vector VPQ and the operating feature vector VPS, a comprehensive feed quality index algorithm is constructed to calculate the feed quality index SIFQ and compare it with the preset feed quality threshold SIFQ_th, and a decision signal Sig is generated according to the comparison result;

[0087] S4, execute the optimization instruction based on the decision signal Sig, calculate the quality index deviation column vector VQE between the predicted quality feature vector VPQ and the preset target quality feature vector VPQ_tgt, and build the influence matrix correction response model based on the historical database HIS to generate the adjustment instruction AIR;

[0088] S5. Adjust the equipment operating status based on the adjustment instruction AIR, continuously monitor and generate the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, upload and update the historical database HIS, and perform iterative optimization based on the historical database HIS.

[0089] In this embodiment, the raw data RAW is precisely preprocessed to construct an operational characteristic vector VPS, laying a solid foundation for in-depth analysis. Combining the historical database HIS with the intrinsic coefficient set SCA, the predicted quality characteristic vector VPQ, which includes key indicators such as starch gelatinization degree Ysg, particle quality index Ypdi, and lysine effective retention rate Ylaa, is dynamically predicted online, thereby enabling advance quality judgment and effectively overcoming the lag of traditional monitoring. On this basis, the system calculates the comprehensive feed quality index SIFQ based on the predicted quality characteristic vector VPQ and the operational characteristic vector VPS, and generates a clear decision signal Sig by comparing it to the feed quality threshold SIFQ_th, providing a holistic and scientific basis for regulation and solving the problem of one-sided evaluation in the past. When the decision signal Sig indicates that adjustment is required, the adjustment instruction AIR can be intelligently generated based on models such as the quality indicator deviation column vector Err, the influence matrix Minf, and the weight W_e, replacing the empirical dependence and blindness of manual adjustment, significantly improving the accuracy and response efficiency of the adjustment. Ultimately, by executing the adjustment instruction AIR, tracking and adjusting data such as the operating characteristic vector VPS_t1, the actual quality characteristic vector VLQ, and the actual feed quality index AFQI, and using this data to update the historical database HIS, the inherent coefficient set SCA is then iteratively optimized, giving the system the core capabilities of self-learning and continuous evolution. This ensures that the monitoring and control model can adapt to raw material fluctuations and changes in operating conditions over the long term, achieving lasting stability and optimization of the production process, and addressing the fundamental flaw of traditional systems, which lack dynamic adaptation and continuous learning and improvement mechanisms.

[0090] Example 2

[0091] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: S1 includes S11;

[0092] S11. Collect raw data RAW from the production line through various sensors, clean the raw data RAW, remove outliers and check data integrity, fill in the data with the average value in the historical database HIS, time-align the cleaned raw data RAW, and generate a structured operational feature vector VPS;

[0093] The raw data RAW includes the material inlet moisture RMc, material crushing particle size distribution index RMs, conditioner steam jacket temperature CTp, conditioner moisture addition amount CMw, material average retention time CRt in the conditioner, pelletizer main motor current load PLd, pelletizer ring membrane outer surface temperature PDt and production line real-time production speed FLr;

[0094] A variety of sensors including near-infrared spectroscopy moisture analyzers, laser diffraction particle size analyzers, temperature detectors, electromagnetic flow meters, precision current transformers, non-contact infrared thermometers, and electronic belt scales;

[0095] S2 includes S21 and S22;

[0096] S21. Based on the running characteristic vector VPS and in combination with the historical database HIS, a key quality indicator prediction algorithm is constructed. The key quality indicator prediction algorithm includes a starch gelatinization prediction algorithm, a particle quality index prediction algorithm, and a lysine effective retention rate prediction algorithm. The starch gelatinization degree Ysg, the particle quality index Ypdi, and the lysine effective retention rate Ylaa are calculated. The inherent coefficient set SCA of the key quality indicator prediction algorithm is obtained through nonlinear regression analysis based on the historical database HIS. The obtained quality indicator values are aggregated to generate a predicted quality characteristic vector VPQ.

[0097] Among them, the intrinsic coefficient set SCA of the key quality indicator prediction algorithm is as follows:

[0098] Starch gelatinization degree prediction algorithm coefficients: tempering effect dominant coefficient Asg, raw material moisture negative effect coefficient Bsg, basic gelatinization degree Csg, excessive tempering temperature penalty factor Dsg and tempering intensity comprehensive response factor ks1;

[0099] Pellet quality index prediction algorithm coefficients: the dominant coefficient Apdi of the contribution of starch gelatinization degree Ysg to the pellet quality index Ypdi, the granulation pressure contribution coefficient Bdpi, the dominant coefficient Cdpi of the negative effect of raw material particle size, the basic particle size Ddpi, the penalty factor Edpi for excessive granulation pressure, the saturation adjustment factor kp1 of starch gelatinization degree Ysg, and the sensitivity factor kp2 for the negative effect of raw material particle size;

[0100] The coefficients of the lysine effective retention rate prediction algorithm are: lysine content benchmark Lint, lysine degradation rate leading factor kl1 in the conditioning stage, lysine degradation rate leading factor kl2 in the granulation stage, and production rate adjustment factor kl3 for lysine retention.

[0101] S22. Based on the current running characteristic vector VPS and in combination with the coefficients of the starch gelatinization prediction algorithm in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a starch gelatinization prediction algorithm is constructed to calculate the starch gelatinization degree Ysg;

[0102] Among them, the starch gelatinization degree prediction algorithm expression is as follows:

[0103] ;

[0104] ;

[0105] Where Psg represents the penalty term for the steam jacket temperature of the conditioner, CTp_opt represents a reference upper limit value of the preset steam jacket temperature CTp of the conditioner, max(a, b) represents the maximum value of a and b, and arctan represents the tangent function;

[0106] Based on the current running feature vector VPS and the particle quality index prediction algorithm coefficients in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a particle quality index prediction algorithm is constructed to calculate the particle quality index Ypdi;

[0107] Among them, the particle quality index prediction algorithm expression is as follows:

[0108] ;

[0109] ;

[0110] Where, Ppdi represents the current load penalty term of the pellet mill, PLd_opt represents a reference upper limit value of the preset pellet mill main motor current load PLd, ln represents the logarithmic function, and exp represents the natural exponential function;

[0111] Based on the current running feature vector VPS and the lysine effective retention rate prediction algorithm coefficients in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a lysine effective retention rate prediction algorithm is constructed to calculate the lysine effective retention rate Ylaa;

[0112] The lysine effective retention rate prediction algorithm expression is as follows:

[0113] ;

[0114] Where, 1+kl3*FLr represents the adjustment term of the production line's real-time production speed FLr on the production environment;

[0115] Specific example of calculating the predicted quality feature vector VPQ:

[0116] Running feature vector VPS={RMc: 0.6, CTp: 0.8, CRt: 0.4, CMw: 0.7, PLd: 0.55, RMs: 0.2, PDt: 0.9, FLr: 0.8};

[0117] The intrinsic coefficient set of the starch gelatinization degree prediction algorithm SCA(Ysg) = {Asg: 0.6, Bsg: 0.02, Csg: 75, Dsg: 0.01, ks1: 0.15, CTp_opt: 0.95};

[0118] The intrinsic coefficient set of the particle quality index prediction algorithm SCA(Ypdi) = {Apdi: 0.5, Bpdi: 0.2, Cpdi: 0.08, Dpid: 80, Edpi: 0.005, kp1: 0.12, kp2: 0.05, PLd_opt: 0.9};

[0119] The intrinsic coefficient set of the lysine effective retention rate prediction algorithm SCA (Ylaa) = {Lint: 90, kl1: 0.002, kl2: 0.001, kpl3: 0.01};

[0120] The starch gelatinization degree Ysg is calculated as follows:

[0121] ;

[0122] ;

[0123] The particle quality index Ypdi is calculated as follows:

[0124] ; ;

[0125] The effective lysine retention rate Ylaa is calculated as follows:

[0126] ;

[0127] Prediction quality feature vector VPQ={Ysg: 0.7495, Ypdi: 0.8122, Ylaa: 0.892}.

[0128] In this embodiment, raw data (RAW) such as material inlet moisture (RMc) and conditioner steam jacket temperature (CTp) collected by sensors such as a near-infrared spectroscopy moisture analyzer and a laser diffraction particle size analyzer are cleaned, outlier-removed, and padded with average values from a historical database (HIS) to generate a structured operational feature vector (VPS). This solves the problem of low raw data availability caused by noise interference or asynchronous acquisition timing, providing a highly reliable data foundation for subsequent analysis. Claims 3 and 4 further construct a key quality indicator prediction algorithm based on the operational feature vector (VPS), including parameters such as the dominant coefficient of the conditioning effect (Asg), the negative effect coefficient of raw material moisture (Bsg), and the penalty factor for excessive conditioning temperature (Dsg). This algorithm accurately quantifies the impact of the conditioning process on starch gelatinization (Ysg). Furthermore, dynamic parameters such as the granulation pressure contribution coefficient (Bpdi) and the negative effect of raw material particle size (Cdpi) enable real-time calculation of the pellet quality index (Ypdi). Furthermore, the lysine effective retention rate (Ylaa) algorithm, combined with the degradation rate factor (kl1) and the production rate adjustment factor (kl3), scientifically reflects the regulatory effect of the production rate (FLr) on the retention efficiency of key nutrients. These algorithms use nonlinear regression to analyze the historical database HIS and dynamically optimize the intrinsic coefficient set SCA to ensure that the prediction model accurately captures the coupling effects of process parameters, avoiding the one-sidedness of traditional single-parameter models and providing an explainable and highly accurate scientific basis for quality prediction and process adjustment.

[0129] Example 3

[0130] This embodiment is explained in Example 2, please refer to Figure 1 ,Specifically: S3 includes S31 and S32;

[0131] S31. Constructing a regional effective function term f(Y, L, U) based on the predicted quality feature vector VPQ, constructing a deviation cost penalty term Pdev based on the operating feature vector VPS, constructing a comprehensive feed quality index algorithm by combining the regional effective function term f(Y, L, U) and the deviation cost penalty term Pdev, calculating a feed quality index SIFQ, and comparing the feed quality index SIFQ with a preset feed quality threshold SIFQ_th;

[0132] If the feed quality index SIFQ ≥ the preset feed quality threshold SIFQ_th, the current feed comprehensive quality is deemed to meet the standard and a decision signal Sig is generated and assigned a value of False;

[0133] If the feed quality index SIFQ is less than the preset feed quality threshold SIFQ_th, the current feed comprehensive quality is deemed to be substandard and a decision signal Sig is generated and assigned a value of True;

[0134] S32, where the regional effective function term f(Y, L, U) is expressed as follows:

[0135] ;

[0136] Wherein, VPQ_sum represents the total number of elements of the prediction quality feature vector VPQ, i represents the sequence number of the i-th element in the prediction quality feature vector VPQ, f_sum(Y, L, U) represents the sum of the regional effective functions of the prediction quality feature vector VPQ, Yi represents the i-th element in the prediction quality feature vector VPQ, Li represents the preset lower limit value of the i-th element in the prediction quality feature vector VPQ, Ui represents the preset upper limit value of the i-th element in the prediction quality feature vector VPQ, C1_i represents the preset lower limit penalty coefficient of the i-th element in the prediction quality feature vector VPQ, and C2_i represents the preset upper limit penalty coefficient of the i-th element in the prediction quality feature vector VPQ;

[0137] Among them, the deviation cost penalty term Pdev is expressed as follows:

[0138] ;

[0139] Wherein, VPS_sum represents the total number of elements of the running feature vector VPS, j represents the sequence number of the j-th element in the running feature vector VPS, Pdev_sum represents the sum of the deviation cost penalties of the running feature vector VPS, Xj represents the j-th element in the running feature vector VPS, Xj_U represents the preset upper limit of the j-th element in the running feature vector VPS, Xj_L represents the preset lower limit of the j-th element in the running feature vector VPS, w1_j represents the preset upper limit penalty coefficient of the j-th element in the running feature vector VPS, and w2_j represents the preset lower limit penalty coefficient of the j-th element in the running feature vector VPS;

[0140] Among them, the algorithm expression of comprehensive feed quality index is as follows:

[0141] ;

[0142] Where δ represents the influencing factor of the real-time production rate FLr of the production line, and FLRbase represents the preset expected production rate.

[0143] In this example, based on specific indicators such as starch gelatinization degree Ysg, pellet quality index Ypdi, and lysine effective retention rate Ylaa in the predicted quality feature vector VPQ, combined with parameters such as the conditioner steam jacket temperature CTp and the pellet mill main motor current load PLd in the operational feature vector VPS, a comprehensive feed quality index (SIFQ) is scientifically calculated through the dynamic weighting of the regional effectiveness function term f(Y, L, U) and the deviation cost penalty term Pdev. The regional effectiveness function term f(Y, L, U) quantifies the degree of deviation of each element in the predicted quality feature vector VPQ based on the preset element lower limit Li, element upper limit Ui, element lower limit penalty coefficient C1_i, and element upper limit penalty coefficient C2_i, thereby avoiding the rigid limitations of single threshold judgment. The deviation cost penalty term Pdev applies a quadratic weighted penalty to the deviation of the upper and lower limits of each parameter in the operational feature vector VPS, thereby enhancing the impact of process stability on overall quality. The influencing factor δ of the production line's real-time production rate FLr was further introduced, and through a comprehensive feed quality index algorithm, the dynamic balance between production efficiency and quality objectives was incorporated into the evaluation system, resolving the issue of quality and capacity separation in traditional methods. Ultimately, the decision signal Sig was generated by intelligently comparing the feed quality index SIFQ with the preset feed quality threshold SIFQ_th. This provided a decision-making basis for process adjustments that took into account multi-dimensional constraints, significantly improving the comprehensiveness of quality assessments and the scientific nature of regulatory strategies.

[0144] Example 4

[0145] This embodiment is explained in Example 3, please refer to Figure 1 , specifically: S4 includes S41 and S42;

[0146] S41, executing the optimization instruction based on the decision signal Sig:

[0147] If the decision signal Sig is False, it is determined that the current comprehensive quality of the feed meets the standard and there is no need to generate the adjustment instruction AIR;

[0148] If the decision signal Sig is Ture, it is determined that the current comprehensive quality of the feed does not meet the standard and an adjustment instruction AIR needs to be generated;

[0149] The adjustment instruction AIR is generated as follows:

[0150] By calculating the difference between the predicted quality feature vector VPQ and the preset target quality feature vector VPQ_tgt, the quality indicator deviation column vector VQE is obtained. Based on the statistical relationship between the changes in the current operating feature vector VPS in the historical database HIS and the changes in the predicted quality feature vector VPQ, an influence matrix correction response model is constructed to obtain the influence matrix Minf. Combined with the deviation importance weight diagonal matrix W_e, the original process parameter adjustment amount vector Cmd is calculated and the adjustment instruction AIR is generated.

[0151] S42, where the influence matrix correction response model expression is as follows:

[0152] ;

[0153] Where ∂ represents the symbol of partial derivative;

[0154] Among them, the deviation importance weight diagonal matrix W_e is expressed as follows:

[0155] ;

[0156] Where we_sg represents the deviation adjustment weight of starch gelatinization degree Ysg set by the staff, we_pdi represents the deviation adjustment weight of particle quality index Ypdi set by the staff, and we_laa represents the deviation adjustment weight of lysine effective retention rate Ylaa set by the staff;

[0157] The calculation formula of the original process parameter adjustment vector is as follows:

[0158] ;

[0159] S5 includes S51;

[0160] S51, based on the adjustment instruction AIR, automatically adjust the production line operating parameters through the PLC interface, continuously monitor and generate the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, and upload and update the historical database HIS;

[0161] If the actual feed quality index AFQI ≥ the preset feed quality threshold SIFQ_th, a yes instruction is output;

[0162] If the actual feed quality index AFQI is less than the preset feed quality threshold SIFQ_th, a no instruction is output;

[0163] If the no instruction ratio is ≤98%, no iterative optimization is required;

[0164] If the proportion of no instructions is greater than 98%, the adjusted operating characteristic vector VPS_t1, actual quality characteristic vector VLQ and actual feed quality index AFQI of the production batch that outputs the no instruction need to be marked as unqualified data, and the inherent coefficient set SCA of the key quality indicator prediction algorithm should be adjusted based on the adjusted operating characteristic vector VPS_t1, actual quality characteristic vector VLQ and actual feed quality index AFQI of the production batch that outputs the yes instruction.

[0165] In this embodiment, by constructing a data-driven adaptive adjustment and iterative optimization mechanism, dynamic correction of the feed production process and continuous evolution of the model are achieved. When the decision signal Sig is judged to be True, the system constructs an influence matrix correction response model based on the quality index deviation column vector VQE of the predicted quality feature vector VPQ and the target quality feature vector VPQ_tgt, combined with the statistical relationship between process parameters and quality indicators in the historical database HIS. The influence matrix Minf is used to quantify the specific impact of the operation feature vector VPS on quality indicators such as starch gelatinization degree Ysg and particle quality index Ypdi, and the deviation importance weight diagonal matrix W_e is used to allocate adjustment priorities and generate adjustment instructions AIR. This mechanism breaks through the ambiguity of traditional manual experience adjustment and realizes the scientific calculation and precise allocation of process parameter adjustment amounts. The adjustment instruction AIR is further automatically executed through the PLC interface, and the adjusted adjustment operation feature vector VPS_t1 and the actual quality feature vector VLQ are generated in real time, and updated to the historical database HIS, providing real production data support for model iteration. Based on the comparison of the actual feed quality index (AFQI) with the preset threshold (SIFQ_th), unqualified batches are dynamically screened, retaining only those that meet quality requirements. This system optimizes the inherent coefficient set (SCA) of the key quality indicator prediction algorithm. This closed-loop iterative mechanism not only addresses the model failure issues caused by operating condition fluctuations in traditional systems, but also, through continuous learning from actual production feedback, enables the model to adapt to changes in raw material properties and equipment aging, significantly improving the system's long-term robustness and process optimization efficiency.

[0166] Example 5

[0167] For the feed online production intelligent monitoring system based on data analysis, please refer to Figure 2 ,Specifically: including production data acquisition and preprocessing module, online quality prediction module, evaluation and decision module, adaptive process optimization module and iterative optimization module;

[0168] The production data acquisition and preprocessing module collects raw data RAW from the production line through various sensors, performs data preprocessing on the raw data RAW and generates a running feature vector VPS;

[0169] The online quality prediction module builds a key quality indicator prediction algorithm based on the running feature vector VPS and combines it with the historical database HIS to generate the predicted quality feature vector VPQ;

[0170] The evaluation and decision module constructs a comprehensive feed quality index algorithm based on the predicted quality feature vector VPQ and the operating feature vector VPS, calculates the feed quality index SIFQ and compares it with the preset feed quality threshold SIFQ_th, and generates a decision signal Sig according to the comparison result;

[0171] The adaptive process optimization module executes the optimization instruction based on the decision signal Sig, calculates the quality index deviation column vector VQE between the predicted quality feature vector VPQ and the preset target quality feature vector VPQ_tgt, and generates the adjustment instruction AIR by constructing the influence matrix correction response model based on the historical database HIS;

[0172] The iterative optimization module adjusts the equipment operating status based on the adjustment instruction AIR, continuously monitors and generates the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, and uploads and updates the historical database HIS, and performs iterative optimization based on the historical database HIS.

[0173] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring method for online feed production based on data analysis, characterized by: The following steps are involved: S1. Collect raw data RAW from the production line through various sensors, perform data preprocessing on the raw data RAW and generate a running feature vector VPS; S2, based on the running feature vector VPS, combined with the historical database HIS to build a key quality indicator prediction algorithm and generate a prediction quality feature vector VPQ; S3. Based on the predicted quality feature vector VPQ and the operating feature vector VPS, a comprehensive feed quality index algorithm is constructed to calculate the feed quality index SIFQ and compare it with the preset feed quality threshold SIFQ_th, and a decision signal Sig is generated according to the comparison result; S4, execute the optimization instruction based on the decision signal Sig, calculate the quality index deviation column vector VQE between the predicted quality feature vector VPQ and the preset target quality feature vector VPQ_tgt, and build the influence matrix correction response model based on the historical database HIS to generate the adjustment instruction AIR; S5. Adjust the equipment operating status based on the adjustment instruction AIR, continuously monitor and generate the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, upload and update the historical database HIS, and perform iterative optimization based on the historical database HIS.

2. The method for online intelligent monitoring of feed production based on data analysis according to claim 1, characterized in that: S1 includes S11; S11. Collect raw data RAW from the production line through various sensors, clean the raw data RAW, remove outliers and check data integrity, fill in the data with the average value in the historical database HIS, time-align the cleaned raw data RAW, and generate a structured operational feature vector VPS; The raw data RAW includes the material inlet moisture RMc, material crushing particle size distribution index RMs, conditioner steam jacket temperature CTp, conditioner moisture addition amount CMw, material average retention time CRt in the conditioner, pelletizer main motor current load PLd, pelletizer ring membrane outer surface temperature PDt and production line real-time production speed FLr; The various sensors include near-infrared spectroscopy moisture analyzer, laser diffraction particle size analyzer, temperature detector, electromagnetic flowmeter, precision current transformer, non-contact infrared thermometer and electronic belt scale.

3. The method for online intelligent monitoring of feed production based on data analysis according to claim 2, characterized in that: S2 includes S21 and S22; S21. Based on the running characteristic vector VPS and in combination with the historical database HIS, a key quality indicator prediction algorithm is constructed. The key quality indicator prediction algorithm includes a starch gelatinization prediction algorithm, a particle quality index prediction algorithm, and a lysine effective retention rate prediction algorithm. The starch gelatinization degree Ysg, the particle quality index Ypdi, and the lysine effective retention rate Ylaa are calculated. The inherent coefficient set SCA of the key quality indicator prediction algorithm is obtained through nonlinear regression analysis based on the historical database HIS. The obtained quality indicator values are aggregated to generate a predicted quality characteristic vector VPQ. Among them, the intrinsic coefficient set SCA of the key quality indicator prediction algorithm is as follows: Starch gelatinization degree prediction algorithm coefficients: tempering effect dominant coefficient Asg, raw material moisture negative effect coefficient Bsg, basic gelatinization degree Csg, excessive tempering temperature penalty factor Dsg and tempering intensity comprehensive response factor ks1; Pellet quality index prediction algorithm coefficients: the dominant coefficient Apdi of the contribution of starch gelatinization degree Ysg to the pellet quality index Ypdi, the granulation pressure contribution coefficient Bdpi, the dominant coefficient Cdpi of the negative effect of raw material particle size, the basic particle size Ddpi, the penalty factor Edpi for excessive granulation pressure, the saturation adjustment factor kp1 of starch gelatinization degree Ysg, and the sensitivity factor kp2 for the negative effect of raw material particle size; The coefficients of the lysine effective retention rate prediction algorithm are: lysine content benchmark Lint, the dominant factor of lysine degradation rate in the conditioning stage kl1, the dominant factor of lysine degradation rate in the granulation stage kl2 and the regulating factor of production rate on lysine retention kl3.

4. The method for online intelligent monitoring of feed production based on data analysis according to claim 3, characterized in that: S22. Based on the current running characteristic vector VPS and in combination with the coefficients of the starch gelatinization prediction algorithm in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a starch gelatinization prediction algorithm is constructed to calculate the starch gelatinization degree Ysg; Among them, the starch gelatinization degree prediction algorithm expression is as follows: ； ； Where Psg represents the penalty term for the steam jacket temperature of the conditioner, CTp_opt represents a reference upper limit value of the preset steam jacket temperature CTp of the conditioner, max(a, b) represents the maximum value of a and b, and arctan represents the tangent function; Based on the current running feature vector VPS and the particle quality index prediction algorithm coefficients in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a particle quality index prediction algorithm is constructed to calculate the particle quality index Ypdi; Among them, the particle quality index prediction algorithm expression is as follows: ； ； Where, Ppdi represents the current load penalty term of the pellet mill, PLd_opt represents a reference upper limit value of the preset pellet mill main motor current load PLd, ln represents the logarithmic function, and exp represents the natural exponential function; Based on the current running feature vector VPS and the lysine effective retention rate prediction algorithm coefficients in the inherent coefficient set SCA of the key quality indicator prediction algorithm, a lysine effective retention rate prediction algorithm is constructed to calculate the lysine effective retention rate Ylaa; The lysine effective retention rate prediction algorithm expression is as follows: ； Where, 1+kl3*FLr represents the adjustment term of the production line's real-time production speed FLr on the production environment.

5. The method for online intelligent monitoring of feed production based on data analysis according to claim 4, characterized in that: S3 includes S31 and S32; S31. Constructing a regional effective function term f(Y, L, U) based on the predicted quality feature vector VPQ, constructing a deviation cost penalty term Pdev based on the operating feature vector VPS, constructing a comprehensive feed quality index algorithm by combining the regional effective function term f(Y, L, U) and the deviation cost penalty term Pdev, calculating a feed quality index SIFQ, and comparing the feed quality index SIFQ with a preset feed quality threshold SIFQ_th; If the feed quality index SIFQ ≥ the preset feed quality threshold SIFQ_th, the current feed comprehensive quality is deemed to meet the standard and a decision signal Sig is generated and assigned a value of False; If the feed quality index SIFQ is less than the preset feed quality threshold SIFQ_th, it is determined that the current comprehensive feed quality does not meet the standard and a decision signal Sig is generated and assigned a value of True.

6. The method for online intelligent monitoring of feed production based on data analysis according to claim 5, characterized in that: S32, where the regional effective function term f(Y, L, U) is expressed as follows: ； Wherein, VPQ_sum represents the total number of elements of the prediction quality feature vector VPQ, i represents the sequence number of the i-th element in the prediction quality feature vector VPQ, f_sum(Y, L, U) represents the sum of the regional effective functions of the prediction quality feature vector VPQ, Yi represents the i-th element in the prediction quality feature vector VPQ, Li represents the preset lower limit value of the i-th element in the prediction quality feature vector VPQ, Ui represents the preset upper limit value of the i-th element in the prediction quality feature vector VPQ, C1_i represents the preset lower limit penalty coefficient of the i-th element in the prediction quality feature vector VPQ, and C2_i represents the preset upper limit penalty coefficient of the i-th element in the prediction quality feature vector VPQ; Among them, the deviation cost penalty term Pdev is expressed as follows: ； Wherein, VPS_sum represents the total number of elements of the running feature vector VPS, j represents the sequence number of the j-th element in the running feature vector VPS, Pdev_sum represents the sum of the deviation cost penalties of the running feature vector VPS, Xj represents the j-th element in the running feature vector VPS, Xj_U represents the preset upper limit of the j-th element in the running feature vector VPS, Xj_L represents the preset lower limit of the j-th element in the running feature vector VPS, w1_j represents the preset upper limit penalty coefficient of the j-th element in the running feature vector VPS, and w2_j represents the preset lower limit penalty coefficient of the j-th element in the running feature vector VPS; Among them, the algorithm expression of comprehensive feed quality index is as follows: ； Where δ represents the influencing factor of the real-time production rate FLr of the production line, and FLRbase represents the preset expected production rate.

7. The method for online intelligent monitoring of feed production based on data analysis according to claim 6, characterized in that: S4 includes S41 and S42; S41, executing the optimization instruction based on the decision signal Sig: If the decision signal Sig is False, it is determined that the current comprehensive quality of the feed meets the standard and there is no need to generate the adjustment instruction AIR; If the decision signal Sig is Ture, it is determined that the current comprehensive quality of the feed does not meet the standard and an adjustment instruction AIR needs to be generated; The adjustment instruction AIR is generated as follows: By calculating the difference between the predicted quality characteristic vector VPQ and the preset target quality characteristic vector VPQ_tgt, the quality index deviation column vector VQE is obtained. According to the statistical relationship between the changes in the current running characteristic vector VPS in the historical database HIS and the changes in the predicted quality characteristic vector VPQ, an influence matrix correction response model is constructed to obtain the influence matrix Minf. Combined with the deviation importance weight diagonal matrix W_e, the original process parameter adjustment vector Cmd is calculated and the adjustment instruction AIR is generated.

8. The method for online intelligent monitoring of feed production based on data analysis according to claim 7, characterized in that: S42, where the influence matrix correction response model expression is as follows: ； Where ∂ represents the symbol of partial derivative; Among them, the deviation importance weight diagonal matrix W_e is expressed as follows: ； Where we_sg represents the deviation adjustment weight of starch gelatinization degree Ysg set by the staff, we_pdi represents the deviation adjustment weight of particle quality index Ypdi set by the staff, and we_laa represents the deviation adjustment weight of lysine effective retention rate Ylaa set by the staff; The calculation formula of the original process parameter adjustment vector is as follows: 。 9. The method for online intelligent monitoring of feed production based on data analysis according to claim 8, characterized in that: S5 includes S51; S51, based on the adjustment instruction AIR, automatically adjust the production line operating parameters through the PLC interface, continuously monitor and generate the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, and upload and update the historical database HIS; If the actual feed quality index AFQI ≥ the preset feed quality threshold SIFQ_th, a yes instruction is output; If the actual feed quality index AFQI is less than the preset feed quality threshold SIFQ_th, a no instruction is output; If the no instruction ratio is ≤98%, no iterative optimization is required; If the proportion of no instructions is greater than 98%, the adjusted operating characteristic vector VPS_t1, actual quality characteristic vector VLQ and actual feed quality index AFQI of the production batch that outputs the no instruction need to be marked as unqualified data, and the inherent coefficient set SCA of the key quality indicator prediction algorithm should be adjusted based on the adjusted operating characteristic vector VPS_t1, actual quality characteristic vector VLQ and actual feed quality index AFQI of the production batch that outputs the yes instruction.

10. A feed online production intelligent monitoring system based on data analysis, applied to the feed online production intelligent monitoring method based on data analysis according to any one of claims 1 to 9, characterized in that: It includes production data acquisition and preprocessing module, online quality prediction module, evaluation and decision module, adaptive process optimization module and iterative optimization module; The production data acquisition and preprocessing module collects raw data RAW from the production line through various sensors, performs data preprocessing on the raw data RAW and generates a running feature vector VPS; The online quality prediction module builds a key quality indicator prediction algorithm based on the running feature vector VPS and combines it with the historical database HIS to generate the predicted quality feature vector VPQ; The evaluation and decision module constructs a comprehensive feed quality index algorithm based on the predicted quality feature vector VPQ and the operating feature vector VPS, calculates the feed quality index SIFQ and compares it with the preset feed quality threshold SIFQ_th, and generates a decision signal Sig according to the comparison result; The adaptive process optimization module executes the optimization instruction based on the decision signal Sig, calculates the quality index deviation column vector VQE between the predicted quality feature vector VPQ and the preset target quality feature vector VPQ_tgt, and generates the adjustment instruction AIR by constructing the influence matrix correction response model based on the historical database HIS; The iterative optimization module adjusts the equipment operating status based on the adjustment instruction AIR, continuously monitors and generates the adjusted operation characteristic vector VPS_t1, the actual quality characteristic vector VLQ and the actual feed quality index AFQI, and uploads and updates the historical database HIS, and performs iterative optimization based on the historical database HIS.

Images