CN-121998479-A - Pig feed production quality tracing method based on cloud computing

CN121998479ACN 121998479 ACN121998479 ACN 121998479ACN-121998479-A

Abstract

The invention provides a pig feed production quality tracing method based on cloud computing, which comprises the steps of whole-process data acquisition and structuring of multisource heterogeneous process parameters and quality detection indexes, sliding window segmentation normalization and time alignment processing, causal relation recognition based on Granger causal inspection and information transfer entropy, dynamic optimization causal map based on improved time perception structural equation model, abnormal triggering inverse fact reasoning for process attribution contribution quantification, and completion of semantic annotation and causal path compression by combining an expert rule knowledge base to generate a visual tracing report.

Inventors

WU HAIKUN
CHENG JINFEI
CHI SHUHONG
LI WENBING
HU XIAOYAN
Hong Aishao
HUANG SIQI
LI YUZHEN

Assignees

广州市江丰生物科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251226

Claims (10)

1. A pig feed production quality tracing method based on cloud computing is characterized by comprising the following steps: S1, collecting technological parameters and quality detection indexes of all links of raw material pretreatment, mixing and proportioning and high-temperature granulation in the whole feed production process, wherein the technological parameters comprise moisture content time sequence data, a temperature gradient curve and a mixing uniformity discrete value, and the quality detection indexes comprise a finished product protein content deviation rate and a particle hardness discrete coefficient; S2, performing self-adaptive filtering and time alignment processing on the acquired multi-source heterogeneous data, and carrying out sectional normalization on technological parameters by adopting a sliding window strategy to generate a standardized time sequence data set containing process characteristic labels; s3, based on Granger causal test and information transfer entropy combined criteria, causal strength calculation is carried out on variable pairs in the standardized time sequence data set, and an initial causal topological structure is constructed; S4, adopting an improved time perception structural equation model to dynamically optimize the initial causal topological structure, and updating causal edge weights in real time through a variable dB leaf inference method to generate a causal map sequence which changes along with production working conditions; s5, when the quality deviation rate of the finished product exceeds a preset threshold value, starting a counterfactual reasoning mechanism, constructing a virtual intervention matrix based on the causal map sequence, calculating a hypothetical intervention effect, and quantifying the attribution contribution degree of each procedure to the final abnormality; s6, inputting the causal path with the attribution contribution degree exceeding a preset significance level into an expert rule knowledge base, executing semantic annotation and path compression operation, and generating a visual traceability report containing abnormal propagation time sequence characteristics, key nodes and propagation intensity.
2. The method for tracing the production quality of the pig feed based on the cloud computing according to claim 1, wherein the step S1 specifically comprises: Acquiring raw material moisture content time sequence data and a temperature gradient curve based on a moisture sensor and a temperature sensor of a raw material pretreatment link; a weighing system and a mixing uniformity detecting device based on a mixing and proportioning link collect discrete value data of the mixing uniformity; a temperature controller and a pressure sensor based on a high-temperature granulating step collect temperature change curve and pressure fluctuation data in the granulating process; based on finished product quality detection equipment, acquiring a deviation rate of finished product protein content and a particle hardness discrete coefficient; And generating an original data set containing a process characteristic label based on the multi-source isomerization process parameters containing the raw material moisture content time sequence data, the temperature gradient curve and the mixing uniformity discrete value and the quality detection index containing the finished product protein content deviation rate and the particle hardness discrete coefficient.
3. The pig feed production quality tracing method based on cloud computing as claimed in claim 2, wherein the raw material moisture content time sequence data and temperature gradient curve are collected by a moisture sensor and a temperature sensor, the moisture data are collected by an analog-to-digital conversion collection module and pass through cyclic redundancy check, the moisture data are filtered and noise reduced by a sliding average value, the temperature data are fitted by a second order polynomial to generate a temperature gradient curve, and the raw material state is quantitatively evaluated by a process stability criterion.
4. The method for tracing the production quality of the pig feed based on the cloud computing according to claim 1, wherein the step S2 specifically comprises: Performing missing value interpolation and outlier rejection operation on the acquired multi-source heterogeneous process parameter data, and performing preliminary cleaning on the moisture content time sequence data, the temperature gradient curve and the mixing uniformity discrete value of each procedure based on a sliding window median filtering algorithm to obtain a cleaned process parameter data sequence; calculating sliding window length parameters based on characteristic time constants of all working procedures, and generating a window configuration parameter set bound with the types of the working procedures; Performing segmentation normalization processing on the cleaned process parameter data sequence by utilizing the window configuration parameter set, and performing dimensionless transformation on the moisture content, the temperature gradient and the mixing uniformity discrete value of each procedure based on a normalization method to obtain a normalization process parameter subsequence; performing time alignment operation on the standardized process parameter subsequence, mapping multiple working procedures under different acquisition frequencies to a unified time reference based on a time stamp matching algorithm, and performing time compensation on unsynchronized sampling points by adopting a linear interpolation method to obtain a working procedure characteristic data set with consistent time axis; and executing feature label labeling operation on the process feature data set, adding a process type label, a time window number and a feature dimension identifier for each process data segment based on a process rule database, and generating a structured standardized time sequence data set.
5. The method for tracing the production quality of the pig feed based on the cloud computing according to claim 1, wherein the step S3 specifically comprises: Executing Granger causal test on the process quality characteristic variable pairs in the standardized time sequence data set, modeling based on lag correlation of a time sequence, calculating linear causal strength among the variables, and identifying process node pairs with obvious linear causal relation; Based on Granger cause and effect test results, carrying out nonlinear cause and effect intensity calculation on the process node pairs by adopting an information transfer entropy method, and evaluating nonlinear dependency relationship among variables by utilizing the difference value of mutual information and conditional mutual information; carrying out weighted fusion on the Granger cause and effect test result and the information transfer entropy result, and generating a comprehensive cause and effect intensity index based on a dynamic weight distribution strategy; Constructing an initial causal topological structure based on the comprehensive causal intensity index to form a directed weighted causal map; and carrying out sparsification treatment on the directed weighted causal map, and removing weak connecting edges with causal strength lower than a set threshold by adopting a threshold pruning strategy.
6. The method for tracing the production quality of pig feed based on cloud computing of claim 5, wherein said initial causal topological structure comprises using process quality characteristics as graph nodes, causal intensity as edge weights, and determining edge direction according to time lag relationship between variables.
7. The method for tracing the production quality of the pig feed based on the cloud computing according to claim 1, wherein the step S4 specifically comprises: Based on the latest quality detection data and process parameter time sequence data in the sliding window, initializing node variables and structure paths of a time-aware structural equation model, and establishing a model initial state; Performing variable decibel leaf-based inference processing on the structure path coefficient in the time-aware structural equation model, and performing approximate posterior distribution estimation on model parameters by using a standardized time sequence data set in a sliding window to obtain an optimal causal edge weight under the current working condition; performing parameter iterative updating on the time-aware structural equation model based on a variation free energy minimization criterion, and performing successive optimization on each path coefficient by adopting a coordinate ascending method until the path coefficients are converged to a local optimal solution to obtain a dynamic causal map reflecting the current production state; binding the updated causal edge weight with the time stamp to generate a causal map sequence with a time tag; And carrying out graph structure stability assessment on the causal graph sequence, calculating graph evolution trend indexes based on graph node degree distribution and edge weight change rate, triggering a causal structure relearning mechanism when the evolution trend indexes exceed a preset fluctuation threshold, re-executing Granger causal inspection and information transfer entropy combined criteria, updating initial causal topology and feeding back to the time perception structural equation model.
8. The method for tracing the production quality of the pig feed based on cloud computing of claim 7, wherein in the variational Bayesian inference process, an average field approximation strategy is adopted in an inference process, and the joint distribution is decomposed into independent approximate distributions of each path coefficient.
9. The method for tracing the production quality of the pig feed based on the cloud computing according to claim 1, wherein the step S5 specifically comprises: Constructing a virtual intervention matrix based on dynamic causal edge weights among nodes in the causal map sequence; Performing a hypothetical intervention operation on each row in the virtual intervention matrix, setting an upstream process variable value corresponding to the row as a reference state, and performing forward propagation simulation based on a causal graph structure to obtain an expected quality output of a downstream node under the intervention condition; calculating the difference value between the expected mass output and the mass deviation rate actually detected to obtain abnormal alleviation effect quantity under each intervention path; Based on the anomaly mitigation effect quantity, carrying out weighted attribution calculation on each intervention path by combining path length and propagation delay information in the causal map, and generating an attribution contribution vector comprising propagation path weights and causal influence intensities; And executing significance test on the attribution contribution degree vector, screening out process nodes with contribution degree exceeding a preset significance threshold and propagation paths thereof, and forming a high-influence abnormal propagation path set.
10. The pig feed production quality tracing method based on cloud computing of claim 9, wherein in the virtual intervention matrix, matrix rows represent upstream process nodes, columns represent downstream process nodes, and matrix elements are normalized propagation intensities of corresponding causal edges.

Description

Pig feed production quality tracing method based on cloud computing Technical Field The invention relates to the technical field of quality tracing and intelligent diagnosis in a feed production process, in particular to a pig feed production quality tracing method based on cloud computing. Background Along with the continuous improvement of the breeding scale and the automation degree of the animal husbandry, the quality management and the traceability of the whole feed production process become core technical links for guaranteeing the food safety and the product quality of the breeding links. The current industry generally adopts a method based on data acquisition and statistical analysis to cover the flow links of raw material pretreatment, mixing and proportioning, high-temperature granulation, finished product detection and the like. The main flow technical scheme comprises the steps of acquiring process parameters (such as moisture, temperature, mixing uniformity and the like) and finished product quality indexes in real time by utilizing an Internet of things sensor and automatic detection equipment, realizing data integration and standardization by a cloud platform or a local area network system, and further carrying out abnormal problem identification and tracing analysis by adopting a correlation analysis, statistical regression modeling, simple rule diagnosis or graph structured tracing mode. Partial enterprises or research teams try to introduce modeling means such as Bayesian networks, graphic neural networks and the like to support association analysis and propagation path mining among multi-level procedures, but the core logic is still mainly based on variable correlation, and lacks in-depth expression and interpretable inference of causality; Existing feed production quality traceability systems typically emphasize complete collection of information streams and data-driven anomaly detection. For example, the production process is tracked by adopting modes such as quality information coding, step identification, batch tracing and the like, so that later-stage problem batch positioning and quality management responsibility division are facilitated. However, in a specific abnormal propagation path identification modeling link, multi-dependency correlation analysis and unidirectional causal attribution cannot reveal the real influence path of each process parameter abnormal in a cross-flow and cross-layer stage, and it is also difficult to effectively distinguish the direct causal driving and indirect association relation between variables under complex working conditions. Partial solutions attempt to model transition probabilities among process variables through a Bayesian network, but face technical bottlenecks such as high preset rigidity of a model structure, difficult real-time dynamic adaptation of parameter updating, insufficient topology inference capability of multi-task complex data and the like. In addition, the abnormal tracing scheme based on the graph neural network has a black box effect, and can output a propagation path, but the model has poor interpretability, so that a specific causal node is difficult to effectively correspond to the actual state of the production process, and the actual diagnosis value of abnormal tracing is weakened. Disclosure of Invention The invention aims to solve the technical problems and provides a pig feed production quality tracing method based on cloud computing. The technical scheme of the invention is realized by a pig feed production quality tracing method based on cloud computing, which comprises the following steps: S1, collecting technological parameters and quality detection indexes of all links of raw material pretreatment, mixing and proportioning and high-temperature granulation in the whole feed production process, wherein the technological parameters comprise moisture content time sequence data, a temperature gradient curve and a mixing uniformity discrete value, and the quality detection indexes comprise a finished product protein content deviation rate and a granule hardness discrete coefficient; S2, performing self-adaptive filtering and time alignment processing on the acquired multi-source heterogeneous data, and carrying out sectional normalization on technological parameters by adopting a sliding window strategy to generate a standardized time sequence data set containing process characteristic labels, wherein the window length is dynamically adjusted according to characteristic time constants of all links; S3, based on Granger causal test and information transfer entropy combined criteria, causal strength calculation is carried out on variable pairs in a standardized time sequence data set, an initial causal topological structure is constructed, wherein nodes represent process quality characteristics, side weights represent inter-process abnormal propagation strength, and propagation time sequence relations are reflected in di