CN-122022612-A - Big data driven tea industry full chain intelligent model and visualization method

CN122022612ACN 122022612 ACN122022612 ACN 122022612ACN-122022612-A

Abstract

The invention relates to the technical field of industrial Internet and discloses a big data driven tea industry full chain intelligent model and a visualization method, which comprises the steps of firstly constructing a full life cycle process topology network according to a process formula of a production batch and production line metadata, and mapping heterogeneous sensing data into a unified event stream; the method comprises the steps of establishing a multi-state coupling digital twin body comprising biochemical reaction and material rearrangement operators for stacking processing links, extracting a mask matrix representing a monitoring blind area through local linearization and spatial modal decomposition of the twin body, calculating a hidden quality out-of-control duty ratio index by combining real-time state energy distribution statistics, and finally constructing a structured digital certificate based on the index and writing the structured digital certificate into a blockchain. The invention can quantify the structural unobservable risk in the pile processing process, and realize visual early warning and trusted evidence storage of hidden quality fluctuation.

Inventors

LI GUOSHI
ZHANG HAIJIN
Zhou Huixu
SHAO ZHENHUA

Assignees

南平市计量所
闽江大学

Dates

Publication Date: 20260512
Application Date: 20260407

Claims (8)

1. Big data driving tea industry full chain intelligent model and visualization method, which is characterized in that the method comprises the following steps: Constructing a full life cycle process topology network according to a process formula and production line information associated with tea batches, and mapping full-chain multi-source heterogeneous sensing data to the full life cycle process topology network to form a time-space aligned unified production event stream; Constructing a polymorphic coupling digital twin body aiming at a tea stacking processing link, wherein the polymorphic coupling digital twin body comprises a dynamic equation describing continuous evolution of a material biochemical reaction and a discrete state equation describing rearrangement of a material structure caused by turning and rolling, and establishing a multi-modal observation mapping associated with an external sensor; Carrying out local linearization analysis on the digital twin body to calculate a perception efficiency matrix representing the internal state capturing capacity of the current sensing network, and extracting a blind area mask representing a monitoring blind area through spatial modal decomposition; utilizing data flow type statistics to evaluate quality fluctuation distribution tensor reflecting dynamic change of actual processing process; calculating and characterizing a hidden quality out-of-control duty ratio index under the current process structure based on the quality fluctuation distribution tensor and the blind area mask of the hidden state; Based on the implicit quality out-of-control duty ratio index and the corresponding non-monitorable modal distribution, structured traceable digital certificates are constructed, and block chain evidence storage is written and full chain risk visual display is driven.
2. The big data driven tea industry full chain intelligent model and visualization method of claim 1, wherein constructing a full life cycle process topology network according to process recipe and production line information associated with tea batches comprises: Assigning a production lot identification to each tea production lot; The tea class code, the technological parameter configuration version number, the processing production line number and the finished product specification code are used as metadata to carry out serialization splicing, and the encryption hash function is utilized to calculate, so that a technological route feature code with a fixed length is generated, and the processing path and the technological logic which are experienced by the batch are locked through the feature code; Initializing a full life cycle process topology network, wherein the full life cycle process topology network is composed of a node set representing tea garden, picking, processing, storage, logistics and sales links and a directed edge set representing material circulation or state transformation events, and binding the structure of the full life cycle process topology network with the process routing feature codes.
3. The big data driven tea industry full chain intelligent model and visualization method of claim 2, wherein mapping full chain multi-source heterogeneous sensory data to the full lifecycle process topology network forms a time-space aligned unified production event stream comprising: Dividing the full-chain data into a tea garden environment flow, a picking flow, a processing sensing flow, a storage logistics flow and a sales transaction flow, and converging the full-chain data into a unified data lake; mapping original sensing data or records of each stream in a data lake into a four-tuple structure comprising a production batch identifier, a topology node identifier, a collection time and a heterogeneous data load, and forming a unified production event stream; Wherein the heterogeneous data load carries multi-modal data including hyperspectral cubes, near infrared spectra, machine vision images, dielectric constants, and warm humid wind environmental parameters.
4. The big data driven tea industry full chain intelligent model and visualization method according to claim 1, wherein constructing a multi-state coupled digital twin body for a tea stacking processing link, the multi-state coupled digital twin body comprising a dynamic equation describing continuous evolution of material biochemical reaction and a discrete state equation describing rearrangement of material structure caused by turning and kneading, and establishing a multi-state observation map associated with an external sensor, comprises: Expanding a water content field in the stacked tea leaves into a group of space modal coefficients by using a fixed space basis function, and introducing a reaction propulsion hidden variable representing coupling of air flow and biochemical reaction to jointly form a batch processing hidden state vector; Constructing a continuous evolution equation describing the change of the batch processing hidden state vector along with the environmental input and time, and a discrete state equation describing the mutation of the batch processing hidden state vector under the turning or rolling operation, wherein the discrete state equation comprises a material structure rearrangement operator for representing the gradient mixing and rearrangement effect in a pile body; establishing a stage-dependent observation map connecting the batch processing hidden state vector with multi-modal observation data including hyperspectral, near infrared and environmental parameters to form a digitized mirror image of the processing procedure.
5. The big data driven tea industry full chain intelligent model and visualization method according to claim 1, wherein the locally linearizing analysis is performed on the digital twin body to calculate a perception efficacy matrix representing the capturing capability of the current sensing network to the internal state, and extracting a blind area mask representing a monitoring blind area through spatial modal decomposition, comprising: Performing discrete local linearization processing on the polymorphic coupled digital twin bodies near the actual running track of the production batch, and calculating a local state transition jacobian matrix and an observation mapping jacobian matrix which are determined jointly by the process route and the processing stage; Calculating a perception efficiency matrix representing the degree to which the internal state of the system can be resolved under the current process structure and observation conditions based on the chain product of the local state transition jacobian matrix and the observation mapping jacobian matrix in a sliding window; And carrying out feature decomposition on the perception efficiency matrix, extracting feature vectors corresponding to near-zero feature values to form unobservable subspaces, and constructing a projection matrix as a blind zone mask of a hidden state according to the feature vectors, so as to quantify the direction of the hidden state which cannot be reconstructed through external observation under the current dimension.
6. The big data driven tea industry full chain intelligent model and visualization method of claim 1, wherein the quality fluctuation distribution tensor reflecting the dynamic change of the actual processing process is evaluated by using data flow statistics, comprising: Setting a sliding time window with a fixed length, and collecting and storing hidden state estimation vector sequences in the window in real time in the process of advancing the window along a time axis; A time decay exponential weighting mechanism is adopted to give weight to each hidden state estimation vector in the fixed-length sliding time window so as to strengthen the influence of the recent state on the current statistical result; And calculating statistical dispersion based on the weighted hidden state estimation vector sequence, and generating a quality fluctuation distribution tensor, wherein the tensor is used for representing the real energy distribution situation of the system state driven by the environment disturbance and the internal reaction together in the actual processing process in each mode direction from a statistical perspective.
7. The big data driven tea industry full chain intelligent model and visualization method of claim 1, wherein calculating a hidden quality out-of-control duty cycle index characterizing the current process structure based on the quality fluctuation distribution tensor and the blind zone mask comprises: Calculating the matrix product of the blind area mask in the hidden state and the quality fluctuation distribution tensor, and solving the trace of the product matrix to obtain the total energy of unobservable fluctuation falling in the monitoring blind area; solving the trace of the quality fluctuation distribution tensor to obtain the actual fluctuation total energy of the system state; Calculating the ratio of the total energy of the unobservable fluctuation to the total energy of the actual fluctuation to obtain a hidden quality out-of-control duty ratio index; The implicit quality out-of-control duty cycle index is used to quantify the proportion of the actual change of a production lot that falls within a structural blind zone that is unrecognizable by the sensing system under the current process routing and observation architecture.
8. The big data driven tea industry full chain intelligent model and visualization method of claim 1, wherein constructing structured traceable digital vouchers based on the implicit quality out-of-control duty cycle index and corresponding non-monitorable modal distribution, writing blockchain evidence and driving full chain risk visualization display comprises: Constructing a structured traceable digital certificate, wherein the data structure of the structured traceable digital certificate comprises a production batch identifier, a process routing feature code, a corresponding time window, a calculated implicit quality out-of-control duty ratio index, an unmonitored modal distribution vector, a model structure verification fingerprint for verifying the consistency of a digital twin body and a multi-modal observation feature abstract; carrying out serialization processing on the structured traceable digital certificate, calculating a cryptographic hash value, generating an on-chain tamper-proof anchor point, writing the anchor point into a blockchain network, and simultaneously storing original sensing data in an off-chain database; constructing a full chain risk visualization interface comprising three view levels: The first level is a full life cycle process topology network view, and each node in the network is color coded by using the implicit quality out-of-control duty ratio index so as to display structural risk heat of the full chain; The second level is a process time evolution view, and a trend curve of the implicit quality out-of-control duty ratio index changing along with time in a key processing window is drawn; And drawing an energy spectrogram through the non-monitorable modal distribution vector after normalization processing to reveal the modal shape of the missing state in the system at the current moment.

Description

Big data driven tea industry full chain intelligent model and visualization method Technical Field The invention relates to the technical field of industrial Internet, in particular to a full-chain intelligent model and a visualization method for big data driving tea industry. Background Along with the rapid development of industrial Internet and agricultural big data, the tea industry is gradually transformed from traditional experience tea making to digital and intelligent tea making. At present, full-chain intelligent monitoring has become an industry research hotspot, and mainstream technologies generally utilize an internet of things sensor to collect planting environments, temperature and humidity of a processing workshop and storage logistics states, and predict tea quality by combining a machine learning algorithm. Meanwhile, the block chain technology is also introduced into supply chain management, and anti-counterfeiting traceability of products is realized through data uplink. However, there are still significant limitations in the prior art in the intelligent monitoring and visual display of tea processing (particularly withering, fermentation, etc.) in three ways: first, existing monitoring means have difficulty penetrating the internal dead zone of the stacked/thick layer of material. In actual industrial production, the withering and fermentation processes generally employ a work form of stacking or thick layer laying (e.g., a thickness of 12cm or more). The existing machine vision or infrared spectrum technology is mainly based on surface scanning, and only can acquire moisture or color information of the surface of a material. However, the interior of the stack is subject to gravity, air permeability and thermal diffusion differences, and there are significant vertical gradients and corner effects, resulting in extremely non-uniform internal moisture distribution and temperature fields. The existing model mostly adopts single-point sampling or surface average value to represent the state of the whole batch of materials, and cannot sense the real distribution in the pile body, so that the processing end point judgment is out of alignment. Second, modeling in a single dimension ignores the deep coupling mechanism of gas flow and biochemical reactions. The prior art often considers the environmental air flow as only the dry power, ignoring the regulation and control effect of the environmental air flow on plant physiology. In fact, withering is a complex spatiotemporal system of thermo-mass transport coupled with biochemical reactions. The air flow not only takes away moisture, but also directly changes the enzymatic reaction rate and the generation path of the fragrance precursor substances by influencing the opening and closing of air holes and the damage degree of cells. The existing scheme is used for independently modeling the decoupling of the moisture migration and the biochemical quality, so that the fluctuation cause of the fragrance quality cannot be accurately explained when the model faces to the disturbance of a complex environment. Third, the traceability and visualization system lacks quantitative expression of the monitoring blind area. The current blockchain tracing system mostly adopts a mode of directly uploading data, namely, the original numerical value acquired by a sensor is stored. However, in the case of a process in which there is a blockage (e.g., inside the stack) or structural rearrangement (e.g., rolling, turning), the data acquired by the sensor is often incomplete. Existing visual signs typically only display data that can be collected, but lack metrics and cues for hidden information that is limited to structures that cannot be collected. This results in full chain tracing, which ensures that the data is not tampered, but cannot prove whether the data itself completely reflects the real quality risk of the processing process, and it is difficult to form responsibility identification evidence of a logic closed loop. Disclosure of Invention The invention provides a big data driven tea industry full chain intelligent model and a visualization method, which solve the technical problems in the background technology. The invention provides a big data driven tea industry full chain intelligent model and a visualization method, which comprises the following steps: Constructing a full life cycle process topology network according to a process formula and production line information associated with tea batches, and mapping full-chain multi-source heterogeneous sensing data to the full life cycle process topology network to form a time-space aligned unified production event stream; Constructing a polymorphic coupling digital twin body aiming at a tea stacking processing link, wherein the polymorphic coupling digital twin body comprises a dynamic equation describing continuous evolution of a material biochemical reaction and a discrete state equation describing rearrangem