CN-122021310-A - Multi-target technological parameter optimization method for tea production line

CN122021310ACN 122021310 ACN122021310 ACN 122021310ACN-122021310-A

Abstract

The invention relates to a multi-target process parameter optimization method for a tea production line, which belongs to the technical field of tea processing and comprises the steps of constructing a data acquisition system of an Internet of things, acquiring data of each process section in real time, fusing a processing mechanism and historical data, establishing a digital twin model capable of predicting quality and energy consumption, defining a multi-target optimization problem aiming at maximizing comprehensive quality scores and minimizing unit product energy consumption, adopting an improved multi-target evolution algorithm, carrying out iterative optimization by taking the digital twin model as an evaluation function to obtain a pareto optimal process parameter solution set, selecting optimal parameters according to decision preference, issuing and executing, and carrying out online correction on the model based on actual production data to realize closed-loop optimization. According to the invention, the intelligent and multi-objective cooperative optimization of the tea processing technological parameters is realized, and the production energy consumption is effectively reduced while the comprehensive quality of the tea is stabilized and improved.

Inventors

LI JIE
GAO LI
WU QINGYANG
Cao Shuci
ZHAO YIQING

Assignees

九江职业大学

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (7)

1. S1, constructing an Internet of things data acquisition system covering the withering, rolling, fermenting and drying procedures of a tea production line, and acquiring equipment operation parameters, environment parameters and material state information of each process section in real time; s2, based on collected data, fusing tea processing heat mass transfer and enzyme dynamics mechanism, and constructing a layered hybrid digital twin model combining a mechanism sub-model and a data driving sub-model, wherein the digital twin model can predict a plurality of tea quality indexes and energy consumption indexes according to input technological parameters; S3, defining a multi-objective optimization problem, wherein an objective function at least comprises maximizing comprehensive quality scores and minimizing unit product energy consumption, and constraint conditions comprise feasible domains of all process parameters and key quality index thresholds; S4, adopting a multi-objective evolutionary algorithm, carrying out iterative optimization in a process parameter space by taking the digital twin model as a simulation evaluation function to obtain a group of pareto optimal process parameter solution sets, and introducing uncertainty of a predicted value of the digital twin model as a disturbance item in the fitness calculation of the multi-objective evolutionary algorithm to carry out robustness optimization; S5, selecting a group of optimal process parameters from the pareto optimal process parameter solution set according to preset decision preferences, and issuing the optimal process parameters to a production line control system for execution; S6, running a production line under the optimal technological parameters, collecting actual production data and final tea quality data, comparing the actual production data and the final tea quality data with predicted values of a digital twin model, and carrying out online correction on the digital twin model based on a comparison result to realize closed-loop optimization.
2. The multi-target process parameter optimization method of the tea production line according to claim 1 is characterized in that in the step S1, the data acquisition system of the Internet of things comprises a temperature and humidity sensor array and a near infrared spectrometer which are deployed in a withering process, a pressure sensor, a torque sensor and a visual sensor which are deployed in a rolling process, a distributed temperature and humidity sensor and a gas concentration sensor which are deployed in a fermentation process, and a multi-stage temperature sensor and a weightlessness metering module which are deployed in a drying process.
3. The multi-target process parameter optimization method of the tea production line according to claim 1, wherein in the step S2, the data driving sub-model adopts an LSTM neural network or a Transformer architecture, the mechanism sub-model and the data driving sub-model are fused through an adaptive weighting module, and the adaptive weighting module dynamically adjusts the output weights of the two sub-models according to the currently input process parameter working conditions.
4. The method for optimizing multi-objective technological parameters of tea production line according to claim 1, wherein in step S3, the comprehensive quality score is obtained by obtaining aroma index, taste index, color score and appearance score of tea by using electronic nose, electronic tongue, machine vision and near infrared spectrum detection equipment, and calculating by using an evaluation model based on combination weighting of analytic hierarchy process and entropy weighting method.
5. A method for optimizing multi-objective process parameters of a tea production line according to claim 1, wherein in step S4, the multi-objective evolutionary algorithm is a reference point-based modified NSGA-III algorithm, and the improvement is that uncertainty of the digital twin model prediction is added to fitness calculation as a disturbance term of an objective function, so as to enhance robustness of the algorithm to model errors.
6. A multi-objective process parameter optimization method for a tea production line according to claim 1, wherein in step S4, in an initialization stage of the multi-objective evolutionary algorithm, heuristic rules based on historical high-quality case data are introduced to generate a partial initial population to accelerate algorithm convergence.
7. The multi-objective process parameter optimization method of a tea production line according to claim 1, wherein in step S6, the online correction adopts a strategy of combining a recursive least square method with bayesian update to dynamically fine-tune the weights of the data driving submodels and the key parameters of the mechanism submodels in the digital twin model.

Description

Multi-target technological parameter optimization method for tea production line Technical Field The invention relates to the technical field of tea processing, in particular to a multi-target process parameter optimization method for a tea production line. Background In the field of tea processing, the intelligent optimization of technological parameters is a key for improving the quality stability of products and reducing the production energy consumption, and currently, the technical development of the field mainly has the following layers that firstly, in the single-process equipment parameter optimization layer, the traditional method depends on experiences of operators or experiments in a small range, for example, a method of combining numerical simulation with orthogonal experiments is studied to optimize structural parameters of a strip tidying machine so as to improve the strip forming rate, and the method is effective in a certain range, but belongs to single-point and static optimization, the coupling influence among working procedures cannot be considered from the aspect of a production line system, and the optimization target is single (such as only appearance), so that multiple targets such as quality, energy consumption and the like are difficult to be comprehensively managed. Secondly, in the multi-objective collaborative optimization of the production line, research on parameter optimization by using an intelligent algorithm starts to appear in recent years, for example, a Chinese patent invention discloses a multi-objective process parameter optimization method of the tea production line based on improved particle swarm, which is implemented by establishing a prediction model of water content and chemical components and utilizing the improved particle swarm algorithm to perform multi-objective optimization, and the method represents the current research front, but has obvious limitations that firstly, the prediction model is a pure data driving model based on historical data, prediction reliability and physical interpretability are insufficient when data are scarce or production conditions are changed drastically, and secondly, once the model is established, self correction cannot be performed according to subsequent actual production feedback, and when the characteristics of processing raw materials fluctuate or the performance of equipment is degraded, the optimization effect is gradually reduced. Thirdly, in the automatic control layer of the production line, in order to improve the control precision, the prior art adopts advanced control algorithms such as fuzzy PID and the like to realize the precise control of key parameters such as fixation temperature and the like, however, the technology still solves the control problem of how to make the parameters stable at a set value, but not the optimization problem of what the set value is, the parameter setting still depends on experience, and the automatic optimizing capability taking the final comprehensive quality and energy efficiency as the guide is lacking. In addition, in other links of the tea industry (e.g. blending), there are studies on the application of fuzzy multi-objective linear programming to address the cost and quality tradeoff, but this is quite different from the dynamic, nonlinear parameter optimization of the process in nature. In summary, the prior art cannot effectively solve the problem of closed loop optimization of the process parameters of the tea production line with multiple targets (quality/energy consumption) and self-adaption, and the specific technical bottlenecks are that 1, a high-fidelity process model with a fusion mechanism and data is lacked, so that the optimization basis is not firm, 2, a robust optimization algorithm considering the uncertainty of the model is lacked, and 3, an online correction mechanism based on real-time feedback is lacked, so that a continuously improved closed loop cannot be formed, therefore, the technical problem to be solved at present is how to construct a dynamic, interpretable and self-adaption process model capable of predicting the multi-target output of the tea processing with high precision, and design a closed loop optimization method capable of tolerating model errors and continuously improving by utilizing production feedback on the basis, so that the continuous autonomous optimization of the process parameters of the tea production line on the whole flow and multiple target dimensions is realized. Disclosure of Invention The invention provides a multi-target process parameter optimization method for a tea production line, which solves the problems of the prior art, and realizes intelligent and multi-target collaborative optimization of tea processing process parameters by constructing a closed-loop framework of perception-modeling-optimization-decision-correction. S1, constructing an Internet of things data acquisition system covering the withering, rolling, fer