CN-121526136-B - Multi-condition complex assembly line resource allocation evaluation method based on instant learning

CN121526136BCN 121526136 BCN121526136 BCN 121526136BCN-121526136-B

Abstract

The invention provides a multi-condition complex assembly line resource allocation evaluation method based on instant learning, which comprises the steps of determining key resource variables influencing production beats based on complex assembly line production modes and process flows, and preparing complex assembly line resource allocation data to construct a sample database. And (3) adopting a Kmeans clustering method to realize working condition division on the resource allocation of the complex assembly line, and classifying the resource allocation schemes with similar production characteristics. And constructing an LWPS-Kmeans dynamic fusion algorithm model based on instant learning, and training an evaluation model by using the resource configuration data. And inputting different complex assembly line resource allocation schemes to be evaluated into a resource allocation evaluation model to realize accurate evaluation of the efficiency of various resource allocation schemes. According to the invention, by introducing an instant learning algorithm and a Kmeans clustering algorithm and optimizing a similarity sample subset screening mechanism, a dynamic fusion resource allocation evaluation model of LWPS-Kmeans based on instant learning is established, and accurate prediction of production beats is realized, so that the accuracy of resource allocation evaluation is improved.

Inventors

TANG WENBIN
RONG YUXIANG
YANG LEIPENG
ZHANG MIN

Assignees

西安工程大学

Dates

Publication Date: 20260512
Application Date: 20251023

Claims (5)

1. A multi-condition complex assembly line resource allocation evaluation method based on instant learning is characterized by comprising the following steps: Step1, determining key resource variables influencing production beats according to a complex assembly line production mode and a process flow, preparing complex assembly line resource configuration data and constructing a sample database; step 2, realizing working condition division on resource allocation of a complex assembly line by using a Kmeans clustering method, classifying the resource allocation of similar production characteristics, and specifically: Carrying out standardized pretreatment on a resource allocation sample data set X= [ X 1 ,x 2 ,…,x N ], wherein a sample X i =[x i1 ,x i2 ,…,x im represents an ith resource allocation sample of a complex assembly line, N is the total number of samples, m is the dimension of a resource allocation variable, clustering the resource allocation sample data set, adopting a contour coefficient as an evaluation index, determining the optimal clustering number by comparing clustering quality under different clustering values, carrying out Kmeans clustering on the standardized resource allocation data by utilizing the optimal clustering value k to obtain a working condition dividing result { G 1 ,G 2 ,…,G k }, wherein the expression of the contour coefficient is as follows: Wherein the method comprises the steps of Intra-cluster average distance for sample x i ; the average distance of nearest neighbor clusters for sample x i ; The objective function for Kmeans clustering is: Wherein k is the number of clusters, v j is the center vector of the jth cluster, C j represents the sample set of the jth cluster; Step 3, constructing a resource configuration evaluation model based on the dynamic fusion of the local weighted partial least square-k-means clustering based on instant learning, and training and optimizing the evaluation model through resource configuration data, wherein the method comprises the following steps of: constructing a resource allocation data set D= [ X, Y ] and carrying out standardization processing, wherein Y= [ Y 1 ,y 2 ,…,y N ],y i ] is the production takt of a complex assembly line of the ith sample; For an evaluation model, in an offline stage, utilizing a standardized resource allocation data set, and determining an optimal parameter combination (h, R, p) by using a leave-one-out cross-validation method and a grid search method through minimizing a prediction error, wherein h and R are core parameters of a local partial least square algorithm, and p is a distance threshold; In the online stage, for each resource allocation sample, calculating the Euclidean distance between the sample and each clustering center, screening by a distance threshold p to obtain a similarity sample subset set Q= (Q 1 ,Q 2 ,…,Q s ), constructing an instant learning model by adopting a local weighted partial least square model aiming at each similarity sample subset Q j , and outputting a corresponding production takt predicted value y j ; And 4, inputting different complex assembly line resource allocation schemes into the evaluation model in the step 3 to obtain the production takt prediction result of each resource allocation scheme, and accurately evaluating the efficiency of each resource allocation scheme according to the production takt prediction result.
2. The multi-condition complex assembly line resource allocation assessment method based on instant learning is characterized in that in the step 1, the pearson correlation coefficient method is adopted to analyze the characteristic correlation of complex assembly line resources, the degree of correlation between different resources and production beats is assessed, based on characteristic correlation sequencing results, variable combinations of resource allocation with higher correlation ranking are selected as input variables of a resource allocation assessment model respectively, based on an actual operation process, a complex assembly line simulation model is established through discrete time simulation software, and the production beats of different resource allocation are obtained through simulation operation.
3. The multi-condition complex assembly line resource allocation evaluation method based on instant learning according to claim 1, wherein in step 3, the process of predicting the production tact by using the local weighted partial least squares model is as follows: Step 3.1 according to the formula Calculating the distance D= [ D 1 ,d 2 ,…,d N ] between the predicted point and each training sample, and then passing through the formula Calculating the corresponding weight of each training sample, and constructing a diagonal matrix according to the weight calculation result : D i is the Euclidean distance between the predicted point x q and the training sample x i ; Step 3.2 according to the formula Performing data averaging, wherein And For the weighted average of the input and output of the training sample, 1 n is an n-dimensional column vector of all 1, and X r 、Y r and X q,r are respectively the input, output and predicted point vector of the training sample after centering; step 3.3, extracting latent variables from the input of training samples and the predicted samples x q,r respectively: Wherein t r and t q,r are the r-th latent variables corresponding to X r and X q,r , respectively, and w r is A feature vector corresponding to the maximum feature value of (a); According to the formula Calculating a load vector p r of the predicted point x q and a model regression coefficient vector q r , and establishing a regression equation Updating the output of the predicted point, if the current latent variable number r reaches the set value, outputting the predicted value of the production takt time of the predicted point Otherwise let r=r+1, by the formula And (3) returning to the step 3.3 for continuous iteration after updating the input and output and predicted point vectors.
4. The multi-condition complex assembly line resource allocation evaluation method based on instant learning according to claim 1, wherein in step 3, the process of obtaining the similarity sample subset by screening according to the distance threshold p is as follows: By the formula The Euclidean distance between the predicted point and each clustering center is calculated and compared with a distance threshold p; If the distance is within the threshold range, the corresponding cluster dataset is selected as a similar sample subset q= (Q 1 ,Q 2 ,…,Q s ) for the predicted point: 。
5. The multi-condition complex assembly line resource allocation evaluation method based on instant learning according to claim 1, wherein in step 3, based on the predicted value y j and the Euclidean distance d j of the production beats corresponding to the clustering center, the predicted results of the instant learning model corresponding to each similar sample subset are weighted and fused, and the process of outputting the predicted production beats y is as follows: Wherein the method comprises the steps of Multiplying the predicted result of the instant learning model corresponding to each similar sample subset by the corresponding weight, and carrying out weighted summation to obtain a final predicted value of the production beat: 。

Description

Multi-condition complex assembly line resource allocation evaluation method based on instant learning Technical Field The invention belongs to the field of complex assembly line resource allocation evaluation, and particularly relates to a multi-condition complex assembly line resource allocation evaluation method based on instant learning. Background The intelligent transformation of the manufacturing industry is accelerated, and the market demand is also rapidly changed. The complex assembly line is used as a core production unit in the high-end manufacturing field, and whether the complex assembly line can respond to market demand change quickly directly influences the competitiveness of enterprises. The takt time (CYCLE TIME, CT) is a key index for directly showing whether the resource allocation is reasonable or not, and is often used for evaluating the resource allocation effect. However, due to the nonlinear coupling relationship and multiple operating characteristics between the complex assembly line resource configuration data and the tact, this presents challenges for evaluating the resource configuration by accurately predicting the tact. Publication nos. Sun Jinhao, yang Yi, du Rui, etc. aircraft assembly line beat closed-loop control model design and analysis [ J ]. Aircraft manufacturing techniques, 2023, 66 (8): 38-46 "," by constructing an aircraft assembly line capacity prediction model using a markov model, and feeding back prediction data to the beat control system in real time, and further adjusting the production plan by adjusting the resource allocation. Although the conventional discrete event simulation method can evaluate the optimal resource allocation scheme based on the production takt, the algorithm of the publication has higher complexity and has the problem of low solving efficiency. To overcome this limitation, advanced technologies such as machine learning and digital twin algorithms are cited. Publication nos. Zhang Qi, jiang Changjian, han Jiawei, etc. aircraft assembly site material configuration optimization based on GRU neural networks and genetic algorithms [ J ]. Aviation manufacturing techniques, 2024, 67 (17): 78-82, 92, "build material configuration scheme assessment simulation agent model with GRU neural networks, take material configuration scheme as input, take projected finishing time at the time of loading and critical material average residence time as output, efficiently assess material configuration scheme. Although this publication achieves significant results in terms of reducing computational complexity and improving accuracy in evaluating resource configurations, the global modeling approach is difficult to accommodate dynamic changes in complex assembly lines, and it is difficult to maintain accurate evaluation accuracy for all resource configurations. Therefore, the accuracy of the existing complex assembly line resource allocation evaluation method still has room for further improvement. Disclosure of Invention In order to solve the problem of insufficient accuracy of multi-working-condition complex assembly line resource allocation evaluation in the prior art, the invention provides a multi-working-condition complex assembly line resource allocation evaluation method based on instant learning. The method introduces an instant learning algorithm and a Kmeans clustering algorithm and optimizes a similarity sample subset screening mechanism, and builds a dynamic fusion resource allocation evaluation model of LWPS-Kmeans based on instant learning. By adopting a prediction result dynamic weighting fusion strategy, the model can realize accurate prediction of the production takt, thereby improving the accuracy of resource allocation evaluation. The technical scheme of the invention is as follows: A multi-condition complex assembly line resource allocation evaluation method based on instant learning comprises the following steps: Step1, determining key resource variables influencing production beats according to a complex assembly line production mode and a process flow, preparing complex assembly line resource configuration data and constructing a sample database; Step2, utilizing a Kmeans clustering method to divide working conditions of resource allocation of a complex assembly line, and classifying the resource allocation of similar production characteristics; Step 3, constructing a resource configuration evaluation model based on the dynamic fusion of the local weighted partial least square-k mean value clustering of the instant learning, and training and optimizing the evaluation model through resource configuration data; And 4, inputting different complex assembly line resource allocation schemes into the evaluation model in the step 3 to obtain the production takt prediction result of each resource allocation scheme, and accurately evaluating the efficiency of each resource allocation scheme according to the production takt prediction result. Furth