Research And Application Of Supplier Health Evaluation Algorithm

Supply chain management plays an increasingly important role in the operation of enterprises.For downstream manufacturers,the risks of upstream suppliers can easily be transmitted through the supply chain,which will have a negative impact on their own operations.Therefore,enterprises are paying more and more attention to health evaluation of their suppliers.The traditional evaluation method is mainly a formulaic judgment method obtained based on experts' analysis of empirical data.However,the labor cost of this method is high,and the subjective opinions of experts have a great impact on the evaluation results.The demand of enterprises for intelligence supply chain management is increasingly strong.This article mainly explores the issue of supplier health evaluation based on traditional machine learning,transfer learning and lifelong learning methods,and constructs a good performance evaluation model.The specific research contents are as follows:(1)In order to mine more information from the training data and further improve the evaluation performance of the model,this paper constructs a health evaluation model based on dimensional division method.In the model building process,the secondary features obtained through the Logistic,Random Forest,and XGBoost base classifiers are combined with the original features to obtain a mixed training set for training the secondary classifier.Then,based on the idea of regional features extraction from CNN,the features in the mixed training set are innovatively classified according to solvency,profitability,operational ability,development ability,and secondary features.Before training the secondary classifier,A network transformation is performed inside the dimension,and finally the evaluation result is obtained through the ANN secondary classifier.By comparing with other commonly used health evaluation methods,the model has achieved better results.(2)Due to the industry differences,establish separate evaluation models for each industry tends to achieve better performance,but too many industries will lead to too many evaluation models.Therefore,this paper combines the ideas of transfer learning and lifelong learning,proposes a novel multi-task learning method based on similarity,Fine-tune + SI.The main idea of this method is to construct different models for tasks with large differences and merge models for tasks with small differences.The differences between different industries make the single-model evaluation method unable to effectively evaluate all industries.This article first analyzes and verifies the industry differences,and then divides each industry into a separate task,and uses the multi-task learning method proposed in this article to build evaluation model.Experiments show that the multi-task learning method based on similarity proposed in this paper can greatly reduce the complexity of the model while ensuring the performance of the model,and reduce the number of models from 14 to 4.In addition,the method can also mine the internal correlation between industries to a certain extent.(3)To accelerate the application of the supplier health evaluation model in the actual production environment,a supplier health evaluation web service based on Flask is constructed.The service integrates modules such as model training,model evaluation and model application.Meanwhile,it provides health evaluation service of supplier information.Users can upload their own data sets and train models on them at the same time.