Zinc ore supplier evaluation and recommendation method based on nonlinear adaptive online transfer learning

Abstract

<p style='text-indent:20px;'>Purchasing decisions determine the purchasing cost, which is the largest section of the production cost of zinc smelting enterprise(ZSE). An excellent supplier recommendation is significant for ZSE to reduce the cost. However, during the supplier recommendation process, the nonlinear demand feature of purchasing department varies with the production environment, and there are wrong samples that can affect the supplier recommendation effect. To handle these problems, the recommendation strategy based on a multiple-layer perceptron adaptive online transfer learning algorithm(AOTLMLP) are proposed. In this method, the original prediction function is modified based on MLP nonlinear projective function and adaptive loss function, which enables the AOTLMLP algorithm to tackle the nonlinear classification problems and efficiently follow the demand change of purchasing department, thereby improving the result of the recommendation. The performance of the AOTLMO algorithm is evaluated through a common dataset and a purchasing dataset from a zinc smelter that generated by a supplier evaluation model. It can be assumed that AOTLMLP can ignore the influence of wrong samples and provide an effective recommendation confronting the characteristic of zinc ore purchasing.</p>