Desenvolvimento de ferramenta computacional para reconhecimento de padrões na indústria sucroenergética

Abstract

Sugar color is one of the main quality parameters for domestic and foreign markets. Color precursors can be generated during production steps or found in sugarcane in the form of amino acids, pigments and phenolic compounds. The implementation of methodologies for detection and quantification of these dyes is not always possible within the routine of industrial laboratory tests. As an alternative to monitoring the use of such detailed development tools, recognizing or increasing machine learning and monitoring the quality of industrial use or a use of work analysis tools or a use of used learning work tools machine to identify quality relationships between the sugar produced and the sugarcane varieties used in the production process. The tools were composed of seven references to sugarcane varieties, five variables identified to the resultant-chemical properties of the mixed dataset and a variable referring to the color of the respective sugar produced. The sugar specifications were originally classified as “In Specification” and “Out of Specification Sugar” according to the color specifications adopted by the International Commission for Uniform Methods of Sugar Analysis (ICUMSA) for the VHP type. The component analysis data known by an autoscaling preprocessing was selected by a middle of the principal exploration (PCA). During an analysis it was possible to visualize a data structure, recognize and eliminate anomalous samples and verify relationships between variables. Four classification models were built in a Python environment, using the scikit-learn library and compared; two of them close networks in the making k-is k-is nearimmost closest (K and k-is close neighbors)3, 45 and the N LPN two N and nm k-is of closest neighbors (K-is LP-K-M-M-N) and in the k-é k-é nextimmost closer (K-ép) The models that present the lowest prediction capacity were the KNN (k = 3 and 4) with 96.4% of the dates correctly classified. The other models showed 100% efficiency in the sample classification of the test set. A graphical interface was developed computationally by people or using models unrelated to knowledge related to programming languages. The one used in the interface was the PCA-KNN for presenting 100% efficiency and requiring less computational time for its construction. The developed tool became accessible to anyone who has access to a useful tool presented and leaves for a specification based on the raw material of the process, making it possible to use it in the strategic planning Sale through the prior evaluation of the lots of sugarcane for production, thus manufacturing off-spec products.