Percorrer por autor "Kochinski, Daniel Tiepolo"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Home energy management systems regression modelsPublication . Kochinski, Daniel Tiepolo; Pereira, Ana I.; Lima, José; Lazzaretti, Andre EugenioRegression Models have good use in the predictability of electrical systems and for Home Energy Management Systems (HEMS) buildings. This master’s thesis performs simulations with data from the Silk House, a building in Bragança. The objective is to determine better parameters in building data collection to improve its efficiency. Several Regression Models in Machine Learning (ML) are in a Python algorithm that constructs different inputs to an output. The Data Set is short, with seven scalar variables of the building’s power flow over a year, measured daily. The algorithm changed the number of variables chosen in the input and ran several models, with and without Principal Component Analysis (PCA). The Coefficient of Determination (R2) measures how well a regression model fits the data and its percentage of results withR2 in the range [0.75, 1] across all simulations. The best results for R2 in the range [0.75, 1] found 45% without PCA and 47.14% with PCA. With just one input, all models initially found 0% R2 in the range [0.75, 1]. The results of R2 in the range [0.75, 1] increased directly with more variables in the input. The variables with the best results were Photovoltaic Production (PP) and Direct Consumption (DC), being consistent with the profile of the building (office), which recommends its expansion. The variable Battery Charge (BC) never reached any R2 in the range [0.75, 1], which indicates possible suppression. It is also concluded that it is prudent to have more data and that non-linear tools are more suitable for site analysis.