Orientador: Anna Diva Plasencia Lotufo / Resumo: A maior inserção de tecnologias da informação nas redes de distribuição de energia elétrica vem permitindo que maiores volumes de dados de consumo sejam capturados em níveis cada vez mais detalhados, menos agregados e com maiores resoluções. Com a evolução dos mercados de energia elétrica, esses tipos de dados alcançam maior importância, uma vez que a comercialização de energia também passa a considerar estes níveis de consumo. Diversas técnicas têm sido aplicadas para previsão de cargas elétricas, como modelos estatísticos, de inteligência computacional e híbridos. Na literatura especializada é possível encontrar trabalhos que aplicam a rede neural artificial ARTMAP Fuzzy para tarefas de previsão de cargas elétricas, no entanto, a técnica ainda é pouco explorada em cenários de consumo menos agregados, e com maiores níveis de detalhe. Neste trabalho a rede ARTMAP Fuzzy é aplicada em tarefas de previsão multi-step de cargas elétricas reais com distintos níveis de agregação. Considerando o impacto do ruído sobre os previsores, sobretudo na capacidade de generalização das redes neurais artificiais, a técnica singular spectrum analysis é aplicada na tarefa de remoção de ruído. Os resultados de previsão permitiram analisar desempenho da rede ARTMAP Fuzzy, que foi comparada com outros dois previsores utilizados como benchmark, a saber, seasonal autoregressive integrated moving average e a rede neural multiLayer perceptron. A remoção de ruído permitiu melhora nos níveis de generaliz... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: The increased insertion of information technologies in electricity distribution networks has allowed larger volumes of consumption data to be captured at increasingly detailed, less aggregated and higher resolution levels. With the evolution of electric energy markets, these types of data become more important, since the commercialization of energy also begins to consider these levels of consumption. Several techniques have been applied to predict electrical loads, such as statistical, computational intelligence and hybrids models. In the specialized literature it is possible to find works that apply the artificial neural network ARTMAP Fuzzy for tasks of prediction of electric charges, however, the technique is still little explored in less aggregated consumption scenarios, and with greater levels of detail. In this work the ARTMAP Fuzzy network is applied in multi-step forecasting tasks of real electric loads with different levels of aggregation. Considering the impact of noise on predictors, especially in the generalization capacity of artificial neural networks, the singular spectrum analysis technique is applied in the noise removal task. The prediction results allowed to analyze the performance of the ARTMAP Fuzzy network, which was compared with other two predictors used as benchmark, namely seasonal autoregressive integrated moving average and the multiLayer perceptron neural network. The noise removal allowed an improvement in the levels of network generalization, po... (Complete abstract click electronic access below) / Doutor

Submitted by MARCOS RICARDO MÜLLER (marcos.ricardo.unesp@gmail.com) on 2018-04-23T16:13:18Z No. of bitstreams: 1 muller_mr_tese_lv.pdf: 4063915 bytes, checksum: 2bd34db37cecdea37a74a93bf0e348b5 (MD5) / Approved for entry into archive by Cristina Alexandra de Godoy null (cristina@adm.feis.unesp.br) on 2018-04-23T17:10:16Z (GMT) No. of bitstreams: 1 muller_mr_dr_ilha.pdf: 4063915 bytes, checksum: 2bd34db37cecdea37a74a93bf0e348b5 (MD5) / Made available in DSpace on 2018-04-23T17:10:16Z (GMT). No. of bitstreams: 1 muller_mr_dr_ilha.pdf: 4063915 bytes, checksum: 2bd34db37cecdea37a74a93bf0e348b5 (MD5) Previous issue date: 2018-02-26 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / A maior inserção de tecnologias da informação nas redes de distribuição de energia elétrica vem permitindo que maiores volumes de dados de consumo sejam capturados em níveis cada vez mais detalhados, menos agregados e com maiores resoluções. Com a evolução dos mercados de energia elétrica, esses tipos de dados alcançam maior importância, uma vez que a comercialização de energia também passa a considerar estes níveis de consumo. Diversas técnicas têm sido aplicadas para previsão de cargas elétricas, como modelos estatísticos, de inteligência computacional e híbridos. Na literatura especializada é possível encontrar trabalhos que aplicam a rede neural artificial ARTMAP Fuzzy para tarefas de previsão de cargas elétricas, no entanto, a técnica ainda é pouco explorada em cenários de consumo menos agregados, e com maiores níveis de detalhe. Neste trabalho a rede ARTMAP Fuzzy é aplicada em tarefas de previsão multi-step de cargas elétricas reais com distintos níveis de agregação. Considerando o impacto do ruído sobre os previsores, sobretudo na capacidade de generalização das redes neurais artificiais, a técnica singular spectrum analysis é aplicada na tarefa de remoção de ruído. Os resultados de previsão permitiram analisar desempenho da rede ARTMAP Fuzzy, que foi comparada com outros dois previsores utilizados como benchmark, a saber, seasonal autoregressive integrated moving average e a rede neural multiLayer perceptron. A remoção de ruído permitiu melhora nos níveis de generalização da rede, impactando positivamente sua capacidade preditiva. / The increased insertion of information technologies in electricity distribution networks has allowed larger volumes of consumption data to be captured at increasingly detailed, less aggregated and higher resolution levels. With the evolution of electric energy markets, these types of data become more important, since the commercialization of energy also begins to consider these levels of consumption. Several techniques have been applied to predict electrical loads, such as statistical, computational intelligence and hybrids models. In the specialized literature it is possible to find works that apply the artificial neural network ARTMAP Fuzzy for tasks of prediction of electric charges, however, the technique is still little explored in less aggregated consumption scenarios, and with greater levels of detail. In this work the ARTMAP Fuzzy network is applied in multi-step forecasting tasks of real electric loads with different levels of aggregation. Considering the impact of noise on predictors, especially in the generalization capacity of artificial neural networks, the singular spectrum analysis technique is applied in the noise removal task. The prediction results allowed to analyze the performance of the ARTMAP Fuzzy network, which was compared with other two predictors used as benchmark, namely seasonal autoregressive integrated moving average and the multiLayer perceptron neural network. The noise removal allowed an improvement in the levels of network generalization, positively impacting its predictive capacity. / 1560734

Long short-term volume forecasting is essential for companies regarding their logistics service operations. It is crucial for logistic companies to predict the volumes of goods that will be delivered to various centers at any given day, as this will assist in managing the efficiency of their business operations. This research aims to create a forecasting model for outbound logistics volumes by utilizing design science research methodology in building 3 machine-learning models and evaluating the performance of the models . The dataset is provided by Tetra Pak AB, the World's leading food processing and packaging solutions company,. Research methods were mainly quantitative, based on statistical data and numerical calculations. Three algorithms were implemented: which are encoder–decoder networks based on Long Short-Term Memory (LSTM), Convolutional Long Short-Term Memory (ConvLSTM), and Convolutional Neural Network Long ShortTerm Memory (CNN-LSTM). Comparisons are made with the average Root Mean Square Error (RMSE) for six distribution centers (DC) of Tetra Pak. Results obtained from encoder–decoder networks based on LSTM are compared to results obtained by encoder–decoder networks based on ConvLSTM and CNN-LSTM. The three algorithms performed very well, considering the loss of the Train and Test with our multivariate time series dataset. However, based on the average score of the RMSE, there are slight differences between algorithms for all DCs.

近年來由於亞洲金融風暴的肆虐，產生經濟不景氣，使得失業的問題逐漸受到社會所關注，本論文企圖以三個時間序列方法：1.單變量ARIMA模型；2.轉換函數(TF)模型；3.向量自迴歸(VAR)模型來建立台灣地區的失業率時間序列預測模型。資料則是利用台灣地區民國75年1月至民國87年12月的失業率月資料作實證預測分析，為了知道資料是否來自時間趨勢模型，測試是否經過差分消掉一部份的記憶會發生預測的誤差，所以先以多步(multi-step)預測和一步(one-step)預測的方法計算出民國88年1月至88年12月預測值，而預測評估準則則採用(1)MAPE、RMSPE、MPE及泰爾不等係數(THEIL)；(2)變化方向誤差與趨勢變化誤差兩大方向來做預測比較。最後將算出的12期預測值與行政院主計處整體統計資料庫中所得到的失業率實際值利用預測評估準則做比較，結果發現一步預測法較多步預測法準確；而向量自迴歸模型(VAR)在大部份的預測期數上有較小的MAPE、RMSPE、MPE及THEIL值，因為此VAR模型考慮了在變數之間的共整合現象，有助於模型的預測，所以有較好預測的能力；反而是較複雜的ARIMA模型及轉換模型預測能力稍差一點。 / In this thesis, we plan to construct three time series models to forecast the Taiwan unemployment Rate. These time series models are ARIMA model、transfer function (TF) model and Vector Autoregressive (VAR) model. The data set consists of monthly observations for the period 75:1-87:12 for unemployment rate. We want to know if the data came from time trend model. First, we use multi-step forecasting and one-step forecasting to calculate 12 forecasted values from 88:01-88:12. Then We compare the prediction performance of these two methods by using：(1) MAPE、RMSPE、MPE and Theil’s Inequality Coefficient (THEIL)；(2) Direction of Change Error and trend Change Error etc. It is found that one-step forecasting is more correct than multi-step forecasting and the forecasting performance of VAR model is improved by explicitly taking account of cointegration between the variables in the model，so VAR model has lower MAPE、RMSPE、MPE and THEIL for most horizons. However，the more parsimonious ARIMA and transfer function models have higher MAPE、RMSPE、MPE for most horizons.