[pt] Em um mundo globalizado, em contínua transformação, são cada vez mais freqüentes mudanças no perfil da demanda. Se não detectadas rapidamente, elas podem gerar impactos negativos no progresso de um negócio devido à baixa qualidade nas previsões de venda, que começam a gerar valores sistematicamente acima ou abaixo da demanda real indicando a presença de viés. Para evitar esse cenário, técnicas formais para detecção de viés podem ser incorporadas ao processo de previsão de demanda. Diante desse quadro, a presente dissertação compara os desempenhos, via simulação, das principais técnicas formais de detecção de viés em previsão de demanda presentes na literatura. Nesse sentido, seis técnicas são identificadas e analisadas. Quatro são baseadas em estatísticas Tracking Signal e duas são adaptadas de técnicas de Controle Estatístico de Processos. Os modelos de previsão de demanda monitorados pelas técnicas em questão são os de séries temporais estruturadas, associados ao método de amortecimento exponencial simples e ao método de Holt, respectivamente, para séries com nível médio constante e séries com tendência. Três tipos de alterações no perfil da demanda – que acarretam em viés na previsão – são examinados. O primeiro consiste em mudanças no nível médio em séries temporais de nível médio constante. O segundo tipo também considera séries temporais de nível médio constante, porém com o foco em surgimentos de tendências. O terceiro viés consiste em mudanças na tendência em series temporais com tendência pré-incorporada. Entre os resultados obtidos, destaca-se a conclusão de que, para a maioria das situações estudadas, as técnicas baseadas nas estatísticas Tracking Signal possuem desempenho superior às demais técnicas com relação à eficiência na detecção de viés. / [en] In a globalized world, in continuous transformation, changes in the demand pattern are increasingly frequent. If not rapidly detected, they can have a negative and persistent impact in the wellbeing of a business due to continuously poor quality sales forecasts, which begin to generate values systematically above or below the actual demand indicating the presence of bias. To avoid this happening, statistical techniques can be incorporated in a prediction process with the objective known as bias detection in demand forecasting. Considering this situation, the present dissertation compares, through simulation, the efficiency performance of the main existing formal techniques of monitoring demand forecasting models, with the view of bias detection. Six of such techniques are identified and analyzed in this work. Four are based on Tracking Signal Statistics and two are adapted from the Statistical Process Control approach. The demand forecasting models monitored by the techniques in question can be classified as structured time series, for a constant level or trend pattern, and using both the simple exponential smoothing and the Holt s methods. Three types of changes in the demand pattern - which result in biased prediction - are examined. The first one focus on simulated changes on the average level of various constant times series. The second type also considered various constant times series, but now simulating the appearance of different trends. And the third refers to simulate changes in trends in various times series with pre-established trends. Among the results attained, one stands out: the techniques based on Tracking Signal Statistics - when compares to other methods - showed superior performance insofar as efficient bias detection in demand forecasting.

Exponential smoothing algorithms are very attractive for the practical world such as in industry. When considering bivariate exponential smoothing methods, in addition to the properties of univariate methods, additional properties give insight to relationships between the two components of a process, and also to the overall structure of the model. It is important to study these properties, but even with the merits the bivariate exponential smoothing algorithms have, exponential smoothing algorithms are nonstatistical/nonstochastic and to study the properties within exponential smoothing may be worthless. As an alternative approach, the (bivariate) ARIMA and the structural models which are classes of statistical models, are shown to generalize the exponential smoothing algorithms. We study these properties within these classes as they will have implications on exponential smoothing algorithms. Forecast properties are studied using the state space model and the Kalman filter. Comparison of ARIMA and structural model completes the study. / Mathematical Sciences / M. Sc. (Statistics)

Demand forecasting performance is subject to the uncertainty underlying the time series an organisation is dealing with. There are many approaches that may be used to reduce demand uncertainty and consequently improve the forecasting (and inventory control) performance. An intuitively appealing such approach that is known to be effective is demand aggregation. One approach is to aggregate demand in lower-frequency 'time buckets'. Such an approach is often referred to, in the academic literature, as temporal aggregation. Another approach discussed in the literature is that associated with cross-sectional aggregation, which involves aggregating different time series to obtain higher level forecasts.This research discusses whether it is appropriate to use the original (not aggregated) data to generate a forecast or one should rather aggregate data first and then generate a forecast. This Ph.D. thesis reveals the conditions under which each approach leads to a superior performance as judged based on forecast accuracy. Throughout this work, it is assumed that the underlying structure of the demand time series follows an AutoRegressive Integrated Moving Average (ARIMA) process.In the first part of our1 research, the effect of temporal aggregation on demand forecasting is analysed. It is assumed that the non-aggregate demand follows an autoregressive moving average process of order one, ARMA(1,1). Additionally, the associated special cases of a first-order autoregressive process, AR(1) and a moving average process of order one, MA(1) are also considered, and a Single Exponential Smoothing (SES) procedure is used to forecast demand. These demand processes are often encountered in practice and SES is one of the standard estimators used in industry. Theoretical Mean Squared Error expressions are derived for the aggregate and the non-aggregate demand in order to contrast the relevant forecasting performances. The theoretical analysis is validated by an extensive numerical investigation and experimentation with an empirical dataset. The results indicate that performance improvements achieved through the aggregation approach are a function of the aggregation level, the smoothing constant value used for SES and the process parameters.In the second part of our research, the effect of cross-sectional aggregation on demand forecasting is evaluated. More specifically, the relative effectiveness of top-down (TD) and bottom-up (BU) approaches are compared for forecasting the aggregate and sub-aggregate demands. It is assumed that that the sub-aggregate demand follows either a ARMA(1,1) or a non-stationary Integrated Moving Average process of order one, IMA(1,1) and a SES procedure is used to extrapolate future requirements. Such demand processes are often encountered in practice and, as discussed above, SES is one of the standard estimators used in industry (in addition to being the optimal estimator for an IMA(1) process). Theoretical Mean Squared Errors are derived for the BU and TD approach in order to contrast the relevant forecasting performances. The theoretical analysis is supported by an extensive numerical investigation at both the aggregate and sub-aggregate levels in addition to empirically validating our findings on a real dataset from a European superstore. The results show that the superiority of each approach is a function of the series autocorrelation, the cross-correlation between series and the comparison level.Finally, for both parts of the research, valuable insights are offered to practitioners and an agenda for further research in this area is provided.

Time series model have been used in many areas of knowledge and have become a current necessity for companies to survive in a globalized and competitive market, as well as climatic factors that have always been a concern because of the different ways they interfere in human life. In this context, this work aims to present a comparison among the performances by the following models of time series: ARIMA, ARMAX and Exponential Smoothing, adjusted to air relative humidity (UR) and also to verify the volatility present in the series through non-linear models ARCH/GARCH, adjusted to residues of the ARIMA and ARMAX models. The data were collected from INMET from October, 1st to January, 22nd, 2014. In the comparison of the results and the selection of the best model, the criteria MAPE, EQM, MAD and SSE were used. The results showed that the model ARMAX(3,0), with the inclusion of exogenous variables produced better forecast results, compared to the other models SARMA(3,0)(1,1)12 and the Holt-Winters multiplicative. In the volatility study of the series via non-linear ARCH(1), adjusted to the quadrants of SARMA(3,0)(1,1)12 and ARMAX(3,0) residues, it was observed that the volatility does not tend to influence the future long-term observations. It was then concluded that the classes of models used and compared in this study, for data of a climatologic variable, showed a good performance and adjustment. We highlight the broad usage possibility in the techniques of temporal series when it is necessary to make forecasts and also to describe a temporal process, being able to be used as an efficient support tool in decision making. / Modelos de séries temporais vêm sendo empregados em diversas áreas do conhecimento e têm surgido como necessidade atual para empresas sobreviverem em um mercado globalizado e competitivo, bem como fatores climáticos sempre foram motivo de preocupação pelas diferentes formas que interferem na vida humana. Nesse contexto, o presente trabalho tem por objetivo apresentar uma comparação do desempenho das classes de modelos de séries temporais ARIMA, ARMAX e Alisamento Exponencial, ajustados a dados de umidade relativa do ar (UR) e verificar a volatilidade presente na série por meio de modelos não-lineares ARCH/GARCH ajustados aos resíduos dos modelos ARIMA e ARMAX. Os dados foram coletados junto ao INMET no período de 01 de outubro de 2001 a 22 de janeiro de 2014. Na comparação dos resultados e na seleção do melhor modelo foram utilizados os critérios MAPE, EQM, MAD e SSE. Os resultados mostraram que o modelo ARMAX(3,0) com a inclusão de variáveis exógenas produziu melhores resultados de previsão em relação aos seus concorrentes SARMA(3,0)(1,1)12 e o Holt-Winters multiplicativo. No estudo da volatilidade da série via modelo não-linear ARCH(1), ajustado aos quadrados dos resíduos dos modelos SARMA(3,0)(1,1)12 e ARMAX(3,0), observou-se que a volatilidade não tende a influenciar as observações futuras em longo prazo. Conclui-se que as classes de modelos utilizadas e comparadas neste estudo, para dados de uma variável climatológica, demonstraram bom desempenho e ajuste. Destaca-se a ampla possibilidade de utilização das técnicas de séries temporais quando se deseja fazer previsões e descrever um processo temporal, podendo ser utilizadas como ferramenta eficiente de apoio nas tomadas de decisão.

Exponential smoothing algorithms are very attractive for the practical world such as in industry. When considering bivariate exponential smoothing methods, in addition to the properties of univariate methods, additional properties give insight to relationships between the two components of a process, and also to the overall structure of the model. It is important to study these properties, but even with the merits the bivariate exponential smoothing algorithms have, exponential smoothing algorithms are nonstatistical/nonstochastic and to study the properties within exponential smoothing may be worthless. As an alternative approach, the (bivariate) ARIMA and the structural models which are classes of statistical models, are shown to generalize the exponential smoothing algorithms. We study these properties within these classes as they will have implications on exponential smoothing algorithms. Forecast properties are studied using the state space model and the Kalman filter. Comparison of ARIMA and structural model completes the study. / Mathematical Sciences / M. Sc. (Statistics)

[pt] Nesta dissertação é desenvolvida uma metodologia para previsão de demanda mensal de energia elétrica considerando cenários de racionamento. A metodologia usada consiste em, a partir das taxas de crescimento da série temporal, identificar e eliminar os efeitos do racionamento de energia elétrica através da aplicação de Modelos Lineares Dinâmicos. São analisadas também estruturas de intervenção nos modelos estatísticos de Box & Jenkins e Holt & Winters. Os modelos são então comparados segundo alguns critérios, basicamente no que tange à sua eficiência preditiva. Conclui-se ao final sobre a eficiência da metodologia proposta, dado a grande dificuldade para solucionar o problema a partir dos modelos estatísticos de Box & Jenkins e Holt & Winters. Esta solução é então proposta como a mais viável para criar cenários de racionamento e pósracionamento de energia para ser utilizado por agentes do sistema elétrico nacional. / [en] In this dissertation, a methodology is developed to forecast monthly demand of electric energy, considering the rationing scenery. The methodology is based on, taking the growth rate from the time series, identify and eliminate the effects of electric energy rationing, using Dynamic Linear Models. It is also analyzed intervention structures in the statistics models of Box & Jenkins and Holt & Winters. The models are compared according to some criterions, mainly forecast accuracy. At the end, we concluded that the methodology proposed is more efficient, due to the difficult to solve the problem using the statistics models with intervention. This solution is proposed as the best among them to create scenery during the energy rationing and after energy rationing, to be used by the national electric system agents.

The subject of this master's thesis is extension of existing data mining system. System will be extended by the module for the time series data mining. This thesis consists of common introduction to data mining issues and continues with time series analysis. Thesis then also contains some of the current tasks and algorithms used in time series data mining, follows by the concept of the implementation and description of the choosen mining method. Possible future system's improvments are disscused at the end of the paper.

Salinity intrusion in a river may have an adverse effect on the quality of life and can be perceived as a modern-day curse. Therefore, it is important to find technical ways to monitor and forecast salinity intrusion. In this paper, we designed a forecasting model using Simple Exponential Smoothing method (SES) which performs weekly salinity intrusion forecast in Ham Luong river (HLR), Ben Tre province based on historical data obtained from the Center for Hydro-meteorological forecasting of Ben Tre province. The results showed that the SES method provides an adequate predictive model for forecast of salinity intrusion in An Thuan, Son Doc, and Phu Khanh. However, the SES in My Hoa, An Hiep, and Vam Mon could be improved upon by another forecasting technique. This study suggests that the SES model is an easy-to-use modeling tool for water resource managers to obtain a quick preliminary assessment of salinity intrusion. / Xâm nhập mặn có thể gây tác động xấu đến đời sống con người, tuy nhiên nó hoàn toàn có thể dự báo được. Cho nên, một điều quan trọng là tìm được phương pháp kỹ thuật phù hợp để dự báo và giám sát xâm nhập mặn trên sông. Trong bài báo này, chúng tôi sử dụng phương pháp Simple Exponential Smoothing để dự báo xâm nhập mặn trên sông Hàm Luông, tỉnh Bến Tre. Kết quả cho thấy mô hình dự báo phù hợp cho các vị trí An Thuận, Sơn Đốc, và Phú Khánh. Tuy nhiên, các vị trí Mỹ Hóa, An Hiệp, và Vàm Mơn có thể tìm các phương pháp khác phù hợp hơn. Phương pháp Simple Exponential Smoothing rất dễ ứng dụng trong quản lý nguồn nước dựa vào việc cảnh báo xâm nhập mặn.

[pt] O objetivo do presente trabalho é o desenvolvimento de um algoritmo estatístico de previsão da potência transmitida pela usina geradora termelétrica de Linhares, localizada no Espírito Santo, medida no ponto de entrada da rede da concessionária regional, a ser integrado em plataforma composta por sistema supervisório em tempo real em ambiente MS Windows. Para tal foram comparadas as metodologias de Modelos Arima(p,d,q), regressão usando polinômios ortogonais e técnicas de amortecimento exponencial para identificar a mais adequada para a realização de previsões 5 passos-à-frente. Os dados utilizados são provenientes de observações registradas a cada 5 minutos, contudo, o alvo é produzir estas previsões para observações registradas a cada 5 segundos. Os resíduos estimados do modelo ajustado foram analisados via gráficos de controle para checar a estabilidade do processo. As previsões produzidas serão usadas para subsidiar decisões dos operadores da usina, em tempo real, de forma a evitar a ultrapassagem do limite de 200.000 kW por mais de quinze minutos. / [en] The objective of this study is to develop a statistical algorithm to predict the power transmitted by a thermoelectric power plant in Linhares, located at Espírito Santo state, measured at the entrance of the utility regional grid, which will be integrated to a platform formed by a real time supervisor system developed in MS Windows. To this end we compared Arima (p,d,q), Regression using Orthogonal Polynomials and Exponential Smoothing techniques to identify the best suited approach to make predictions five steps ahead. The data used are observations recorded every 5 minutes, however, the target is to produce these forecasts for observations recorded in every five seconds. The estimated residuals of the fitted model were analysed via control charts to check on the stability of the process. The forecasts produced by this model will be used to help not to exceed the 200.000 kW energy generation upper bound for more than fifteen minutes.

共變數的值會隨著時間而改變時，我們稱之為時間相依之共變數。時間相依之共變數往往具有重複測量的特性，也是長期資料裡最常見到的一種共變數形態；在對時間相依之共變數進行重複測量時，可以考慮每次測量的間隔時間相同或是間隔時間不同兩種情形。在間隔時間相同的情形下，我們可以忽略間隔時間所產生的效應，利用分組的Cox模式或是合併的羅吉斯迴歸模式來分析，而合併的羅吉斯迴歸是一種把資料視為“對象 時間單位”形態的分析方法；此外，分組的Cox模式和合併的羅吉斯迴歸模式也都可以用來預測存活機率。在某些條件滿足下，D’Agostino等六人在1990年已經證明出這兩個模式所得到的結果會很接近。 當間隔時間為不同時，我們可以用計數過程下的Cox模式來分析，在計數過程下的Cox模式中，資料是以“對象 區間”的形態來分析。2001年Bruijne等人則是建議把間隔時間也視為一個時間相依之共變數，並將其以B-spline函數加至模式中分析；在我們論文的實證分析裡也顯示間隔時間在延伸的Cox模式中的確是個很顯著的時間相依之共變數。延伸的Cox模式為間隔時間不同下的時間相依之共變數提供了另一個分析方法。至於在時間相依之共變數的預測方面，我們是以指數趨勢平滑法來預測其未來時間點的數值；利用預測出來的時間相依之共變數值再搭配延伸的Cox模式即可預測未來的存活機率。 / It is so called “time-dependent covariates” that the values of covariates change over time. Time-dependent covariates are measured repeatedly and often appear in the longitudinal data. Time-dependent covariates can be regularly or irregularly measured. In the regular case, we can ignore the TEL(time elapsed since last observation) effect and the grouped Cox model or the pooled logistic regression model is employed to anlalyze. The pooled logistic regression is an analytic method using the“person-period”approach. The grouped Cox model and the pooled logistic regression model also can be used to predict survival probablity. D’Agostino et al. (1990) had proved that pooled logistic regression model is asymptotically equivalent to the grouped Cox model. If time-dependent covariates are observed irregularly, Cox model under counting process may be taken into account. Before making the prediction we must turn the original data into“person-interval”form, and this data form is also suitable for the prediction of grouped Cox model in regular measurements. de Bruijne et al.(2001) first considered TEL as a time-dependent covariate and used B-spline function to model it in their proposed extended Cox model. We also show that TEL is a very significant time-dependent covariate in our paper. The extended Cox model provided an alternative for the irregularly measured time-dependent covariates. On the other hand, we use exponential smoothing with trend to predict the future value of time-dependent covariates. Using the predicted values with the extended Cox model then we can predict survival probablity.