In times of demand shocks, when quantitative forecasting based on historical time series becomes obsolete, the only information about future demand is "advance demand information", i.e. interpreting early customer bookings as an indicator of not yet known demand. This paper deals with a forecasting method which selects the optimal forecasting model type and the level of integration of advance demand information, depending on the patterns of the particular time series. This constitutes the applicability of the procedure within an industrial application where a large number of time series is automatically forecasted in a flexible and data-driven way. The architecture of such a planning system is explained and using real-world data from a make-to-order industry it is shown that the system is flexible enough to cover different demand patterns and is well-suited to forecast demand shocks. (authors' abstract)

This thesis describes a short term water demand forecasting application that is based upon a combination of a neural network forecast generator and a rule based system that modifies the resulting forecasts. Conventionally, short term forecasting of both water consumption and electrical load demand has been based upon mathematical models that aim to either extract the mathematical properties displayed by a time series of historical data, or represent the causal relationships between the level of demand and the key factors that determine that demand. These conventional approaches have been able to achieve acceptable levels of prediction accuracy for those days where distorting, non cyclic influences are not present to a significant degree. However, when such distortions are present, then the resultant decrease in prediction accuracy has a detrimental effect upon the controlling systems that are attempting to optimise the operation of the water or electricity supply network. The abnormal, non cyclic factors can be divided into those which are related to changes in the supply network itself, those that are related to particular dates or times of the year and those which are related to the prevailing meteorological conditions. If a prediction system is to provide consistently accurate forecasts then it has to be able to incorporate the effects of each of the factor types outlined above. The prediction system proposed in this thesis achieves this by the use of a neural network that by the application of appropriately classified example sets, can track the varying relationship between the level of demand and key meteorological variables. The influence of supply network changes and calendar related events are accounted for by the use of a rule base of prediction adjusting rules that are built up with reference to past occurrences of similar events. The resulting system is capable of eliminating a significant proportion of the large prediction errors that can lead to non optimal supply network operation.

No description available.

Made available in DSpace on 2016-08-10T10:40:38Z (GMT). No. of bitstreams: 1 Paulo Henrique Borba Florencio.pdf: 1950329 bytes, checksum: bd2a6b0f6c7cf875f0ccbd8d5c9cafcf (MD5) Previous issue date: 2016-03-16 / The constant changes on the national car sales scenario brought one more factor in the search of management of spare parts inventory; the need to reduce the amount invested in stocks and avoid obsolescence. For this purpose, the work aims to evaluate the performance of Artificial Neural Networks in predicting demand for vehicles of engine parts identifying, among the studied networks, which are best suited to each evolution of consumer model and how it applies in each case. Furthermore, it proposes the use of a method of assessment and monitoring of selected models by analyzing the root mean square errors of prediction. The determination of predictive methods with a higher degree of accuracy, is a critical step in the process of inventory management. If the forecast has a low accuracy can generate excess or lack of inventory and this excess, if not properly treated, it can lead to the obsolescence and generate unnecessary costs. To achieve the objective, sought first, the study of the major theoretical aspects of the methods of inventory management processes and demand forecasting. Later, after the selection process two artificial neural networks, Elman network and TDNN Network. To ensure the accuracy of the demands were used four items that differ by the type of evolution of consumption chart, looking for items with constant consumption, increasing, decreasing and items with smaller amounts of observed periods. The results obtained through the use of the proposed methodology, showed that neural networks have the necessary characteristics for your application with a higher degree of accuracy. / As constantes mudanças no cenário nacional de venda de automóveis trouxeram um fator a mais na busca do gerenciamento dos estoques de peças de reposição: a necessidade de diminuir o valor investido em estoques e evitar a obsolescência. Com esse propósito, este trabalho tem por objetivo avaliar o desempenho de Redes Neurais Artificiais na predição de demanda de peças de reposição de veículos automotores identificando, dentre as redes estudadas, quais se adaptam melhor a cada modelo de evolução de consumo e como se aplica em cada caso. Outrossim, propõe a utilização de um método de avaliação e monitoramento dos modelos selecionados através da análise dos erros médios quadráticos da previsão. A determinação de métodos preditivos com maior grau de precisão, constituise em etapa fundamental do processo de gerenciamento de estoques. Se a previsão apresentar uma baixa acurácia, pode-se gerar excesso ou falta de estoques e esse excesso, se não tratado adequadamente, pode culminar em obsolescência e gerar custos desnecessários. Para alcançar o objetivo proposto, buscou-se, em primeiro lugar, o estudo dos principais aspectos teóricos relacionados ao processo de gestão de estoques e aos métodos de previsão de demanda. Posteriormente, segue o processo de seleção de duas redes neurais artificiais, Rede de Elman e Rede TDNN. Para certificar a acurácia das demandas, foram utilizados quatro itens que se diferem pelo tipo de gráfico de evolução de consumo, buscando itens com consumo constante, crescente, decrescente e itens com quantidades menores de períodos observados. Os resultados obtidos, mediante a utilização da metodologia proposta, mostraram que as Redes Neurais possuem as características necessárias para sua aplicação com um grau de acurácia mais elevado.

Os serviços de emergências hospitalares (EH) desempenham um papel fundamental no sistema de saúde, servindo de porta de entrada para hospitais e fornecendo cuidados para pacientes com lesões e doenças graves. No entanto, as EH em todo o mundo sofrem com o aumento da demanda e superlotação. Múltiplos fatores convergem simultaneamente para resultar nessa superlotação, porém a otimização do gerenciamento do fluxo dos pacientes pode auxiliar na redução do problema. Nesse contexto, o tempo de permanência dos pacientes na EH (TPEH) é consolidado na literatura como indicador de qualidade do fluxo de pacientes. O tema desta dissertação é a previsão e gestão da demanda em EH com alto grau de congestionamento, que é abordado através de três artigos científicos. O objeto de estudo é o Hospital de Clínicas de Porto Alegre (HCPA). No primeiro artigo, são aplicados quatro modelos de previsão da procura por atendimento na EH, avaliando-se a influência de fatores climáticos e de calendário. O segundo artigo utiliza a técnica de regressão por mínimos quadrados parciais (PLS – partial least squares) para previsão de quatro indicadores relacionados ao TPEH para hospitais com alto grau de congestionamento. O tempo médio de permanência (TM) na EH resultou em um modelo preditivo com melhor ajuste, com erro médio absoluto percentual (MAPE - mean absolute percent error) de 5,68%. O terceiro artigo apresenta um estudo de simulação para identificação dos fatores internos do hospital que influenciam o TPEH. O número de exames de tomografias e a taxa de ocupação nas enfermarias clínicas e cirúrgicas (ECC) foram as que mais influenciaram. / Emergency departments (ED) play a key role in the health system, serving as gateway to hospitals and providing care for patients with injuries and serious illnesses. However, EDs worldwide suffer from increased demand and overcrowding. Multiple factors simultaneously converge to result in such overcrowding, and the optimization of patient flow management can help reduce the problem. In this context, the length of stay of patients in ED (LSED) is consolidated in the literature as a patient flow quality indicator. This thesis deals with forecast and demand management in EDs with a high degree of congestion. The subject is covered in three scientific papers, all analyzing data from the Hospital de Clínicas de Porto Alegre’s ED. In the first paper we apply four demand forecasting models to predict demand for service in the ED, evaluating the influence of climatic and calendar factors. The second article uses partial least squares (PLS) regression to predict four indicators related to LSED. The mean length of stay in the ED resulted in a model with the best fit, with mean percent absolute error (MAPE) of 5.68%. The third article presents a simulation study to identify the internal hospital factors influencing LSED. The number of CT exams and the occupancy rate in the clinical and surgical wards were the most influential factors.

Esta tese apresenta uma sistemática de seleção dos especialistas mais consistentes e dos fatores de ajuste mais relevantes com vistas ao aprimoramento da acurácia da previsão de demanda gerada por métodos quantitativos. Para tanto, são testados sete modelos quantitativos: Médias Móveis (MM-3, MM-6 e MM-9), Suavização Exponencial Simples e Dupla e o modelo de Holt-Winters multiplicativo e aditivo. O modelo utilizado na previsão quantitativa foi aquele que gerou a melhor aderência aos dados e acurácia preditiva com base nos indicadores R2 e Erro Percentual Médio Absoluto (MAPE), respectivamente, extraídos mediante a quebra da série histórica na proporção 80% (banco de treino) e 20% (banco de teste) para cada produto. Com base nesse critério, tanto o leite UHT quanto o queijo mussarela foram modelados através da Suavização Exponencial Dupla (SED). Na sequência, especialistas e fatores utilizados para ajuste qualitativo da demanda foram selecionados de forma a reter somente os especialistas mais consistentes e os fatores mais influentes para tal fim. O método reteve os 5 especialistas mais consistentes dos 15 inicialmente entrevistados. Dos 23 fatores iniciais, apenas os 13 mais representativos foram retidos. Através da previsão corrigida para o leite UHT, o MAPE foi reduzido de 14,29% para 6,44%. Já previsão ajustada do queijo mussarela possibilitou reduzir o MAPE de 15,25% para 8,72%. / This thesis presents a systematic selection of the most consistent experts and most relevant adjustment factors aimed at improving the accuracy of forecasting demand generated by quantitative methods. For this, seven quantitative models are tested: Moving Averages (MM-3, MM-6 and MM-9), Single and Double Exponential Smoothing and Holt-Winters multiplicative and additive model. The model used in quantitative forecasting was one that generated the best adherence to data and predictive accuracy based on the indicators R2 and Mean Absolute Percentage Error (MAPE), respectively, extracted by breaking the time series in the ratio 80 % (workout bench) and 20% (test bank) for each product . Based on this criterion , both UHT milk and mozzarella cheese were modeled by Double Exponential Smoothing (SED). Further, experts and qualitative factors used to adjust demand were selected so to retain only the most consistent experts and the most influential factors for this purpose. The method retained the 5 most consistent experts of the 15 interviewed initially. Of the 23 initial factors, only the 13 most significant were retained. Through prediction corrected for UHT milk the MAPE was reduced from 14.29 % to 6.44 %. It had forecast adjusted mozzarella cheese possible to reduce the MAPE of 15.25% to 8,72.

This thesis explores forecasting for intermittent demand requirements. Intermittent demand occurs at random, with some time periods showing no demand. In addition, demand, when it occurs, may not be for a single unit or a constant size. Consequently, intermittent demand creates significant problems in the supply and manufacturing environment as far as forecasting and inventory control are concerned. A certain confusion is shared amongst academics and practitioners about how intermittent demand (or indeed any other demand pattern that cannot be reasonably represented by the normal distribution) is defined. As such, we first construct a framework that aims at facilitating the conceptual categorisation of what is termed, for the purposes of this research, “non-normal” demand patterns. Croston (1972) proposed a method according to which intermittent demand estimates can be built from constituent elements, namely the demand size and inter-demand interval. The method has been claimed to provide unbiased estimates and it is regarded as the “standard” approach to dealing with intermittence. In this thesis we show that Croston’s method is biased. The bias is quantified and two new estimation procedures are developed based on Croston’s concept of considering both demand sizes and inter-demand intervals. Consequently the issue of variability of the intermittent demand estimates is explored and finally Mean Square Error (MSE) expressions are derived for all the methods discussed in the thesis. The issue of categorisation of the demand patterns has not received sufficient academic attention thus far, even though, from the practitioner’s standpoint it is appealing to switch from one estimator to the other according to the characteristics of the demand series under concern. Algebraic comparisons of MSE expressions result in universally applicable (and theoretically coherent) categorisation rules, based on which, “non-normal” demand patterns can be defined and estimators be selected. All theoretical findings are checked via simulation on theoretically generated demand data. The data is generated upon the same assumptions considered in the theoretical part of the thesis. Finally, results are generated using a large sample of empirical data. Appropriate accuracy measures are selected to assess the forecasting accuracy performance of the estimation procedures discussed in the thesis. Moreover, it is recognised that improvements in forecasting accuracy are of little practical value unless they are translated to an increased customer service level and/or reduced inventory cost. In consequence, an inventory control system is specified and the inventory control performance of the estimators is also assessed on the real data. The system is of the periodic order-up-to-level nature. The empirical results confirm the practical validity and utility of all our theoretical claims and demonstrate the benefits gained when Croston’s method is replaced by an estimator developed during this research, the Approximation method.

Considering the international scenario, in a recent past, the electrical industry was based on the concepts of monopolistic concessions and vertical utilities structures. In Brazil, until recently, the electricity companies were all governmental properties that served restricted monopolized areas. In a similar manner, in the United States, monopolies for certain concession areas were assigned to vertically integrated electric utilities. This monopolistic portfolio brought to the industry, in a generic sense, a lack in the interface between companies and consumers. This fact established a low capacity of obtaining consumer's information and consequently, a low capacity of developing precise demand forecasts. Lately, the industry of electrical energy around the world has passed through immense structural changes, not only in developed countries, but also in developing countries. In this new environment, competition and private capital are fundamental agents. Now, demand forecasting represents a key factor to support decision-making for planning strategies of electricity utilities. In addition to immediate potential benefits to commercial decisions, a framework for electricity demand forecasting can help to take actions aiming to develop more precise yearly budgets as well as to make accurate investments in the infrastructure expansion. With the objective of improving the supply chain perception of the electricity industry, this work analyzes the electricity industry and develops a mathematical tool to accurately support the decision-making process of electricity utilities. CPFL Energy Co., a holding that controls companies and private enterprises in the generation area, electric power distribution, and trading in Brazil, was chosen as the case of study. CPFL Energy Co.'s supply chain was studied to find out the right explanatory variables and an electricity forecasting mathematical model was created through the stepwise regression procedure.

A model is proposed to provide inventory and maintenance guidance for a system of operating parts. This model is capable of handling a system with multiple operating components, unknown part lifetime failure distribution, and separately maintained parts. In this model, part reliability characteristics are used along with system costs to predict the required stocking levels and part replacement times. Two maintenance strategies are presented that have the unique characteristic of allowing flexible scheduling of replacements. A case study is completed comparing developed stocking policies to an existing policy. An estimation selection method is introduced and fit into the model for computing Weibull distribution parameters when part reliability is not well known. An algorithm is displayed that describes the implementation of the system model and data from practical case scenarios are conducted using this algorithm.

La presente investigación resalta la importancia de aplicar una mejora en la gestión de inventarios para una empresa dedicada a la fabricación y comercialización de prendas de vestir, específicamente camisas. El objetivo principal de la investigación es reducir los costos generados por una deficiente gestión de inventarios, es decir, reducir costos por sobre stock de inventarios y baja rotación de inventarios, además planificar el requerimiento de materiales. El primer capítulo describe los artículos e investigaciones relacionadas a la propuesta en los últimos años. Además, se explica los conceptos relacionados a la gestión de inventarios, pronósticos, requerimiento de materiales, y otras herramientas como matriz de kraljic y el modelo de cantidad económica. En el segundo capítulo se brinda una visión general de la empresa en estudio, se describe y reconoce los problemas existentes, priorizando el más significativo a fin de identificar las causas que lo originan para posteriormente seleccionar la mejor alternativa de solución. En el tercer capítulo se presenta la propuesta de mejora para la gestión de inventarios de acuerdo a la información obtenida de la empresa cuya justificación está en el beneficio que se podría obtener. Finalmente, se expone las conclusiones y recomendaciones para la propuesta de mejora para la gestión de inventarios. / This thesis below highlights the importance of developing a proposal for improvement inventory management system for a company dedicated to manufacture and marketing men’s clothing, specifically shirts. The main objective of the investigation is to reduce the costs originated by a deficient inventory management, which means, reduce the costs by excessive stock and low inventory turnover, and material requirement planning. The first chapter describes the different research articles in the last years that are related to the proposal. Besides, the concepts related to the inventory management are explained, as well as demand forecasting, MRP and some other inventory management tools. The second chapter presents a general vision of the company; describing and recognizing the existing problems, for later prioritize the most significant problem in order to identify the causes that have its origin to finally select the best solution. The third chapter introduces the proposal for improvement inventory system according to the information given by the company and the benefit that can be acquired. Finally, conclusions and recommendations will be presented for the proposal. / Trabajo de Investigación