Thesis (Ph. D.)--University of Wisconsin--Madison, 1947. / Typescript. Vita. eContent provider-neutral record in process. Description based on print version record. Includes bibliographical references (leaves 147-151).

An electricity demand profile is a graph showing the amount of electricity used by customers over a unit of time. It shows the variation in electricity demand versus time. In the demand profiles, the shape of the graph is of utmost importance. The variations in demand profiles are caused by many factors, such as economic and en- vironmental factors. These variations may also be due to changes in the electricity use behaviours of electricity users. This study seeks to model daily profiles of energy demand in South Africa with a model which is a composition of two de Moivre type models. The model has seven parameters, each with a natural interpretation (one parameter representing minimum demand in a day, two parameters representing the time of morning and afternoon peaks, two parameters representing the shape of each peak, and two parameters representing the total energy per peak). With the help of this model, we trace change in the demand profile over a number of years. The proposed model will be helpful for short to long term electricity demand forecasting.

Energy demand forecasting, and specifically electricity demand forecasting, is a fun-damental feature in both industry and research. Forecasting techniques assist all electricity market participants in accurate planning, selling and purchasing decisions and strategies. Generation and distribution of electricity require appropriate, precise and accurate forecasting methods. Also accurate forecasting models assist producers, researchers and economists to make proper and beneficial future decisions. There are several research papers, which investigate this fundamental aspect and attempt var-ious statistical techniques. Although weather and economic effects have significant influences on electricity demand, in this study they are purposely eliminated from investigation. This research considers calendar-related effects such as months of the year, weekdays and holidays (that is, public holidays, the day before a public holiday, the day after a public holiday, school holidays, university holidays, Easter holidays and major religious holidays) and includes university exams, general election days, day after elections, and municipal elections in the analysis. Regression analysis, cate-gorical regression and auto-regression are used to illustrate the relationships between response variable and explanatory variables. The main objective of the investigation was to build forecasting models based on this calendar data only and to observe how accurate the models can be without taking into account weather effects and economic effects, hence weather neutral models. Weather and economic factors have to be forecasted, and these forecasts are not so accurate and calendar events are known for sure (error-free). Collecting data for weather and economic factors is costly and time consuming, while obtaining calendar data is relatively easy.

In terms of economic development, internal availability of energy in a region means the capability to produce essential goods and services for the improvement in the quality of life of all the economic agents. Economic development consists in large part of harnessing increasing amounts of energy for productive purposes or by making more efficient use of available energy resources. In this dissertation, the future electricity supply and demand interactions are examined for seven countries in West Africa: Benin, Ghana, Ivory Coast, Niger, Nigeria, Togo, and Upper Volta. A description of the primary energy resources (coal, hydro, natural gas, and oil) available in each country is presented. The future demands for electricity in the medium term (1980-1989) are projected through econometric models developed in the study. Two sectorial models for each country's economy, the residential sector, and the commercial and industrial sector, are presented. Multiple regression analysis is applied in the estimation of all demand equations. Major determinants for electricity demand used in the estimation for the residential sector were average price of electricity, real personal income, and the number of households with access to electricity. Data on these variables were obtained from international organisations such as the United Nations and from government publications for the period 1960-1977. Each of these determinants was found to be significant for most countries; however, their relative importance differ across countries. Similarly, average price of electricity, real output, and employment were major determinants used and found to be significant in the demand for electricity in the commercial and industrial sector of all countries. Price and income elasticities were obtained from the estimated equations. A general multi-region supply model was developed to structure the future electricity supply possibilities in the countries involved. The objective of this model, using linear programming, was to seek the least-cost combination of resources (primary energy, capital, and technology) for the production of electricity. The impacts of various levels of resource availability on average cost of electricity were examined for each country, and for joint development efforts using a non-integer, deterministic, linear version of the general model. The application of the supply and demand models to West Africa over the decade to 1989 reveals that except for Nigeria, all countries in the region will require fossil fueled systems to supply additional demands for electricity, because all hydro resources would have to be exploited by the mid-1980s. This will lead to higher costs in producing electricity. However, Nigeria is expected to have excess electrical energy if plans initiated in its third development plan are completed. The extension of transmission lines between Nigeria and Benin could effectively distribute the relatively cheaper energy from Nigeria to other countries, since adequate transmission network already exists between most of the countries.

Typescript (photocopy). / Digitized by Kansas Correctional Industries

The design, simulation and evaluation of an intelligent and integrated load management system is presented in this dissertation. The objective of this research was to apply modern computer and communication technology to influence customer use of electricity in ways that would produce desired changes in the utility's load shape. Peak clipping (reduction of peak load) using direct load control is the primary application of this research. The prototype computerized communication and control package developed during this work has demonstrated the feasibility of this concept. The load management system consists of a network of computers, data and graphics terminals, controllers, modems and other communication hardware, and the necessary software. The network of interactive computers divides the responsibility of monitoring of meteorological data, electric load, and performing other functions. These functions include: data collection, processing and archiving, load forecasting, load modeling, information display and alarm processing. Each of these functions requires a certain amount of intelligence depending on the sophistication and complication of that function. Also, a high level of reliability has been provided to each function to guarantee an uninterrupted operation of the system. A full scale simulation of this concept was carried out in the laboratory using five microcomputers and the necessary communication hardware. An important and integral part of the research effort is the development of the short-term load forecast, load models and the decision support system using rule-based algorithms and expert systems. Each of these functions has shown the ability to produce more accurate results compared to classical techniques while at the same time requiring much less computing time and historical data. Development of these functions has made the use of microcomputers for constructing an integrated load management system possible and practical. Also, these functions can be applied for other applications in the electric utility industry and maintain their importance and contribution. In addition to that, the use of rule-based algorithms and expert systems promises to yield significant benefits in using microcomputers in the load management area. / Ph. D.

O presente trabalho apresenta a previsão de demanda de água em sistemas urbanos de abastecimento através de Rede Neural Artificial (RNA) utilizando dados de consumo de água e variáveis meteorológicas e socioambientais. A RNA utilizada foi uma de três camadas chamada de rede de múltiplas camadas alimentadas adiante com o algoritmo de treinamento LLSSIM (Hsu et al., 1996). Neste estudo, foram utilizados os dados de consumo de água (SABESP) e meteorológicos (IAG/USP) para o período de 2001 a 2005 para Região Metropolitana de São Paulo (RMSP). As variáveis socioambientais e meteorológicas que podem afetar o consumo de água foram analisadas. A ETA Cantareira e o setor Itaim Paulista foram utilizados para avaliar a relação entre o consumo e as variáveis antrópicas e meteorológicas para o ano de 2005. Esses conjuntos de dados foram utilizados para o treinamento, o teste e a previsão da RNA. Para a ETA Cantareira, foram criados 8 modelos e para o setor Itaim Paulista 57, sendo que os modelos 9 a 57 correspondem à previsão ideal. O desempenho dos modelos foi avaliado pelo o erro médio, erro médio absoluto, erro médio quadrático, o coeficiente de correlação, exatidão, viés, POD, FAR, CSI e POFD. Para a ETA Cantareira o melhor desempenho ocorreu para a média de 12 horas e para o Itaim Paulista a média de 6 horas. Na previsão ideal observou-se que a memória do sistema é um fator importante, principalmente quando se tem dois intervalos de tempo anterior. Os resultados mostraram a importância da memória, pois ela ajuda a melhorar o desempenho da previsão A previsão horária foi obtida com níveis de erros aceitáveis. Comparando os resultados de todas as configurações dos modelos, observou-se que há uma tendência para pequenos erros. Finalmente, conclui-se que o método proposto pode ser utilizado para previsão de consumo obtendo uma boa previsão. / This work is concerned with the prediction of water demand in urban water supply systems using water consumption, meteorological and socioenvironmental variables in an Artificial Neural Network (ANN) system. The ANN is a three layer feed-forward network with the LLSSIM training algorithm (Hsu et. al., 1996). In this study, water consumption (SABESP) and meteorological (IAG USP) data sets between 2001 and 2005 were used for studying the Metropolitan Area São Paulo (MASP). Possible socio-environmental and meteorological conditions affecting water consumption in the MASP were analyzed. Two water treatment stations (ETA), namely, Cantareira and the Itaim Paulista were used to evaluate the relationship between water consumption against anthropic and meteorological conditions for the year 2005. These data sets were also used for training, testing and forecasting of the water consumption model with the ANN. For the Cantareira ETA, 8 model configurations were tested and 57 for the Itaim Paulista ETA. In this late case, configurations 9 to 57 were for ideal forecasts. The various model configurations were evaluated by the mean error, mean absolute error and mean square root error, correlation coefficient, bias, POD, FAR, CSI e POFD. The best performance for the Cantareira ETA was obtained for a 12-hour average of the input variables, and for the Itaim Paulista ETA, for the 6-hour average. The ANN model configurations fed with variables of previous three times steps (memory) performed best, followed by two previous time steps. The results indicate the importance of these memory to improving the performance of the forecasting. The hourly forecasting was obtained with acceptable error levels. Comparing the results of all model configurations, there is an overall tendency for minor errors. The proposed method can be used to demand forecast a good prediction.

O presente trabalho apresenta a previsão de demanda de água em sistemas urbanos de abastecimento através de Rede Neural Artificial (RNA) utilizando dados de consumo de água e variáveis meteorológicas e socioambientais. A RNA utilizada foi uma de três camadas chamada de rede de múltiplas camadas alimentadas adiante com o algoritmo de treinamento LLSSIM (Hsu et al., 1996). Neste estudo, foram utilizados os dados de consumo de água (SABESP) e meteorológicos (IAG/USP) para o período de 2001 a 2005 para Região Metropolitana de São Paulo (RMSP). As variáveis socioambientais e meteorológicas que podem afetar o consumo de água foram analisadas. A ETA Cantareira e o setor Itaim Paulista foram utilizados para avaliar a relação entre o consumo e as variáveis antrópicas e meteorológicas para o ano de 2005. Esses conjuntos de dados foram utilizados para o treinamento, o teste e a previsão da RNA. Para a ETA Cantareira, foram criados 8 modelos e para o setor Itaim Paulista 57, sendo que os modelos 9 a 57 correspondem à previsão ideal. O desempenho dos modelos foi avaliado pelo o erro médio, erro médio absoluto, erro médio quadrático, o coeficiente de correlação, exatidão, viés, POD, FAR, CSI e POFD. Para a ETA Cantareira o melhor desempenho ocorreu para a média de 12 horas e para o Itaim Paulista a média de 6 horas. Na previsão ideal observou-se que a memória do sistema é um fator importante, principalmente quando se tem dois intervalos de tempo anterior. Os resultados mostraram a importância da memória, pois ela ajuda a melhorar o desempenho da previsão A previsão horária foi obtida com níveis de erros aceitáveis. Comparando os resultados de todas as configurações dos modelos, observou-se que há uma tendência para pequenos erros. Finalmente, conclui-se que o método proposto pode ser utilizado para previsão de consumo obtendo uma boa previsão. / This work is concerned with the prediction of water demand in urban water supply systems using water consumption, meteorological and socioenvironmental variables in an Artificial Neural Network (ANN) system. The ANN is a three layer feed-forward network with the LLSSIM training algorithm (Hsu et. al., 1996). In this study, water consumption (SABESP) and meteorological (IAG USP) data sets between 2001 and 2005 were used for studying the Metropolitan Area São Paulo (MASP). Possible socio-environmental and meteorological conditions affecting water consumption in the MASP were analyzed. Two water treatment stations (ETA), namely, Cantareira and the Itaim Paulista were used to evaluate the relationship between water consumption against anthropic and meteorological conditions for the year 2005. These data sets were also used for training, testing and forecasting of the water consumption model with the ANN. For the Cantareira ETA, 8 model configurations were tested and 57 for the Itaim Paulista ETA. In this late case, configurations 9 to 57 were for ideal forecasts. The various model configurations were evaluated by the mean error, mean absolute error and mean square root error, correlation coefficient, bias, POD, FAR, CSI e POFD. The best performance for the Cantareira ETA was obtained for a 12-hour average of the input variables, and for the Itaim Paulista ETA, for the 6-hour average. The ANN model configurations fed with variables of previous three times steps (memory) performed best, followed by two previous time steps. The results indicate the importance of these memory to improving the performance of the forecasting. The hourly forecasting was obtained with acceptable error levels. Comparing the results of all model configurations, there is an overall tendency for minor errors. The proposed method can be used to demand forecast a good prediction.

published_or_final_version / Urban Studies / Master / Master of Social Sciences

A previsão do consumo de energia elétrica do Brasil é muito importante para os órgãos reguladores do setor. Uma série de metodologias têm sido utilizadas para a projeção desse consumo. Destacam-se os modelos de regressão com dados em painel, modelos de cointegração e defasagem distribuída, modelos estruturais de séries temporais e modelos de Box & Jenkins de séries temporais, dentre outros. Neste trabalho estimar-se um modelo de previsão do consumo comercial, industrial e residencial de energia brasileiro por meio de modelos SARMAX. Nesses modelos o consumo de energia pode ser estimado por meio de uma regressão linear múltipla considerando diversas variáveis macroeconômicas como variáveis explicativas. Os resíduos desse modelo são explicados por meio de um modelo de Box & Jenkins. Neste estudo realiza-se uma pesquisa bibliográfica sobre fatores que influenciam no consumo de energia elétrica e levantam-se variáveis proxies para prever este consumo no Brasil. Utiliza-se uma base de dados mensal no período entre Janeiro de 2003 e Setembro de 2010 para construção de cada um dos três modelos de previsão citados. Utilizase uma amostra de validação de Outubro de 2010 até Fevereiro de 2011. Realiza-se a avaliação dos modelos estimados em termos de adequação às premissas teóricas e ao desempenho nas medidas de acurácia MAPE, RMSE e coeficiente de determinação ajustado. Os modelos estimados para o consumo de energia elétrica dos setores comercial, industrial e residencial obtêm um MAPE de 2,05%, 1,09% e 1,27%; um RMSE de 144,13, 185,54 e 158,40; e um coeficiente de determinação ajustado de 95,91%, 93,98% e 96,03% respectivamente. Todos os modelos estimados atendem os pressupostos de normalidade, ausência de autocorrelação serial e ausência de heterocedasticidade condicionada dos resíduos. Os resultados confirmaram a viabilidade da utilização das variáveis macroeconômicas testadas para estimar o consumo de energia elétrica por setores e a viabilidade da metodologia para a previsão destas séries na amostra de dados selecionada. / The prediction of electricity consumption in Brazil is very important to the industry regulators. A number of methodologies have been used for the projection of this consumption. Noteworthy are the regression models with data in panel, co-integration and distributed lag models, time series structural models and Box & Jenkins time series models among others. In this work we intend to estimate a forecasting model of the Brazilian commercial, industrial and residential consumption of energy by means of SARMAX models. In these models the power consumption can be estimated by a multiple linear regression considering various macro-economic variables as explanatory variables. The residues of this model are explained by a Box & Jenkins model. In this study it is carried out a bibliographic research on factors that influence energy consumption and proxy variables are risen to predict the consumption in Brazil. The consumption of electricity is estimated for the commercial, industrial and residential sectors. It is used a monthly data base over the period between January 2003 and September 2010 for the construction of each of the three prediction models mentioned. It is used a validation sample from October 2010 to February 2011. It is carried out the assessment of the estimated models in terms of compliance with the theoretical premises and the performance on measures of accuracy MAPE, RMSE and adjusted determinant coefficient. The estimated models for the energy consumption of commercial, industrial and residential sectors obtain a MAPE of 2.05%, 1.09% and 1.27%; a RMSE of 144.13, 185.54 and 158.40; and a adjusted determinant coefficient of 95.91%, 93.98% and 96.03% respectively. All estimated models satisfy the assumptions of normality, absence of serial autocorrelation and absence of conditioned heteroscedasticity of the residues. The results confirmed the viability of the usage of the macroeconomic variables tested to estimate the energy consumption by sector and the viability of the methodology for the prediction of these series in the selected data sample.