| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 6.88 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Esta dissertação apresenta o desenvolvimento de uma ferramenta de previsão de consumos de água
mediante a utilização de inteligência artificial, nomeadamente de modelos de redes neuronais artificiais.
Uma adequada previsão de consumos de água a curto, médio e longo prazo possibilita às empresas de
abastecimento e distribuição de água uma informação imprescindível para estimar a capacidade de
planeamento, atividades de manutenção, melhorias do sistema e otimização da operação de sistemas
elevatórios e de tratamento.
São examinados, para além das redes neuronais artificiais, outros modelos estatístico-matemáticos
aplicados à previsão dos consumos, como o modelo ARIMA, baseado em médias móveis integradas
autorregressivas.
Em virtude do escasso número de trabalhos nesta área, considerou-se necessário aprofundar os
conteúdos teóricos em matéria de redes neuronais e séries temporais, para depois aplicar uma
metodologia própria que permitiu implementar o modelo na linguagem Python, sendo também utilizados
outros softwares para explorar a capacidade das redes neuronais artificiais, como o Neural Network
Toolbox para Matlab e o modelo ARIMA.
Como caso de estudo, a metodologia foi aplicada aos registos de consumos do concelho de Arouca, cujo
abastecimento em alta corresponde à Águas do Douro e Paiva (AdDP), do grupo Águas de Portugal. Como
dados de base foram considerados os valores de consumo a cada meia hora, bem como dados de
precipitação e temperatura.
Os resultados da aplicação de redes neuronais foram satisfatórios, uma vez que o modelo aplicado sobre
a série diária produziu previsões com uma precisão próxima de 96%, enquanto o modelo aplicado à série
horária forneceu precisões próximas de 87 %. Este rigor nas previsões demonstra a efetividade da
aplicação das redes neuronais na previsão dos consumos de água em vários horizontes de tempo.
This thesis presents the development of a tool for water consumption forecasting through the use of artificial intelligence, especially of artificial neuron models. An adequate prediction of water consumption in the short, medium and long term provides essential information for water supply and distribution companies planning capacity, maintenance activities, system improvements and optimization of pumping treatment operation. In addition to the artificial neural networks, other statistical and mathematical models are applied to the forecast of the time series of consumption, such as the ARIMA model, based on autoregressive integrated moving averages. Due to the small number of available works in this area, it was necessary to deepen the theoretical contents regarding neural networks and time series, for the later development of a specific methodology and the respective implementation using Python language. Moreover, the capacity of artificial neural networks and of the ARIMA model were also explored using Neural Network Toolbox for Matlab. As case study, the methodology was applied to the consumption records for the municipality of Arouca, whose water supply is delivered by the Águas do Douro and Paiva (AdDP) company, from Águas de Portugal group. The main data used was water consumption, precipitation and temperature records available for the area. The results of the application of neural networks were satisfactory, since the model applied over a series of daily records produced predictions with an accuracy close to 96%, while the model applied to the hourly series provided accuracies close to 87%. This rigor in predictions demonstrates an effectiveness of the application of neuronal networks in the face of the problem for water consumption forectasting in various periods of time.
This thesis presents the development of a tool for water consumption forecasting through the use of artificial intelligence, especially of artificial neuron models. An adequate prediction of water consumption in the short, medium and long term provides essential information for water supply and distribution companies planning capacity, maintenance activities, system improvements and optimization of pumping treatment operation. In addition to the artificial neural networks, other statistical and mathematical models are applied to the forecast of the time series of consumption, such as the ARIMA model, based on autoregressive integrated moving averages. Due to the small number of available works in this area, it was necessary to deepen the theoretical contents regarding neural networks and time series, for the later development of a specific methodology and the respective implementation using Python language. Moreover, the capacity of artificial neural networks and of the ARIMA model were also explored using Neural Network Toolbox for Matlab. As case study, the methodology was applied to the consumption records for the municipality of Arouca, whose water supply is delivered by the Águas do Douro and Paiva (AdDP) company, from Águas de Portugal group. The main data used was water consumption, precipitation and temperature records available for the area. The results of the application of neural networks were satisfactory, since the model applied over a series of daily records produced predictions with an accuracy close to 96%, while the model applied to the hourly series provided accuracies close to 87%. This rigor in predictions demonstrates an effectiveness of the application of neuronal networks in the face of the problem for water consumption forectasting in various periods of time.
Description
Keywords
Abastecimento Consumos de água Inteligência artificial Redes previsão Otimização Supply Water consumption Artificial intelligence ANN forecasting Optimization