Short-term demand forecast using a bank of neural network models trained using genetic algorithms for the optimal management of drinking water networks
Efficient management of a drinking water network reduces the economic costs related to water production and transport (pumping). Model predictive control (MPC) is nowadays a quite well-accepted approach for the efficient management of the water networks because it allows formulating the control prob...
| Autores: | , , , |
|---|---|
| Tipo de recurso: | artículo |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2017 |
| País: | España |
| Institución: | Consejo Superior de Investigaciones Científicas (CSIC) |
| Repositorio: | DIGITAL.CSIC. Repositorio Institucional del CSIC |
| OAI Identifier: | oai:digital.csic.es:10261/166485 |
| Acceso en línea: | http://hdl.handle.net/10261/166485 |
| Access Level: | acceso abierto |
| Palabra clave: | Water distribution network Model predictive control Artificial neural networks Demand forecast Genetic algorithms |
| Sumario: | Efficient management of a drinking water network reduces the economic costs related to water production and transport (pumping). Model predictive control (MPC) is nowadays a quite well-accepted approach for the efficient management of the water networks because it allows formulating the control problem in terms of the optimization of the economic costs. Therefore, shortterm forecasts are a key issue in the performance of MPC applied to water distribution networks. However, the short-term horizon demand forecast in a horizon of 24 hours in an hourly based scale presents some challenges as the water consumption can change from one day to another, according to certain patterns of behavior (e.g., holidays and business days). This paper focuses on the problem of forecasting water demand for the next 24 hours. In this work, we propose to use a bank of models instead of a single model. Each model is designed for forecasting one particular hour. Hourly models use artificial neural networks. The architecture design and the training process are performed using genetic algorithms. The proposed approach is assessed using demand data from the Barcelona water network. |
|---|