Publicação
Load Forecasting Benchmark for Smart Meter Data
| Resumo: | Actual integration of high-tech devices brings opportunities for better monitoring, management and control of low voltage networks. In this new paradigm, efficient tools should cope with the great amount of dispersed and considerably distinct data to support smarter decisions in almost real time. Besides the use of tools to enable an optimal network reconfiguration and integration of dispersed and renewable generation, the impact evaluation of integrating storage systems, accurate load forecasting methods must be found even when applied to individual consumers (characterized by the high presence of noise in time series). As this effort becomes providential in the smart grids context, this article compares three different approaches: one based on Kernel Density Estimation, an alternative based on Artificial Neural Networks and a method using Support Vector Machines. The first two methods revealed unequivocal benefits when compared to a Naive method consisting of a simple reproduction of the last available day. |
|---|---|
| Autores principais: | Viana, João |
| Outros Autores: | Bessa, Ricardo J.; Sousa, João |
| Assunto: | Forecasting low voltage smart meter density estimation neural networks |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | comunicação em conferência |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Instituto Politécnico de Leiria |
| Idioma: | inglês |
| Origem: | IC-online |
| Resumo: | Actual integration of high-tech devices brings opportunities for better monitoring, management and control of low voltage networks. In this new paradigm, efficient tools should cope with the great amount of dispersed and considerably distinct data to support smarter decisions in almost real time. Besides the use of tools to enable an optimal network reconfiguration and integration of dispersed and renewable generation, the impact evaluation of integrating storage systems, accurate load forecasting methods must be found even when applied to individual consumers (characterized by the high presence of noise in time series). As this effort becomes providential in the smart grids context, this article compares three different approaches: one based on Kernel Density Estimation, an alternative based on Artificial Neural Networks and a method using Support Vector Machines. The first two methods revealed unequivocal benefits when compared to a Naive method consisting of a simple reproduction of the last available day. |
|---|