Load forecasting by ANN

A description of artificial neural networks (ANNs) is given. Reasons why interest in ANNs has increased are discussed. Steps used to train neural networks (NNs) are described, including gathering and normalizing data, selecting NN architecture, training and testing networks, selecting alternative network architectures, and performing additional training. A case study in load forecasting performed by Associated Electric Cooperative, Inc. (AECI) is discussed. The ANN method was chosen for its ability to learn historical data, draw inferences, and adapt to new situations. The software used to simulate the ANN was developed in-house, allowing a custom interface to be built to the specifications of the system dispatchers. How data is selected, the training process, guidelines for designing neuron configurations, and error tolerances are discussed.<
>