This document evaluates the performance of deep learning algorithms and data preprocessing for demand power prediction. Experiments were conducted using raw and preprocessed daily electricity load, temperature, and weather data from Australia. Recurrent neural networks and convolutional neural networks generally had better prediction accuracy when trained on preprocessed rather than raw data. Preprocessing scaled the temperature data and increased the data points, leading to more stable and accurate results across all tested algorithms. The best performance was achieved using a recurrent neural network on preprocessed data. Further analysis of extreme learning machine algorithms was recommended.