Desarrollo de software de predicción de carga de energía basado en Machine Learning

Abstract

This research focuses on the development of an electrical power charge prediction system to validate power behavior using Convolutional Neural Networks (CNN), Artificial Neural Networks (ANN) and Long and Short Term Memory (LSTM). Metrics such as the Mean Absolute Percentage Error (MAPE) and the coefficient of determination (R2) are used to assess the accuracy of the predictions. The methodology used in the research is mixed, combining quantitative and qualitative approaches. The CRISP-DM methodology is followed, which consists of different stages such as understanding the business, understanding the data, data preparation, modeling, evaluation and use. The research approach is experimental, where variables are manipulated and monitored according to the established metrics. In addition, the study is carried out in a descriptive context, presenting and analyzing the relevant variables. The main objective of the work is to develop an electric power charge prediction system using the CNN, ANN and LSTM neural networks, and to evaluate its performance using the MAPE and R2 metrics. The purpose is to improve the predictability of electrical power load and provide reliable and accurate results. Tests and experiments are carried out to compare the performance of the different models and techniques used. It seeks to identify the most efficient and effective model in the prediction of the electric power load. In summary, the thesis focuses on the development of an electrical energy prediction system using CNN, ANN and LSTM neural networks. Metrics such as MAPE and R2 are used, and the CRISP-DM methodology is followed. The research is carried out using a mixed approach, with an experimental and descriptive approach. The goal is to improve the accuracy and reliability of electrical power load forecasting, which may have important applications in efficient power planning and management.