Use of AI for Energy Management and Predictive Maintenance

Artificial Intelligence (AI) is rapidly transforming various sectors, including the energy industry, redefining processes and optimizing operational efficiency. It is therefore essential to understand how AI can integrate into this domain and what impact it may have on a key figure such as the energy manager. This thesis aims to achieve two main objectives: first, to explore the use of intelligent and customized chatbots to support and enhance the efficiency of energy managers; second, to develop a predictive maintenance model based on Machine Learning techniques to analyze the performance degradation of two photovoltaic (PV) plants, with a particular focus on progressive soiling and anomalies. The chatbots developed in this research are based on two main platforms: ChatGPT and Botpress. The use of OpenAI's custom GPTs enabled rapid personalization through prompt engineering, making chatbot creation, adaptation, and usage highly user-friendly and flexible. On the other hand, Botpress stands out for its advanced capabilities in managing conversational flows and integrating with other platforms, making it a more suitable solution for commercial deployment and enterprise applications. Data analysis was conducted using Python and a set of advanced metrics to assess plant operating conditions and monitor their performance evolution. A key role was played by the CUSUM of residuals, normalized to generated power, which allowed for the identification of anomalies and performance degradation over time. To ensure reliable analysis, a structured pipeline was developed for data pre-processing, model training, and result evaluation, automating monitoring and improving decision-making processes. The integration of historical data analysis and predictive models enabled a more accurate assessment of degradation, reducing uncertainty and optimizing maintenance interventions. The use of advanced dashboards with various types of automated alerts allows for a clear and immediate visualization of plant conditions, enabling energy managers to detect anomalous trends and take timely action. This system not only improves data interpretability but also provides practical support for implementing evidence-based strategies, making plant management more proactive and efficient. The results demonstrate that the integration of Machine Learning, intelligent chatbots, and advanced dashboards significantly enhances the monitoring and management of PV plants. The CUSUM of residuals, combined with automated alerts and interactive analysis, has made it possible to precisely identify performance degradation and optimize maintenance. Additionally, customized chatbots provide immediate access to data, supporting energy managers in operational decision-making. This solution not only reduces costs and improves efficiency but also paves the way for further advancements in predictive maintenance and automation within the energy sector.