Multi-source-data-driven microgrids reliability analysis via power supply chain using deep learning

Abstract

Microgrid reliability is the ability to maintain a stable energy supply in a variable environment. However, such an environment (wind direction, temperature, humidity, pressure, and wind speed) renders the power supply with randomness, intermittency, and volatility. To ensure power stability in variable environments, a data-driven microgrid (DDMG) reliability analysis method is proposed based on the power supply chain (PSC) model, which fully considers the data-dependent output power. Firstly, a convolutional neural network-support vector machine (CNN-SVM) model is developed to effectively fuse multi-source data features. Secondly, a temporal convolutional network-bidirectional gated recurrent unit (TCN-BiGRU) approach is introduced to capture temporal dependencies to predict equipment states. The above two deep learning models provide accurate input values for reliability assessment. Then, a reliability assessment model is established based on the PSC model, complemented by an importance-measure-based reliability improvement strategy. Finally, the feasibility of methodology is validated with a case. The results show that compared with the traditional methods, the classification accuracy of CNN-SVM is up to 98.9747 %, the R² of TCN-BiGRU is up to 96.1962 %, and the recovered ranking based on the importance measure markedly and stably improves the reliability, which would guide microgrid reliability design.