Develop an algorithm to calculate the area of a chimney's shadow on a tilted roof in the Himalayas. The algorithm must consider the specific location (latitude/longitude), the roof's slope and orientation, and the chimney's dimensions. The goal is to provide the shadow area for any given date and time, and to find the date and time when the smallest shadow occurs during the year. Build a machine learning or deep learning model to detect anomalies in a photovoltaic (PV) power system. Using provided time-series data (expected and actual power output), the model should identify when the system is underperforming due to factors like shading or component failure. The task involves explaining the data analysis, the chosen model, the training process for unlabeled data, and visualizing the detected anomalies.