Accurate collection system models rely on well-calibrated hydrology parameters that can closely simulate the inflow and infiltration of rainwater into the system under various conditions. Traditional hydrology methods use simple statistical models, such as diurnal patterns, or simplified groundwater models to approximate the complex process of RDII (rain derived inflow and infiltration).
However, these methods have several limitations, including the inability to directly measure most of the necessary parameters and the oversimplification of the real RDII process. As a result, manual calibration is required, which can be time-consuming and error prone.
With the increasing adoption of the Internet of Things (IoT) and the advancement of artificial intelligence (AI), more utilities are collecting flow monitoring data at permanent locations and using AI models to analyze this data. These models can overcome the limitations of traditional methods by automating the training process and allowing the model to learn the real hydrological process directly from the growing dataset.
This presentation aims to share with the audience the basic ideas behind artificial intelligence and machine learning methods, and the typical workflow of preparing the data, training the model, and then evaluating the models. To help illustrate key concepts, we will train AI models using flow data simulated using synthetic rainfall, thus we can stay focused on understanding how AI models can learn the true hydrological process, how does the amount, accuracy, variety of the training data and other factors that can impact the results.
Click here to watch recorded presentation on YouTube.