Real-Time Multi-Agent Based Flood Forecasting and Warning System Model: A Malaysia Perspectives

Abstract

The development of a precise flood forecasting methodology necessitates the utilization of an automated data collection system for the examination of a comprehensive range of hydrographic catchment parameters that are continuously monitored. Monitoring river basins is a topic of significant strategic importance. In recent years, researchers have introduced several cutting-edge technologies to enhance this process, including the utilization of artificial intelligence (AI). Notably, AI has been applied in various techniques such as knowledge-based systems, agent-based modelling, and neural networks. These AI-based approaches have shown promise in improving the monitoring and management of river basins. The nationwide flood forecasting and warning system, known as 'NaFFWS', has been implemented in Malaysia through the PRAB program. The establishment was created with the purpose of facilitating the advancement of mitigation technologies aimed at minimizing the potential consequences of forthcoming flood events. The current utilization of modelling tools incorporates multiple factors that contribute to uncertainty, which can be attributed to the specific characteristics of the system. This review paper aims to explore the potential capabilities of an integrated multi-agent system specifically designed for the purpose of monitoring flood events. The proposed system is composed of logical agents and utilizes deep reinforcement learning (MADRL) techniques. This approach introduces a conceptual framework wherein a collection of intelligent agents collaborates to accomplish diverse tasks and effectively exchange information, ultimately facilitating the generation of timely alerts in the context of flood crises. The agents in question operate in collaboration with a hybrid approach that combines the DQN and TD3 algorithms. This combination is utilized to mitigate the various challenges arising from uncertainty. The proposed model's contribution is notable in enhancing flood forecasting accuracy amidst diverse sources of uncertainty.