👈🏼 Back to list

Elisa Castro

Università di Catania
Artificial Neural Networks
wave climate
maritime accidents
PHD school
Evaluation and Mitigation of Urban and Land Risks
PhD Cycle
37
List of Supervisors
Luca Cavallaro
Main research approches
Numerical analysis
Research abstract
Sea state prediction in coastal and port areas through Artificial Neural Network
Background And Research Gaps
Knowing local weather and sea conditions well in advance is undoubtedly helpful in mitigating the risk of accidents in port areas. Complex numerical models are used to define the wave climate near port areas. However, due to their high computation cost, such models may be useless for forecasting and nowcasting applications. To overcome this limitation, Artificial Intelligence algorithms, such as ANNs, can be used. Indeed, the scenarios simulated with ANNs can be trained with the data obtained from the models mentioned above and can provide the required metocean information almost instantaneously. Although the applications of ANN are becoming more widespread, most of them are limited to single-site forecasting and do not consider the spatial correlation with the other surrounding points.
Research Goals
The research project aims to reduce the risks associated with maritime accidents and their consequences on human life and the environment, creating a coordinated system of innovative methods. In particular, the main goal is the development of dynamic forecasting of the meteomarine climate in harbour areas through the implementation of an ANN as an element of an integrated system for safe navigation in the port area of Augusta (Sicily), which represents one of the most important Italian ports. In particular, this research aims to contribute to developing a new strategy to adopt ANNs for evaluating and predicting nearshore wave characteristics in actual conditions and over entire areas.
Methods
Numerical models are utilised to reconstruct nearshore wave climate characteristics, considering the diverse morphological conditions present in the Augusta area. However, the computational requirements of analysing extensive areas with varied physical attributes can be substantial. Therefore, the second step involves implementing clustering algorithms to identify areas with homogeneous conditions, thereby reducing computational demands and improving the efficiency of subsequent applications of ANNs for each cluster.
Results
The K-means algorithm was applied to each macro-area. Various evaluations of the distance metrics and the optimal number of clusters were conducted to find the algorithm's best configuration. The proposed methodology is promising since clustering can effectively capture wave height distribution in vast areas. However, the analysis reveals that the choice of distance metrics can significantly impact the performance of standard clustering algorithms such as K-means. ANNs were applied to each cluster to forecast the wave climate not only for specific points but for the entire cluster as a whole.