This site uses cookies cookie policy, here. By continuing your visit to our website, you are accepting our policy.

Puertos de Estado

Improve the quality of your predictions thanks to Machine Learning

Sales Contact | (+34)976 512 433

The Client

Puertos del Estado has developed and maintains measurement&forecasting systems of the marine environment with the fundamental objective of providing the essential meteorological data for the exploitation of the port system.

The system consists of measurement networks (buoys, tide gauges and high frequency radars), prediction services (waves, sea level, currents and water temperature) and climate sets, which describe both the maritime climate at present and its scenarios of change in the 21st century.

The quality of this data has made possible to increase the efficiency, sustainability and security of port operations over the years and its improvement may allow us to continue developing more secure port operations with cost reduction.

The Challenge

Predict the oceanographic variables through historical data.

When one of the measuring instruments breaks down, it leaves the area without reliable reference data, and the entire prediction model is affected.

The challenge faced by the Nologin researchers was to be able to estimate the behavior of the different oceanographic variables, starting with a huge amount of historical data, adjacent to the instrument, to estimate the measurements of the damaged measuring instruments..

Using data from wave, wind and currents numerical models as inputs, it was accomplished to fill in the information gaps (buoys measures gaps) and therefore, an improvement in the accuracy of wave model was achieved. This improvement has direct implications for the logistic activities: it allows to optimize the routes of the vessels, making possible to save fuel and time, as well as to support the port staff's decisions on the safety of the port.

coches autopista

Results of the predictive model and results of the neural network at the Tarifa buoy (variable Tm02). An improvement of 74% can be seen.

The Results

Filling historical gaps in a reliable way has been possible through the use of Machine Learning models (neural networks and convolutional neural networks), improving by up to 74% the baseline wave model.

As a result, State Ports has been able to improve the reliability of its forecasting wave model, which is vital for many port operations.

The neural networks yielded more accurate results than the wave numerical model in different locations of the Spanish coasts. Several buoys were studied and the wave model estimations were improved considerably, obtaining error reductions of up to 74%.

The obtained results allows us to glimpse that, in the future, the complete replacement of the physical model by a calculation based on neuronal networks can be considered.

Currently the wave numerical model requires a calculation time of about 1 hour to obtain the predictions for the Spanish area, while convolutional neural networks (CNN) once properly trained could have an estimated calculation time in the order of minutes. Assuming that the networks are capable of making forecasts similar to the model, this would involve saving time and energy in the calculation of the forecast.

More in-depth

Improve your infrastructure with Nologin