Resumen:
Background: Colombia is the sixth country with the greatest water supply in the world, but the Ministry of Environment estimates that approximately 50% of water resources have quality problems. Often to control proper water quality conditions, it is not sufficient to establish monitoring activities in addition to this there is a need models and mechanisms to anticipate the risk materialized with enough time range to prevent negative effects disturbing the quality of water resources. In this sense, predicting water quality plays a very important role for many socio-economic sectors that depend on the use of this liquid.
Goals: Propose a mechanism for water quality prediction through an adaptive approach that supports decision-making processes on different uses of water resources.
Methods: An adaptive prediction mechanism of water quality using Computational Intelligence techniques is proposed; the main focus of this mechanism is the ability to be applied to datasets from different water uses without the prediction accuracy is affected in a drastic way. This mechanism consists of a parameter calibration component, a predictive component, responsible for performing prediction operations using CI techniques and finally an adaptive component, which implements the CI algorithm to adjust the predicted values to actual values in the water use selected.
Results: Three water quality datasets (aquaculture, human consumption and recreational use), an adaptive prediction mechanism of water quality based on Computational Intelligence techniques.
Conclusions: Support Vector Regression configured with PUK kernel presented a better performance in the accuracy of predictions compared to other techniques of Computational Intelligence. In order that the predicted values by the mechanism from approaching the actual values in different uses of water, it was necessary to use the Particle Swarm Optimization (PSO) technique.