Summary: |
There has been a significant increase in the usage of mathematical models across almost all fields of science over the last decades, certainly related to the rapid progresses in hardware and software development . Models are now
intensively used in theoretic and applied Ecology. Recently, a new field of research has been emerging from the linkage of Artificial Intelligence (AI) techniques like Machine Learning (ML) and Autonomous Agents with
Environmental Sciences, as reflected in several international workshops.
After a model is implemented, the first round of simulations is generally designed to test it s internal logic against common sense knowledge. Once this
task is accomplished, it is necessary to calibrate the model, i.e., to perform a second round of simulations and to tune model parameters in order to reproduce observed data. In every mathematical model, parameters regulate the behavior of equations describing temporal and spatial changes of model state variables and their interactions. Generally, there is some uncertainty associated with each parameter. This "tuning" procedure is a hard and
"tedious" work requiring a good understanding of the effects of different parameters over the available variables. A third round of simulations is generally performed to validate the model, i.e., to compare predicted with observed results that were not used in the calibration process. Once the model is validated it may be used as a predictive tool and simulations may be set up depending on the purposes for which it was developed. When the goal is to find some optimal solutions, a large number of trial and error simulations may be required.
Some of the learning processes associated with model implementation may be automated to save man power and increase the performance of some tasks, such as model calibration and search for optimum solutions. Automatic calibration procedures already exist , based on systematic and exhaustive generation o ![Ver mais. Adequado para parcelas de texto incompletas e que, através deste ícone, permite-se que o utilizador leia o texto todo.](/fep/pt/imagens/VerMais) |
Summary
There has been a significant increase in the usage of mathematical models across almost all fields of science over the last decades, certainly related to the rapid progresses in hardware and software development . Models are now
intensively used in theoretic and applied Ecology. Recently, a new field of research has been emerging from the linkage of Artificial Intelligence (AI) techniques like Machine Learning (ML) and Autonomous Agents with
Environmental Sciences, as reflected in several international workshops.
After a model is implemented, the first round of simulations is generally designed to test it s internal logic against common sense knowledge. Once this
task is accomplished, it is necessary to calibrate the model, i.e., to perform a second round of simulations and to tune model parameters in order to reproduce observed data. In every mathematical model, parameters regulate the behavior of equations describing temporal and spatial changes of model state variables and their interactions. Generally, there is some uncertainty associated with each parameter. This "tuning" procedure is a hard and
"tedious" work requiring a good understanding of the effects of different parameters over the available variables. A third round of simulations is generally performed to validate the model, i.e., to compare predicted with observed results that were not used in the calibration process. Once the model is validated it may be used as a predictive tool and simulations may be set up depending on the purposes for which it was developed. When the goal is to find some optimal solutions, a large number of trial and error simulations may be required.
Some of the learning processes associated with model implementation may be automated to save man power and increase the performance of some tasks, such as model calibration and search for optimum solutions. Automatic calibration procedures already exist , based on systematic and exhaustive generation of parameter vectors and using several convergence methods. However, they require a large number of model runs and are, therefore, not applicable to complex ecosystem models demanding large computational times. A possible alternative may be to develop a selflearning tool that simulates the learning process of the modeler about the simulated system. A similar approach
may be used to find optimal solutions to environmental problems. In both cases, Autonomous Agents may be a sound way to implement the self- learning tools. Autonomous Agents also enable to introduce in the simulations, in a
very natural way, the human element , whose reasoning process is very difficult to model by traditional simulation processes.
This project aims at developing a complete multi-agent simulation system, including an ecological simulator, a Calibration Agent based on ML techniques capable of calibrating complex ecological models and autonomous agents representing the intelligent entities present in the simulation. The system will be applied to ecological models of coastal ecosystems and used for aquaculture optimization. |