Prediction of the microclimate through NAR and NARX neural networks: application to Rosenborg Castle, museum partner of the CollectionCare project

The microclimate influences the chemo-physical and structural properties of cultural objects preserved in the museum being a determining factor for art collections' safety and, at the same time, for the comfort of visitors and staff. For this reason, over the years, the monitoring of the internal climate has become a common practise to study the main agents of deterioration for artistic artefacts In fact, it is possible to act in a preventive manner, , having available the forecasts of the dangerous agent trends. In recent decades, knowing in advance has become synonymous of Artificial Intelligence (AI). Machine learning (ML) has in fact shown surprising results in the field of prediction, classification and decision-making applied to very different contexts, from the economic to the medical one. However, one of the limitations of Machine Learning lies in the availability of very large datasets, characterized by long-time data series and by many independent variables. If this is not possible, the trained neural network may "get used" to the data and be unable to predict any changes or trends far from the models. The application of two specific neural models for time series, Nonlinear Autoregressive Neural Networks (NARNNs) and Nonlinear Autoregressive with Exogenous Inputs Neural Networks (NARXNNs), has proved to be extremely powerful for training on not-extensive datasets. These two models can be applied to any data context: the simpler NARNNs networks can be trained on a single time series. If several variables are available, the more complex NARXNNs networks can be used, which take as input the variable to be predicted as output and other historical series called exogenous. These two models have been used on temporal datasets recorded at the Rosenborg Castle (Copenhagen-Denmark), museum partner of the European CollectionCare project. The results of these preliminary applications are very promising and Future developments and applications of ML models for microclimate predictions could lead to important benefits in terms of safeguarding the cultural heritage, people's well-being and energy saving. Indeed, the networks trained on this data have shown that they are able to predict future trends very precisely. The knowledge of future microclimatic trends can help to act in advance, controlling the environmental characteristics through the management of heating and cooling systems, to keep the artistic artefacts in a safe microclimatic habitat.