HTML5 Webbook

30/40

theme f. Investigation of reproducibility of spatio-temporal evolution model obtained by AI learning (Junichiro Tahara, Kengo Nakai) Research is emerging that learns time-series data of phenomena such as the actual atmosphere and ocean and uses it for future prediction. .However, simply being able to predict time series is not enough to construct a model of a phenomenon.Therefore, we clarify how well the time evolution model obtained by learning is reproduced from a mathematical point of view.In particular, we focus on the reproducibility of the dynamical system structure in which the hyperbolicity is broken. (Planning/Method) High-dimensional Lorentz model is one of the dynamical structures that we want to see.In addition, it has been found that machine learning called reservoir computing is effective for time-series forecasting of deterministic dynamics.Therefore, first, by using this learning method, we obtain a machine learning model that can predict time-series data with high accuracy from the time-series data of the high-dimensional Lorenz model. (Results and future prospects) In high-dimensional Lorentz systems, it is known that the dynamical system structure changes greatly depending on the parameters, and there are parameters where the local instability is one-dimensional and two-dimensional. ing.We succeeded in constructing a machine learning model that learns the time series of the high-dimensional Lorentzian system with this parameter and predicts its time evolution.Figure 1 below shows the prediction results of the constructed machine learning model.The horizontal axis is time, and the prediction was started from time 2, and the movement of a variable with the predicted high-dimensional Lorentz was written out as a solid red line.For comparison, the correct time-series data obtained by directly calculating the high-dimensional Lorenz equation is drawn with a blue dotted line.Although the prediction error spreads due to chaos, it can be seen that the prediction is very good until about time 0.The results of long-term calculations are plotted in Figure 6 below.It can be seen that the model can be calculated without failure up to a certain amount of time.It is a future task to clarify the reproducibility of the model constructed in this research, focusing on the dynamical structure such as the dimensional switch.If this reproducibility is clarified, it is expected that the effectiveness of machine learning modeling on real data will become clearer.Fig. XNUMX Prediction result by configured machine learning model Fig. XNUMX Result of long time calculation XNUMX.Acquisition of marine big data and promotion of AI analysis research (XNUMX) Development of a fishery integrated support system that realizes resilient and sustainable fisheries using marine organism big data and training of marine AI human resources