Initialization of multilayer forecasting artifical neural networks
This work addresses initialization challenges in neural networks for time series prediction, but it appears incremental as it builds on existing linear filter methods.
The authors tackled the problem of initializing multilayer neural networks for time series forecasting by developing a method that sets initial weights based on linear prediction filters and uses matrix decomposition variants to improve accuracy, demonstrating its efficiency on the Lorentz chaotic system.
In this paper, a new method was developed for initialising artificial neural networks predicting dynamics of time series. Initial weighting coefficients were determined for neurons analogously to the case of a linear prediction filter. Moreover, to improve the accuracy of the initialization method for a multilayer neural network, some variants of decomposition of the transformation matrix corresponding to the linear prediction filter were suggested. The efficiency of the proposed neural network prediction method by forecasting solutions of the Lorentz chaotic system is shown in this paper.