Time series prediction can be applied to a lot of fields such as financial analysis etc.Recently
there are growing interests in neural network based time series predictors.However
neural network based time series predictor often gives invalid predictions.In this paper
the probability of invalid prediction given by neural network based is analyzed firstly.And then a new Regularized Learning algorithm is proposed for Neural Network based time series Predictors (RLNNP).After sufficient training
neural network based time series predictor with RLNNP algorithm will give a valid prediction for time series.