In this paper, a novel algorithm named SLS-Elman is put forward, which aims at more effectively training and learning small scale samples with many characteristic variables, and takes Sectional Least Squares principle and some structural property of Elman neural network into consideration. When doing characteristic variable reduction on high-dimension of small scale samples, the novelty algorithm takes relativity among dependent variables into account. Obtained data by this algorithm carry on training and simulating on a neural network; outputted network is more simplified in view of structure, and presents a more precise network model. The statistics of case analysis demonstrates novelty algorithm improves convergence rate and forecast precision, and efficiency. In the mean time, on purpose to test efficiency of novelty algorithm, it is compared with some algorithms such as Elman neural network algorithm based on Principal Component Analysis and so on, as a result, it presents more advantages.
