TIME SERIES-DYNAMIC NEURAL NETWORK FORECAST ON DREDGER FILL SETTLEMENT

Reclamation through technology of dredger fill can relieve the problem of short earth resources effectively. So it is urgent to improve this technology. Dredger fill has a high content of clay, organic and moisture, high compressibility and low strength, which cause the characteristic of low consolidation efficiency and slow settling velocity for the reclaimed land. For most of the projects, long-term settlement observation has been omitted due to the big requirement of resources. It usually takes 2-3 years to form the hard mantle layer on the surface of dredger fill. Such duration is too long. The effect is not ideal. Furthermore, there is a big difference between the actual settlement after construction and the expected one. In order to satisfy the deformation requirements, the problems of the prediction of long-term settlement based on the observation data of short-term settlement have to be addressed. In addition, the method that can be taken based on the long-term settlement prediction needs to be solved. The time series-dynamic neural network is established through self programming in this article. This method is applied in the prediction of long-term settlement and the analysis of results in Dredger Fill. The results show that the method of dynamic neural network can be reasonably applied to the prediction of soft soil consolidation settlement with minor error and better feasibility. The prediction has high precision and stability.