Two improvement methods were presented to avoid ineffective circulation and invalid waiting state problems when running algorithm Randomized Switching for Maximizing Lifetime (RaSMaLai) and then a new random switching algorithm New Randomized Switching for Maximizing Lifetime (NRaSMaLai) was put forward: the first improvement was to conduct the initialized inspection in the process of traversing tree nodes in order to prevent the tree from entering into invalid waiting state; the second improvement was to do the state inspection for the maximum load node and all its descendant nodes in the operation process of updating the tree to avoid ineffective circulation. The tree balance was achieved by increasing the load of the minimum node and its descendant nodes with NRaSMaLai. The simulation experiment shows that these two methods can make the tree achieve the balance state or at least get closer to the presupposed state. When the sink node was located in the regional center, the iterative steps which make the tree balanced can be reduced to 1/5 of the original by NRaSMaLai and also it appears little oscillation. This is significant for the data collection tree's rapid convergence and the extension of the network's lifetime.