This paper proposed a continuous algorithm for distribution network reconfiguration (DNR) based on efficient encoding of solution space. By estimating the exchange range of switch exchange through deduced approximate formula of network loss, the solution space was compressed and the efficiency of optimization was improved. To avoid directly solving discrete variables, a double-population solution space including discrete variables and continuous variables solutions was established, which can reduce the amount of calculation and is suitable to apply continuous variable optimization method to solve DNR problem. After reducing the radius of the confidence interval variables, two kinds of non-effective solutions were avoided, and the number of times of power flow calculation was also decreased. The improved crisscross optimization algorithm improves the convergence speed and precision. Simulation results verify the correctness and effectiveness of the proposed algorithm.