Aiming at the heavy computational burden in reliability assessment of microgrid（MG）

a selective data-driven algorithm（SDDA） is proposed. Firstly

a series of continuous random variables with proper probabilistic performances are defined to model the sequential multi-state uncertainties of components and load of MG. Secondly

the reliability index is expressed as the multidimensional integration in terms of random variables

which is compatible with sequential Monte Carlo（SMC） simulation. In addition

the analytical expressions of optimal probability density functions（PDFs） of random variables are worked out based on the Lagrange multiplier method

and the numerical solutions are estimated by Monte Carlo integration. At the same time

a density variance index is defined to evaluate the contribution of each random variable to the computation cost. Lastly

the SDDA is proposed to overcome the prohibitive computational burden by assigning selective data-driven procedure to determine the dominant variables and optimal PDFs

and efficient simulation procedure to decrease computation burden. An evaluation program based on SDDA is coded on the platform of MATLAB

and comprehensive case studies are implemented. Results show that dominant random variables will change along with system conditions and be identified by SDDA

which will propose an effective data-driven strategy accordingly. Then the computation cost of reliability assessment is greatly reduced.