Abstract: Under the fluctuating power spot price, micro-energy-grid often faces the uncertainty of dispatch costs and additional costs. However, the commonly-used stochastic optimization methods, namely, typical scenario optimization and conditional value at risk (CVaR), have the problem of ignoring the loss of scenario reduction and the subjective selection of confidence value. To this end, an improved comprehensive scenario-reduction-optimization method in which both the scenario similarity and the combined loss are taken into account is proposed and the loss degree after scenaio reduction is used as the basis for the selection of confidence value, forming an improved CVaR day-ahead economic dispatch model. The cases study shows that the method based on scenario-reduction-optimization guiding confidence value selection can effectively reflect the actual dispatch loss cost by CVaR, and is closer to the theoretical minimum tail risk loss than the subjective value, thus the day-ahead dispatch cost reaches a more optimal level in the actual market environment. Furthermore, the applicability of improved CVaR model is discussed when further considering the uncertainty of wind power and different schemes of scenario reduction.