Over the past few decades, distributed renewable generation has experienced rapid growth in distribution networks. These generation units can not only encourage prosumers to participate in electricity markets, but also provide more flexible options for electricity usage of customers. However, the reliability service is uniform for urban customers. In order to maximize the reliability of the grid, customers can choose their preferred reliability service level based on the availability and price of distributed energy resources (DERs) when an outage occurs. Thus, a novel double-side auction mechanism that considers customer classification for half-hourly intervals is proposed in this paper, triggered by an early warning mechanism from the power grid. The price of DERs in this condition can be the same with the grid retail price, or even higher. Customers will be classified first by hierarchical clustering analysis to provide reliability service levels and purchase capability. Then the guidance price level and the trading amount for each cluster are indicated for prosumers. To obtain the maximum social welfare, the transaction sets for market clearing are refreshed until the winning bidders are satisfied. Simulations are conducted to verify the proposed model, using data collected from Smart Grid Smart City project in Australia.