Abstract: In order to solve the problem of the high risks and low efficiency caused by the inconsistency of the day-ahead and real-time electricity market prices in the electricity spot market, virtual bidding (VB) was used to arbitrage on difference between such two market prices that are unknown to virtual bidders to promote the fair competition. The virtual bidding model was established from the dimensions of time and space. In order to maximize the cumulative payoff of virtual bidders, the proposed model took the budget constraints of virtual bidders into account, as well as conducted two types of virtual bidding of decrement and increment bids for multiple locations on a period of time. And the problem was formulated as a 0-1 knapsack problem. Meanwhile, the conditional value-at-risk was used to quantify the risks faced by virtual bidders by risk pursuing and aversion. A virtual bidding model under risk measurement was also established. In order to solve the problem, through the design of state space, action space and reward function, a deep reinforcement learning network framework was built. Meanwhile, the deep Q network was used to interact with the environment to obtain feedback and the parameters of the neural network was optimized to achieve an effective solution to the optimal bidding strategy. The PJM data from June to December in 2018 was used to calculate the cumulative profits and Sharpe ratio of virtual bidders. Compared with greedy algorithm and dynamic programming, the the effectiveness and superiority of deep reinforcement learning algorithm is verified in this paper.