| Abstract: |
| To enhance the deployment capability and low-carbon degree of virtual power plants (VPPs), a novel optimized scheduling model is proposed in this paper for a multi-energy VPP. To explore the distribution potential of the VPP and bolster its multi-energy complementarity, an architecture integrated with electric vehicle (EV) charging stations is introduced, and a battery health degradation mechanism is constructed. To address the uncertainty exhibited by EV behaviors, a feature extraction method based on deep Q-network and maximum relevance-minimum redundancy (mRMR) is then proposed. This method optimizes the applicability of mRMR in large datasets, thereby improving the accuracy of charge behavior prediction. Next, to achieve a complex optimization dispatch, a twin delayed deep deterministic policy gradient algorithm is employed. The twin Q-value truncation mechanism and smooth regularization effectively suppress the issue of policy overestimation biases. Further-more, to validate the performance of the proposed model and algorithm, four different cases are designed, and the scheduling effects achieved for EVs are compared with those of the traditional battery energy storage system framework. The simulation results show that the proposed model significantly reduces both the operational cost and carbon emission level while slowing the battery health degradation process. |
| Key words: Energy distribution, twin deterministic policy gradient, deep Q-network, electric vehicles, virtual power plant. |
| DOI：10.23919/PCMP.2025.000017 |
|
| Fund:This work is supported by the National Nature Science Foundation of China (No. 72401182). |
|
