Modular multilevel converter (MMC) is a possible solution for integrating an EV fleet into the smart grid. Considering the differences of initial state of charge (SOC) and the inherent randomness in charge/discharge demands from the various EVs connected to the MMC, it becomes challenging to control the MMC in order to satisfy the network and EV fleet requirements. This paper proposes a multi-objective power management strategy for a MMC-Based EV fleet to be integrated into the smart grid using a virtual SOC model and coordinated control of the power flow among the arms of the same and different phase-units by taking advantage of circulation current. The model of the MMC-based EV fleet system is developed in Matlab/Simulink and simulation results show that the multi-objectives of differential charging/discharging power control and interactive power management between EVs and utility can be effectively realized, while the three-phase output power of MMC is kept in balance and circulation current is controlled, thus, mitigating the negative impact of large-scale application of EVs on the utility grid.

Fast and accurate forecasting of schedulable capacity of electric vehicles (EVs) plays an important role in enabling the integration of EVs into future smart grids as distributed energy storage systems. Traditional methods are insufficient to deal with large-scale actual schedulable capacity data. This paper proposes forecasting models for schedulable capacity of EVs through the parallel gradient boosting decision tree algorithm and big data analysis for multi-time scales. The time scale of these data analysis comprises the real time of one minute, ultra-short-term of one hour and one-day-ahead scale of 24 hours. The predicted results for different time scales can be used for various ancillary services. The proposed algorithm is validated using operation data of 521 EVs in the field. The results show that compared with other machine learning methods such as the parallel random forest algorithm and parallel k-nearest neighbor algorithm, the proposed algorithm requires less training time with better forecasting accuracy and analytical processing ability in big data environment.

This paper presents a single-phase differential buck-boost inverter that possesses both buck-boost and active power decoupling functions without adding any power electronics or increasing the complexity of driver circuits as compared with previous differential inverters. Two types of operating principles of the differential buck-boost inverter are introduced: the unipolar operation with new pulse energy modulation (PEM) technique enabling the inverter to operate under both discontinuous conduction mode (DCM) and continuous conduction mode (CCM); the bipolar operation with energy-based active power decoupling delivering the second-order ripple power into the output film capacitors, thus eliminating the large electrolytic capacitor at the DC side. Small-signal modeling analysis is conducted to show the characteristics of the proposed system and control. Simulation and experimental results verified the feasibility of both operating principles for the differential buck-boost inverter and successful power decoupling with substantial reduction of second-order component in the DC current.

Accurate forecasting of short-term wind speed is a key technology to enable efficient and reliable operation of microgrids with wind generators. Results of wind speed prediction (WSP) methods in the current literature are subject to errors due to the random nature of wind speed and the limited generalization of forecast algorithms. In this paper, a short-term WSP method based on a model of forecast, error correction, wind power generation, support vector machines (SVMs), and prediction algorithms is proposed. In the proposed method, the error model is built to predict the errors of the original predicted wind speed. Subsequently, the predicted errors are incorporated into the original predicted sequence to produce the final forecasts. The final predicted results can be brought to the system operator step by step for use in scheduling strategies. Using SVM and back propagation (BP) prediction algorithm as examples for the basic prediction algorithm, the proposed method is applied to a day-ahead WSP with five daily deliveries and compared with the results by a single SVM or a BP prediction algorithm. The test results demonstrate significant improvement in prediction accuracy using the proposed short-term WSP based on a model of forecast error correction method, independent of the basic prediction algorithm.

In this paper, a power-frequency (P-ω) controller is presented for voltage source converters (VSC). The approach is intended for multiple parallel VSCs forming a microgrid operating in both grid-connected and islanded modes. The proposed controller allows a VSC to mimic the operation of a synchronous generator (SG) by implementing the swing equation of SG with a primary frequency controller. In addition, a generalized model of the active power generation dynamics is developed in order to analyze the stability and to design the main control parameters. In contrast with the conventional droop control method, the proposed controller improves the close-loop system dynamic response without changing the frequency accuracy. The obtained results show the good performance of the proposed controller.

The application of large-scale electric vehicles (EVs) into the future smart grid may bring about serious power quality problems. But EVs can provide ancillary services for the power system as distributed energy resources through Vehicle-to-grid (V2G) technology. The fast and accurate prediction of schedulable capacity (SC) of EVs is the first step to enable this benefit. In this paper, two different time scales schedulable capacity forecasting (SCF) methods are proposed, including the real-time SCF method and a day head SCF method. Firstly, the real time of SCF of EVs is calculated by collecting the real-time charging or discharging data of individual EVs and a distributed data parallel calculation method to solve problems of the mass data process and storage problem. Secondly, the large amount of results of the real-time SCF, is used as historical data, to train the SCF models one day ahead by parallel decision tree regression algorithm (PDTR) and parallel random forest regression algorithm (PRFR). These proposed methods are evaluated by the real time operation data involving 521 EVs in the filed in China on Spark. The test results show the advantage of using big data technology in the ability and speed of dealing with large-scale data, and PDTR is faster and better than PRFR in one day ahead SCF.