Font Size: a A A

Intelligent Optimal Control Strategies For Microgrid Based On Distributed Generation

Posted on:2014-07-27Degree:DoctorType:Dissertation
Country:ChinaCandidate:F D LiFull Text:PDF
GTID:1262330401979022Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
The operation control for distributed generations (DGs) has become an important issue in smart grid. Because of variety of the configurations of DGs, diversity of operation properties, and complex fluctuation in realtime operation, the operating condition of microgrid including distributed generation units, loads, and energy storage elements seems more complex and dynamic. Hence, it has been one of the key technologies to develop feasible control methods for variant operating condition of microgrid, with DGs properties analyzed in depth.This thesis studies the optimal control strategies for the microgrid control. To meet the requirements of practice, a wind power intelligent prediction method based on slope events forecasting and wind speed forecasting is proposed. For the islanding mode of microgrid, an improved control strategy of load distribution is established, to deal with changes about distributed power source and loads. For the grid-connected mode of microgrid, a multi-agent system has been established. Finally, an improved reinforcement learning method is proposed to minimize electricity costs on the premise of satisfying the power balance and generation limit of units in a microgrid. The main study achievements include:(1) A wind power intelligent prediction method based on slope events forecasting and wind speed forecastingDue to large prediction errors existed in medium-term forecast, it is difficult to realize long-term management. Thus, this thesis defines the wind power slope events, and investigates its change characteristics. Then the classification of slope down/up events for both one-step and multi-step ahead is developed by using MSVMs. At the same time, an adaptive wavelet neural network is used to predict wind speed up to30h ahead. Based on slope events forecasting and wind speed forecasting, an improved radial basis function neural network (RBFNN) is proposed to predict wind power up to30h ahead. A novel adaptive learning method has been developed for on-line learning of the applied RBFNN. The proposed model is tested using wind power data collected from a real wind farm. The effectiveness of the proposed method is compared with FFNN, persistence (PER) and new-reference (NR) benchmark models. The results show that both the prediction time horizons and the prediction accuracy are guaranteed, and the proposed method can be applied to the optimal scheduling of wind farms for one day in advance.(2) An improved control strategy of load distribution in an autonomous microgridFor autonomous microgrid with multiple distributed generation units, a load sharing method is investigated. To overcome the shortages of traditional V/f droop control method, a new control based on V/δ droop control is presented, in which, the P-δ droop function and Q-V droop function employs additional derivative terms with modified gain respectively, so that the stability and good transient response can be guaranteed. A supplementary loop is proposed around the primary droop control loop of each distribution generator (DG) converter, which can modulate the d-axis voltage reference of each converter. Comparative experiments are carried out in the autonomous microgrid under significantly change state and steady state, and both the higher and lower droop gains are employed. The simulation results verify that, the proposed control strategy is effective for stabilizing the system and resulting in optimal load distribution in the autonomous microgrid.(3) An multi-agent system for microgrid under grid-connected mode and real-time electricity price designedA hierarchical control architecture including local controllers, microgrid manager and distribution management system is proposed. Then, the microgrid control requirements, such as distributed control, real-time pricing and operational cost, are analyzed, in order to set up objective function for optimization. Then a multi-agent system to describe microgrid under grid-connected mode is built, with state variables, action variables and reward function are formulated. A state variable "Trend" is defined to express the most possible transitions of the system, in order to reduce the complexity of states. Such multi-agent system of microgrid needs less communication between different units, and reflects the actual situation. It serves as the foundation of optimal control for the microgrid under real-time electricity price.(4) DHRL-RCSV algorithm for the optimization in microgrid under grid-connected modeRL is perplexed by the problem of’curse of dimensionality’in the large-scale microgrid environment, which results in lower efficiency. Therefore, a novel dynamic hierarchy based on change rate of key state variable is proposed, with which a hierarchical reinforcement learning, named DHRL-RCSV is developed. As the electricity price is taken as the key state, a similar MAXQ hierarchy can be constructed automatically by identifying bottleneck states. Then an improved algorithm named Bayesian-MAXQ learning is applied to explore the recursively optimal policy for microgrid management. The simulation results show that DHRL-RCSV performs with good efficiency to generate manage strategy, which is able to optimize loads, DGs, and energy storage elements, according to the changes of electricity price in power markets.
Keywords/Search Tags:Distributed generation(DG), Microgrid, Slope events, Support vector machines, Neural networks, Intelligent prediction, V/δ control, Multi-agent system, Dynamic hierarchical reinforcementlearning
PDF Full Text Request
Related items