Stochastic Optimization Model And Method For Energy Management Of Stand-alone PV-Battery-Diesel Systems

Microgrid is an important component and technology of Smart Grid. The operation optimization is an important research focus of microgrid. With the energy management system, efficient utilization of renewable energy and economy of microgrid can be realized. According to the operation information (e.g. output of distributed generators, load demand, power market), optimization objectives and constraints, energy control strategy of microgrid is made by the energy management system.In this thesis, we present an optimal control model for stand-alone PV-diesel-battery system. To minimize the long-term average operating cost, the control strategy for the battery and diesel generators are optimized. The stochastic models of the photovoltaic and load demand are established and energy control of the system is modeled as a semi-Markov decision process (SMDP). The system state is formed by the levels of photovoltaic power and load demand, operating numbers of diesel generators and levels of battery state of charge (SOC). We use the stochastic dynamic programming method to solve the problem and get the optimal action in every state. The simulation results show that the optimal energy management strategy can decrease the operation cost of the system and demonstrate the rationality of the model and effectiveness of the optimization method. The initial investments, optimal operation costs and payback periods of different system configurations are illustrated.In addition, SOC of battery in microgrid system can’t be directly measured. Since the charging and discharging efficiency are difficult to obtain and there is self-discharge phenomenon in the battery, it will produce a big error to calculate the SOC with Ah counting method. And the SOC can’t be obtained accurately in real-time. Considering that there is a relationship between SOC and voltage of the battery, and the voltage is easy to measure, we make battery voltage as an observation. And observation states are composed by the power of photovoltaic, load demand, operating numbers of diesel generators and voltage of battery. Partially observable Markov decision process model of the system is established. The problem is solved by theory and learning algorithms and the optimal control strategy under belief states of the system is got. Finally, simulation results and analysis are given.