Studies On Regulation And Storage Allocation Strategies Of Wind Farm Self-discipline

As one of the most potential renewable energy, Wind power generation is developing rapidly all around the world. In order to achieve the rapid development of wind power, on one hand, an optimized generation schedule should be made to deal with output uncertainty of wind power generation and promote the ability of integrating wind power. On the other hand, an efficient control strategy should be explored to enhance self-discipline of wind farms, while keeping profit and reducing the burden of power grid. In this paper, the uncertainty of wind power is analyzed. Reinforcement learning method is introduced into scheduling control problem of stored energy wind farms, and a wind power-energy storage system scheduling control model based on Q-learning is constructed, which can realize wind farms self-discipline. A method is given to assess the suitability of energy storage capacity by monitoring SOC curves. The main research content is as follows.The distribution model of wind power forecast error is built by the non-parametric estimation method. From the perspective of system backup, the influences of wind power uncertainty on power system are studied. A method for determining wind power reserve capacity is given based on reliability indices. The effects of wind farms self-discipline and control strategies are present by the wind power reserve capacity. A scheduling control architecture of stored energy wind farm is introduced, which with the goal of maximizing the wind farm long term operating income and controlled by energy storage action threshold and wind power schedule output adjust value. The rationality and necessity of control quantity are proved by analysis of examples.From the angle of view of time series, wind power forecast errors show that the nonstationarity of error sequences may increase the demand of storage capacity, so it is necessary to control wind power-energy storage systems in real time.Q-learning algorithm can get decision-making ability from the interaction through the controller and actual operation environments, without the need for prior knowledge. For this purpose, control of wind power-energy storage system can be regarded as multi-stage decision problem, and Q-learning algorithm can be used to construct the scheduling control model. Simultaneously, an action selection strategy is given, which may balance optimality and exploratory. Examples prove that wind power-energy storage system scheduling control model based on Q-learning can realize wind farms self-discipline, ensure the rationality of charge and discharge and improve the operating income of wind farms.Considering the particularity of energy storage when participating in power system dispatching, an energy storage capacity utilization coefficient is defined to reflect the sufficiency of capacity. Adjusting charge and discharge sequence can hold the external characteristics of energy storage, after the compression of capacity, which proves that optimizing control strategy can improve energy storage’s utilization efficiency and reduce storage capacity requirements. Based on the differences of complementary degree between wind power reserve capacities and reserve prices, the influences of capacity constraints on the satisfaction level of self-discipline and the conservative degree of Q-learning controller are discussed. Cases prove that energy storage capacity allocation conforms to the law of diminishing marginal utility, and a method is given to find optimal energy storage capacities and capacity utilization coefficients, according to energy storage income increment curves and operating costs curves. Under the premise of the strong correlation between income incremental curves and energy storage capacity utilization coefficient curves, Online assessments of allocation suitability can be realized by monitoring SOC curves, which provide reference for the adjustment of energy storage capacities.