|关键词||智能电网 神经网络 自适应动态规划 电能管理|
神经网络是一种基于计算的智能控制算法，它模拟人脑的信息处理机制，将系统的特性赋值于神经网络的连接权值之中, 然后利用大量数据进行自我学习，在复杂系统控制、机器学习、模式识别等领域有广泛的应用。自适应动态规划(Adaptive dynamic programming)是一种集成了动态规划、强化学习和神经网络三者优势的智能控制方法，解决了传统动态规划的“维数灾”问题。自适应动态规划基于系统的大量已知数据进行计算实验，在求解复杂非线性系统的优化控制问题方面有很好的应用效果。因此，本文将研究自适应动态规划方法在智能微网中的电能优化协调控制问题。
With the rapid growth of the world’s population and the continuous development of the society in the past decades, the demand for energy has continuously increased, and the global energy shortage problem is becoming increasingly serious. Energy saving, emission reduction and improving the utilization of energy have attracted extensive attention from the worldwide scientists. In the electricity filed, in order to achieve the goal of improving the utilization of electricity and reducing the cost of electricity, countries have begun investing significant resources in the research of smart grids. With the development of power-related technologies, real-time pricing mechanisms are gradually being adopted to guide users in optimizing power usage. The reduction of the energy storage equipment storage could enable the ordinary users to use energy storage equipment to achieve the storage and release of electrical energy. In addition, advances in renewable energy power generation technologies such as solar power generation have reduced the amount of pollutants emitted by conventional power generation methods and protected the environment. At the same time, renewable energy generation has also reduced the economic cost of electricity for users.
Artificial neural network is a computation-based intelligent control algorithm. It simulates the information processing mechanism of human brain, assigns the characteristics of the system to the connection weights of the artificial neural network, and then uses a large amount of data for self-learning, which has been widely used in different areas such as complex system control, machine learning, and pattern recognition. Adaptive dynamic programming (ADP) is an intelligent control method that integrates the advantages of dynamic programming, reinforcement learning, and artificial neural network, and solves the "dimensional disaster" problem of traditional dynamic programming. ADP performs computational experiments based on a large amount of known data from the system, and has a good application effect in solving optimization control problems for complex nonlinear systems. Therefore, this paper will explore the application of ADP in the problem of adaptive power planning and coordination in the intelligent microgrid.
The main work and contributions of this article are presented in the following areas:
1. Aiming at the problem of power coordination between two smart residential users in an intelligent micro-grid system, an adaptive energy management algorithm based on ADP is proposed. In the case of real-time electricity prices, when all the single intelligent residential users have energy storage devices, the power transmission scheme between any two smart residential users (and external public power grids) is very complicated and difficult to solve. To solve this problem, an ADP-based energy optimization transmission management algorithm between two smart homes based on was proposed for the first time. Second, this work also proposed that the efficiency of energy storage equipment could be increased by the usage of penalty costs, which could control the cost of the usage of store equipment. Finally, re-planning the power of the external public power grid using energy storage equipment, which could improve the balance of the total user load, reduce the peak-valley load difference of the external power grid, and save the user's economic costs at the same time.
2. For the real-time electricity price situation, there is a problem of optimal coordination and control between two residential users (and external public grids) of the smart microgrid who have energy storage equipment. The incorporation of solar power makes the optimization of the energy between the two users more complicated, especially considering the volatility of solar power generation. A ADP-based coordinated power control optimization method was designed to prioritize the power generated by solar power generation. In the penalty function, a variable weight function was proposed to regulate the influence of the weather on solar power generation. Secondly, the outside air temperature was also used as an input variable of the neural network to increase the influence of solar power generation as much as possible. The results of the experiments show that this method effectively saves the user's economic costs and improves the utilization rate of the electricity energy.
3. Taking into account the influence of solar power generation on the optimal control of power coordination between two residential users (and outside public power grids), a new power optimization coordination and control algorithm based on weather classification prediction was proposed. Based on the mature weather forecasting system, the weather is divided into three types (sunny, partially cloudy, and cloudy). Each day is divided into a type of weather, and the corresponding network is trained according to the type of weather, which could minimize the influence of weather factor on solar power generation and the coordinated optimization of energy control for the entire residential users. Then the accuracy of the corresponding network and the effectiveness and pertinence of the algorithm could be improved. At the same time, the corresponding power transmission priority scheme has been designed to maximize the utilization of solar energy and reduce the energy loss during power conversion. According to the control problem of energy storage equipment, the corresponding energy storage equipment charging and discharging control strategy adjustment plan was proposed. The results of the experiment show that this method could reduce the user's economic costs and improve the utilization of solar energy effectively.
|徐延才. 基于自适应动态规划方法的智能微网系统电能优化协调控制[D]. 北京. 中国科学院大学,2018.|
|Yancai Xu-Thesis.pdf（2080KB）||学位论文||暂不开放||CC BY-NC-SA||请求全文|