Research On The Approach Of The Reactive Power And Voltage Control Of Distribution Network Based On Incremental Learning Theory

The distribution network is an important part of the power system as the most closely links with users. With the development of economy, science and technology, the automation degree of power grid related equipment and control is increasing. Even so, the management of our distribution network still has many problems. Such as power system reactive power planning, construction and management are still relatively weak, reactive power capacity is insufficient, sometimes power factor is lower and line loss rate is higher.Under the above background, the paper studies the current status of automatic reactive power and voltage control for distribution network and summarizes the characteristic of the global optimization based on the entire network and the local optimization based on partial information. And then, starting from the actual project needs and making support vector machine (SVM) learning methods apply to reactive power and voltage control. It uses the set of training samples to generate multi-classification model, which is used to control the online system. Therewith the control results are compared with ones of the reactive power and voltage control equipment based on nine-zone principle. Again according to the incremental learning ideal, the paper uses Q-learning algorithm to form online learning mechanism and construct an online incremental method for reactive power and voltage control of the distribution network. For that, establishing an effective reward function using for measuring the quality of each action is the key of control effect. Practice shows that the reactive power and voltage control based on Q-learning algorithm is more excellent than that based on SVM. In the end, the paper illustrates the engineering value of the method by regional power grid reactive power and voltage optimization control system applying to Shandong Binzhou and Dezhou.The automatic reactive power and voltage control based on incremental learning theory in this paper is able to avoid the existing problems in global optimization method, which is too dependent on data quality and poor convergence. In addition, it can coordinately control the compensation equipments between transformer substations, improving the control quality of local compensation based on partial information. So the method is suitable for the grid development direction under "Big Data" background, possessing remarkable theoretical research significance and engineering practice value.