Research On Voltage Control Of Smart Grid Based On Expandable Deep Width Learning

The access of large-scale distributed energy resources has brought greater risks and greater flexibility to the voltage control of the power system.On the one hand,the grid connection of numerous distributed energy resources increases the risk of grid voltage violation.But on the other hand,the reactive power regulation capability of distributed energy resources also enhances the flexibility of the voltage control process.The highly information-based smart grid provides a technical guarantee for the grid connection of numerous distributed energy resources.How to quickly and effectively control the grid voltage in the smart grid is an urgent problem for electric power practitioners to solve.This thesis studies the voltage control problem of smart grids with the participation of distributed energy resources,and designs model-free deep reinforcement learning algorithms for smart grid voltage control aiming at the difficulty of accurate modeling due to the increasing complexity of smart grids.The main innovations of this thesis are as follows:(1)Based on the deep neural networks of the deep learning and the wide neural networks of the broad learning system,an expandable deep width neural network is proposed.Then,an expandable deep width learning algorithm is proposed by combining the expandable deep width neural networks with the deep deterministic policy gradient.In addition,the concept of quantum state in quantum mechanics is introduced into the data distribution process of the expandable deep width neural networks,and an expandable quantum deep width learning algorithm is proposed.(2)The reactive power compensation characteristics of flexible loads and traditional reactive power compensation equipment are analyzed,and a three-state energy voltage regulation model is proposed.The proposed model is applicable to numerous devices with reactive power compensation capability.To solve the problem that the traditional automatic voltage control system has the risk of uncoordinated control due to different time scales,a smart grid voltage coordination control framework is proposed based on the proposed three-state energy voltage regulation model.The proposed framework can eliminate the risk of uncoordinated control between the secondary voltage control system and the primary voltage control system due to different time scales.In addition,to adjust the grid voltage by controlling the various voltage regulators quickly and efficiently,a unified time-scale coordination primary voltage controller based on the expandable deep width learning algorithm is designed.The designed controller can coordinately control each voltage regulator quickly and effectively.Finally,the feasibility and effectiveness of the proposed control framework,controller and algorithm are verified by simulation results.(3)To solve the problems of heavy communication and computation burden,susceptibility to the single point of failure,lack of protection of information privacy,and different time scales in the traditional centralized voltage control framework,a real-time smart grid distributed voltage control framework is proposed based on the three-state energy voltage regulation model.The proposed framework can coordinately control the voltage in all regions of the entire grid.In addition,to coordinately control the voltage regulators in the region in real-time,a real-time region voltage controller based on the expandable quantum deep width learning algorithm is designed.The designed controller can coordinately control the voltage regulators in the region quickly and accurately.The simulation results demonstrate that the proposed control algorithm has higher control accuracy and faster control speed than the deep learning method and traditional proportional-integral-derivative method.