Active Optimal Dispatch Of Virtual Power Plants Based On Model Predictive Control

In order to achieve complete consumption of high-permeability distributed power energy and coordinated control of multiple resources,it is an effective solution to construct a virtual power plant to aggregate different distributed power resources and energy storage devices to achieve overall active power coordinated control.However,the randomness and volatility of wind power and photovoltaic power poses a huge challenge to the control process of virtual power plants.In the process of real-time control of virtual power plants,in order to achieve accurate tracking of the reference trajectory of power generation,it is necessary to introduce feedback links for closed-loop control,and model predictive control is an effective tool that can be utilized,and the prediction accuracy of the predictive model is undoubtedly another key.In order to deal with the uncertainty of the intermittent distributed power energy in virtual power plants,and to solve the problem that the prediction accuracy decreases with the increase of time scale,resulting in inaccurate prediction,this paper proposes the active power coordination optimization of virtual power plant based on improved model predictive control under the structure of multi-agent system.The dispatching strategy realizes the maximum consumption of distributed power and the accurate tracking of the virtual power plant power generation reference trajectory.Using the deep learning model based on long short-term memory,the ultra-short-term accurate prediction of new energy output is realized under the condition of sufficient historical data and insufficient historical data to cope with the randomness and volatility of its power generation.The closed-loop rolling optimization dispatching model of predictive control adopts multi-step dynamic rolling optimization,with the minimum deviation of the relative reference trajectory as the target,and the sequential quadratic programming method is used to solve the active power increment,and the coordination control of resources within the virtual power plant is completed.This paper simulates the proposed model based on the measured data of distributed power stations and virtual power plants in some regions inside China.In the case that the distributed power supply has just been put into operation and the historical data is not enough to complete the model training alone,the unit with long running time and strong output characteristics can be utilized by transfer learning to complete the training of the model;The LSTM prediction model is used as the prediction module of the model predictive control method.The active power of the virtual power plant is optimized,and compared with the traditional model predictive control method based on continuous prediction.The results of the example verify that the proposed method is effective and superior on the mask of virtual power plant active power reference trajectory tracking.