Online Equivalent Modeling Of Terminal Power Grids And Its Application Based On An Enhanced Reinforcement Learning Method

The accuracy of load modeling directly influences power system operation and control.The expansion of power grids and the increasing integration of distributed generations(DGs)by renewable energy pose challenges for load modeling.Previous modeling studies have mainly concentrated on the loads connected to a single load bus.This thesis is dedicated to study the equivalent modeling of terminal grids in power systems,from the perspective of the parameter identification,the topological structure and its application.The aim of this thesis is to precisely describe the aggregated load characteristics of the terminal grids and the spatial uncertainty of internal DGs in time-varying operational conditions.On the aspect of parameter identification,this thesis employs the measurement-based strategy,and proposes an enhanced reinforcement learning(ERL)algorithm to identify the model parameters in an online manner.Adaptive learning rates and parallel tables of value functions are implemented in the ERL algorithm to improve its online-tracking performance.Three simulation cases are addressed.The results demonstrate that the ERL algorithm outperforms an existing reinforcement learning algorithm and the improved least-squares method in terms of convergence and the ability to track time-varying load characteristics.On the aspect of equivalent models,this thesis first proposes an equivalent model for area loads with multiple boundary buses in transmission networks based on Ward equivalence.This equivalent model considers the voltage-dependent static load characteristics,and is confirmed to be more accurate than a previously introduced model according to simulation cases.Subsequently,this thesis proposes an equivalent model for active distribution networks(ADNs)considering the spatial uncertainty of DG,which is caused by the dispersed locations of distributed renewable energy resources.The boundary injections of this model are composed of deterministic and uncertain components.The deterministic component is calculated based on the fitted power characteristics,while the uncertain component is described using a probability distribution.Simulation results demonstrate that the equivalent ADN model can accurately represent the aggregated feature of the ADN considering the spatial uncertainty caused by distributed renewable energy resources.On the aspect of application,the equivalent model of ADN is utilized in stochastic reactive power optimization of transmission networks based on the multi-scenario technique.the multiple scenarios caused by the probability distribution of the equivalent model are obtained using the Latin hypercube sampling.Based on this,the optimal solution is optimized,and probabilistic power flows of the equivalent system and the actual system are compared.Simulation results demonstrate that the error between the two systems is low and the ADN equivalent model is accurate in terms of stochastic reactive power optimization.