A Study On Load Forecasting And Risk Assessment Of Distribution Network Under The Increasing Growth Of Electric Vehicles

With the growing environmental issues and global energy crises,the demand for clean energy is becoming more and more significant.The advancement and deployment of electric vehicle(EV)technologies provide an emergent solution to green the transportation sector.The in-depth research on applications of EV technology will be necessary to encounter increasing issues associated with energy security,climate change,air quality improvement,and clean energy environment.However,operational challenges exist in order to enhance sustainability and to obtain clean energy systems.Large-scale integration of EVs into residential distribution networks(RDNs)is an evolving issue of paramount significance for utility operators.Negative impacts are observed in low-voltage(LV)RDNs due to the added demand of EVs during uncontrolled charging scenarios at different residential locations.Large-scale deployment of EVs may deteriorate the performance of distribution components and create potential issues for the stable operation of RDNs.One of the possible solutions to overcome the risks associated with RDNs for massive penetration of EVs is to control the charging behavior of EVs in a coordinated way,which could further lead to defer the high investment needed for infrastructure upgradation of power networks.Consequently,coordinated charging schemes become an imperative solution for the smooth and effective integration of EVs.Since massive integration of EVs usually occurs at distribution level,therefore,one of the main objectives of this thesis is to present a deep insight into the impacts of EVs on RDNs based on controlled and uncontrolled EV charging schemes with various charging power levels under different EV penetration levels.In the competitive market of electricity,electric load forecasting is a substantial operational process,which permits utilities to overcome imminent demand management issues and rising concerns for optimal energy utilization.Development in load forecasting methods with new approaches suggests effective solutions to utilities towards reduction in operational cost.As the residential load is substantially increased with added demands of EVs,consequently,a forecasting model is necessary to predict the peak residential demand for effective load management in RDNs.Therefore,the second objective is to formulate a precise and optimum residential load forecasting method based on artificial neural networks(ANNs)for the effective load prediction with least mean absolute percentage error(MAPE).In addition,this research contributes regarding the operational process of future power networks by modeling the stochastic charging behavior of EVs and developing a prediction model for predicting the peak load demand of distribution networks.Unbalanced voltages prevent the effective and reliable operation of a distribution grid.This study implements a novel framework to examine risks associated with RDNs by integrating a residential forecasting model along with a stochastic model of EVs charging pattern.The main research work and contributions of this thesis are:(1)This study proposed an improved load forecasting method,by implementing an optimum non-linear auto-regressive neural-network with time-varying external input vectors(NARXNN).It can be precisely used in all seasons of a year to predict the peak demand of various domestic houses.Different variables associated with weather and seasonal variations are applied in terms of exogenous inputs,and endogenous inputs of NARX model contain historical load profiles.Unlike the earlier conventional short term load forecasting(STLF)methods based on ANNs,the proposed model is employed in a recursive approach by applying predicted open-loop output as a feedback input to reduce dependency on external data,which makes the forecasting model more robust and it further leads to improved accuracy of the model.Stability issues are also encountered in the proposed method,and optimum values of input time-lag and hidden layer neurons are determined.The results achieved from the proposed method shows notable improvements with less prediction error in comparison with other statistical approaches.(2)The forecasting model is further improvised by obtaining the optimized parameters of the NARX network by using the particle swarm optimization(PSO).In comparison with the conventional methods,optimal values of recurrent NARX-NN model are determined associated with feedback delays,hidden layer neurons along with input delays of the prediction model for peak demand estimation of various residential houses.Comparative analysis in demand forecasting is also presented for different seasons with traditional forecasting methods to validate the efficacy of the proposed model.Forecasting results are comprehensively improved by achieving the reduced forecasting error,fast learning and execution time,less variance and fluctuations,and better convergence speed.The intelligence and robustness of recurrent load forecaster are enhanced by implementing the optimization approach.(3)The prospect of EV technology demands several realistic scenarios to be considered for a profound assessment of EV impacts on the existing infrastructure of LV networks.Stochastic modeling of EVs covering behavioral characteristics of drivers along with other realistic factors for the actual estimation of an expected EV load on the power grid is much needed effort to make necessary up-gradation in distribution network infrastructure.In previous studies,the charging model of EV is either based on a deterministic approach or a stochastic model has not covered many important realistic aspects.Consequently,this study develops a comprehensive integrated stochastic EV charging model in which mobility data of EVs are used in a novel and detailed manner by considering several realistic aspects as well as behavioral characteristics of drivers comprising charging time,battery capacity,arrival and departure time,driving mileage,state-of-charge,traveling frequency,charging power level,and time-of-use mechanism under peak and off-peak charging strategies.In addition,the current EV market share is also incorporated for estimating the actual EV charging demand.Finally,the Monte-Carlo simulations are performed for all probabilistic calculations and estimation of EV charging demand.(4)Random phase-wise distribution of houses as well as EVs is identified as a common residential distribution network problem,which causes performance parameters of distribution network to surpass the permitted limit in multiple circumstances according to the international standards.With massive penetrations of EVs in RDNs,unbalancing issues are bound to enhance.In previous studies,the main focus is to analyze the voltage unbalance factor(VUF)only and none of them majorly focus to reduces the VUF.Therefore,this study develops a smart EV charging algorithm for mitigating the adverse effects of random allocation of EVs.An effective algorithm is proposed to reduce the VUF for optimal integration of EVs,which leads to the safe and reliable operation of LV networks.The VUF is reduced as an objective function subject to various constraints of EVs as well as constraints of network security.Charging and discharging states,charging/discharging power and connecting points(feeder phases: a,b,c)of EVs are optimally selected to minimize the VUF.The outcomes of the proposed strategy reveal that the controlled charging and discharging of EVs along with variable power rating have a considerable impact on the VUF and VUF values are substantially reduced in this approach by the proposed method.(5)As far as impacts of RDNs are concerned,the previous studies assumed average load profiles of residential houses for assessment purpose.Therefore,this study develops a mathematical modeling based framework,which is first to carry out the collective technical assessment of forecasted residential load along with stochastic charging load of EVs in RDNs.The output of the EV stochastic model obtained from Monte-Carlo simulations and output of residential forecasting model obtained from optimized NARX-NN model are utilized to evaluate power quality parameters of RDNs.The equipment capability of RDNs is evaluated to determine the potential overloads.In particular,grid contingencies are assessed including low voltage issues faced by the customers,domestic transformer limits,feeder losses along with voltage unbalance factor due to the random phase-wise house and EV distribution at various feeders of implemented benchmark system.A profound examination of EV charging strategies is presented for analyzing performance parameters of RDNs.Moreover,issues related to EV charging demand with location along with residential power demand with location of domestic houses are well investigated.The simulation results reveal that EV level of penetrations can be further enhanced along with reduced value of VUF by implementing controlled EV charging/discharging with the proposed method.