An Evolutionary Load Forecast Approach For Comprehensive Electric Vehicle Charging Allocation

The anticipation of large-scale Electric Vehicles(EVs)charging and discharging load could bring security and reliability challenges to the power system.As a smart load,EV requires an intelligently designed scheduling and pricing algorithm that takes into account the stochastic EVs user behaviour,grid charging capacity,battery characteristics,and real-time electricity price variations.The research is distributed into two phases,at the first stage,the focus is to develop a comprehensive understanding of the baseload forecast accuracy generated by employing state of the art optimal autoregressive neural network(NARX)for multiple,nonlinear,dynamic,and exogenous time-varying input vectors.To deal with the variable selection issues in day-ahead forecasting,a comparative empirical study is conducted with up-to-date benchmark expert models,considering load datasets from the northern Chinese power grid.Five different additional methods,i.e.Single Step elimination(SSE),step-wise regression(SR),least absolute shrinkage and selection operator(LASSO),ridge regression(RR)and a combination of RR and LASSO,i.e.elastic nets(EN)are considered for significant variable selection in the benchmark model category.The input predicting variables selection is exceedingly crucial in optimizing the performance of time series forecasting models,highly correlated variables are determined from the considered out of sample data set further to enhance the performance and accuracy of the proposed forecasting model.Automated variable selection and shrinkage method is applied to test and choose the best expert model from statistical state of the art benchmark algorithms.A Diebold Mariano(DM)test is carried out to validate the significant difference in the applied expert variable selection methods.An optimal hyper tuned NARX architecture is devised for day a head baseload forecast,the training of the proposed model is realized in a closed loop by feeding back the predicted results obtained from the open-loop model which,made the implemented model more robust compared with conventional feedforward and NARX approaches.The network is further improved by proposing a lightning search algorithm(LSA)to further optimize NARX network parameters with exponential weight decay(EWD)technique to control the input error weights.The recurrent nature of the applied model reduces its dependency on the external data and produce mean absolute percentage error(MAPE)of below 1%.Subsequently,more precision in handling daily grid operations with an average error improvement of 16-20% in comparison with existing computational techniques such as bagged regression tree(BRT),autoregressive integrated moving average with external inputs(ARIMAX)and conventional feedforward artificial neural network is achieved.In the second stage,a Markov Chain Monte Carlo Simulation is proposed for EV charging load estimation.A combination of two models is utilized to formulate a stochastic EV charging and travel schedule.Both temporal and categorically spatial uncertainties are taken into account by formulating arrival,departure,travel time,activity and mobility model individually,this helps in keeping the rationality among original data set and avoids any uncorrelated/ unrealistic samples to be selected during Monte Carlo estimation.Finally,a multi-objective comprehensive stand-alone solution is proposed considering the dynamic pricing computed from both baseload and charging load forecast to regulate EVs charging/discharging schedule intelligently.The proposed alternative heuristic charging strategy optimally configures solution indices.It provides a tradeoff between considered evaluation parameters taken from the perspective of power suppliers and EV users,thereby mitigating the effect of uncontrolled charging introduced by stochastic charge/discharge activities.The objective is to shift the peak hours load to non-peak hours with a reduction in average-to-peak ratio,charging cost minimization and maximize the availability of charging capacity for pledging the travelling plan determined by EV users.Different EVs penetration rates are tested to validate the performance of the proposed charging solution under massive EV integration,based on the driving pattern obtained from the Beijing National Travel Survey.