Enhanced EV Spatial Distribution Estimation for Grid Planning using Neural Architecture Search-Guided Physics-Informed Neural Network and Pufferfish Optimization Algorithm

Electric vehicles (EV) play a vital role in modern transportation, and accurately estimating their spatial distribution is essential for effective grid planning. The utility grid management system will optimize energy utilization while balancing network capacity and develop new infrastructure components which reduce the electrical grid stress during EV integration increases. However, estimating EV distribution in grid planning is challenging due to unpredictable charging patterns, dynamic mobility behavior, and varying energy demands, which complicate load forecasting and infrastructure optimization. To overcome these drawbacks, this paper proposes a hybrid approach for spatial distribution of EV. The process begins by gathering data from vehicle registration dataset, which is then passed through a pre-processing phase. Regularized Bias-Aware Ensemble Kalman Filter (RBAEKF) is employed to clean and remove the missing value in the input data. The pre-processed output was fed to Neural Architecture Search-Guided Physics-Informed Neural Network (NASPINN) the data enters the classification phase, to enhance the accuracy of classifications. The class 0, class 1 and class 2 of EV distribution is successfully classified by using NASPINN. The weight parameter of NASPINN is optimized using Pufferfish Optimization Algorithm (POA). The NASPINN-POA technique is implemented in MATLAB and evaluated using various performance metrics, including accuracy, precision, recall, F1-score, specificity and Root Mean Squared Error (RMSE). The results show that the NASPINN-POA method outperforms existing approaches, such as Sparrow Search Algorithm-Back Propagation Neural Network (SSA-BPNN), Particle Swarm Optimization (PSO), Eurasian Oystercatcher Optimizer-Quantum Neural Network (EOO-QNN), Bayesian Network (BN) and Long Short-Term Memory (LSTM). The proposed NASPINN-POA method enables an accurate spatial distribution of EV with 98.8% accuracy, 98.4% recall and achieves an MAE of 1.07 to optimize grid planning by minimizing errors.