Variational Decision Models for Strategic Resource Allocation Under Market Uncertainty

Abstract

As electric vehicles (EVs) become more popular, the challenge of allocating resources is complicated by the variable loads on existing networks, differences in EV charging demand, capacity availability, electricity prices, and user preferences. Current optimisation methodologies for EVs use preprogrammed models or uncertainty scenarios that are inflexible in learning the market and congestion conditions. This paper presents an EV charging resource allocation model using a Learning-Based Markov Switching Dynamic VRI (LB-MS-DVI) framework in the presence of market uncertainties. The framework is based on the interaction between the charging parties, represented by variational inequalities, and changes in market conditions, represented by a finite-state Markov process, which includes conditions such as price, volatility, and congestion. The online stochastic-approximation mechanism sequentially updates the equilibrium operator by observing the data, which allows the model to change its operator to adapt to the changing operating conditions. Variational iteration projections are applied to obtain feasible charging allocations while applying the shared capacity constraints. Cost, energy shortage, equilibrium residual, convergence, regime-recovery time and utilization are key performance measures evaluated. An experimental study on the sensitivity of the proposed framework regarding learning rate, market volatility, risk aversion, and congestion coupling is conducted through proof-of-concept experiments with synthetic price, congestion, and market demand sequences. These results illustrate the computational behaviour of the proposed formulation and serve as a stepping stone for subsequent computational analysis of actual EV charging traces provided by the ACN-Data environment.