A Sustainable AI-Based Inventory Optimization Framework with Carbon Cost Minimization Using Hybrid LSTM–GA–DQN

Abstract

Managing inventory has become more challenging because of the fluctuation in customer demand, product deterioration, market volatility, and environmental sustainability laws. Traditional inventory optimization techniques struggle to incorporate and optimize for non-linear demand forecasts and minimize carbon emissions. Recognising these limitations, the present research proposes a sustainable framework for inventory optimization by applying different techniques such as Long Short-Term Memory (LSTM), Genetic Algorithm (GA), and Deep Q-Network (DQN). The historical relationship between demand is captured in an LSTM model to predict future demand, while the Genetic Algorithm is used for replenishment decisions and their associated safety stock levels to optimize the model. Furthermore, the DQN agent can acquire the optimal inventory policy as it moves in an unknown environment. A cost component for the carbon emissions is added to the objective function, so that sustainable green supply-chain operations are possible. The hybrid model will have the cost of holding and ordering cost will come down, there will be no shortage cost, diminishing cost, carbon emission cost, and service level & operational efficiency will be increased as a result of the hybrid model. Besides this, the model proposed here is better than the traditional inventory models from various numerical and sensitivity analysis that proves the sustainability, adaptability, and cost reduction capabilities of the model.