One of themajorresponsibilities oftheindustrial unitsisinventoryplanningand control. Nowadays, production, service and industrial systemsutilizeof aninventory management system. Manyrealsystemsdealwith different items. For instance, the distribution center of a chain-store manages its inventory system for a lot of items. Similar modes exist in wholesalers, third-party logistics centers, as well as department stores. Inventory system design has an important effect on cost reduction of firms in order to determine optimal or near optimal value of order quantity and reorder point. Inventory systems have constraints as many other systems. While the constraints make a model more practical, they increase the complexity of the model and limit the solution approaches of the model. This thesis deals with an inventory system under continuous reviewwith multiple items and resource constraint. Resource constraint is in fact the budget constraint that can be spend to purchase items. Itisasoftconstraint that Includedintheobjective function.If consumption of the budget become greater than available budget, shortage cost is incurred. Customers demand have Poisson discrete distribution and inventory system receive orders after fixed lead time. In the considered model, four types of costs are incured: orderingcost, holding cost, penalty cost for customer backorders and resource shortage cost. The goal is to determine the order quantity and reorder point so that the total system cost is minimized. A heuristic method have been presented to determine the ‌(r,Q) policy. Thismethodconsistsoftwo stages. Thefirststageis based onreduction of item'sinventoryposition. If the solution is not optimal in first stage, a local searchis usedinthe second stage. Numerical examples demonstrate that proposed method reduces total cost more than other methods. Computational times of both proposed and existing methods increases dramatically by increasing problem dimension. Inorder toovercomethis incompetence new method has been presented based on neural network method. It is shown that this approach is able toobtainorderingpoliciesforlarge-scaleproblemsin a resonabletime in comparison withexisting methods