The SNP optimizer allows you to execute a cost-based planning. This means that it searches through all feasible plans to try and find the most cost-effective. Total costs refers to the following:

• Production, procurement, storage, and transportation costs
• Costs for increasing storage resources, transportation resources, production resources, and handling resources
• Safety stock violation costs
• Late delivery costs
• Shortfall quantity costs

As you are able to enter PPMs with fixed resource consumption in master data, you can also perform setup procedures in Supply Network Planning. You can therefore also use the SNP Optimizer for lot size planning.

The optimizer uses linear programming to consider all planning problem-relevant factors simultaneously in one optimum solution. As more constraints are activated, the optimization problem becomes more complex and, as a general rule, the time required to solve this problem increases. Optimization should generally be executed as a batch activity.

The Optimizer differentiates between continuous and discrete linear optimization problems.

There are three methods for solving continuous linear optimization problems with the Optimizer:

• Primal Simplex Method
• Dual Simplex Method
• Interior Point Method

All three methods arrive at an optimal solution. The main difference in the application of these methods may be the runtime. There is no general rule known for selecting the best method for a given problem (except testing each one of them). A good measure for the application is benchmarking on a test scenario because the optimal choice of the method depends mainly on the structure of the supply chain and less on the given input data. Therefore, in a productive environment, daily benchmarking is not necessary.

A problem is not continuous (that is, discrete) in Supply Network Planning, when the model contains:

• Discrete lot sizes for PPMs (in integers)
• Discrete means of transport
• Discrete capacities for transportation or production
• Minimum lot sizes for transportation or PPMs
• Piecewise linear cost functions for transportation, production, or procurement
• Fixed PPM resource consumption

If the optimizer is to include one of these named restrictions, you must use a discrete optimization method that can be selected in the SNP optimizer profile. Discretization can be performed in two ways: full search or partial search.

The full search uses Branch&Cut techniques. To avoid excessively long runtimes, you should limit the number of improvement steps and/or the runtime (in minutes) the Optimizer is allowed to use. An improvement step means that the system moves from one valid solution to the next improved valid solution. If you enter a search value in the appropriate field in the SNP Optimizer profile, the system checks both criteria simultaneously. If you make no entry, it results in an unlimited runtime, which means an unlimited number of improvement steps are allowed.

The partial search algorithm solves relaxations of the original problem sequentially to determine values for the discrete variables. A relaxation means that the discretization constraint is not taken into account. For the relaxed problem, the LP solver (the linear programming solution process) generates an optimal solution but with some decision variables violating their discreteness constraint. These violations are solved incrementally by fixing the decision variables to neighboring discrete values.

The piecewise linear cost function that you can define in the master data, differentiates between two important cases: convex cost function (cost per unit increases for higher volumes, for example, modeling over-time or night shifts) and concave cost function (cost per unit decreases for higher volumes, for example, modeling freight rates).

Convex cost functions do not complicate the planning problem and can be solved efficiently. Alternatively, they could even be modeled without using the feature of piecewise linear cost functions with alternative modes.

• Mode 1 with $50 per unit and a limited capacity of eight models;
• Mode 2 with $100 per unit and limited capacity of six models
• Convex functions of labor cost per day, assuming eight normal working hours and a maximum of six hours of over-time with doubled per hour wages

In contrast, concave piecewise linear cost functions could not be solved by an LP solver but using discretization techniques (mixed integer linear programming). If piecewise linear functions are modeled, but the optimizer is run without discretization, or the discretization horizon is less than the planning horizon, then the optimizer includes the definable linear cost function in addition to the piecewise linear cost function.

The Optimizer can differentiate between the priority of customer orders and forecasted demand. With strict prioritization, sales orders always have priority 1, the corrected demand forecast priority 5, and the demand forecast priority 6. Within each priority class all available cost information is used by the system to determine the final solution. When cost-based prioritization is employed, the optimizer uses penalty cost information from the product master data (SNP1 tab) to determine the optimal solution.

The primary focus of decomposition is to reduce the runtime, and memory requirements of optimization. Both Time decomposition and Product decomposition are strategies that can be used in conjunction with continuous and discrete optimization.

Decomposition gives the user a flexible tool for balancing the tradeoff between optimization quality and required runtime. Time decomposition speeds up the solution process by dividing the source problem into a series of sub-problems. These sub-problems are then solved sequentially. In general the quality of the solution is lower than when using the optimizer without time decomposition. Product decomposition speeds up the solution process by building groups of products. The complete model solves one product group at a time according to the window size selected. The rule of thumb is as follows: the smaller the window size, the less time it will take to find a solution, but the larger the window size, the better the quality of the solution found.

To reduce the size of the model to be optimized, the optimizer can plan only on the location product group level (assuming you have defined the demands at the lower level). Plans are distributed to lower level products based on the demand for the lower level products. To plan at the product group level, you must define hierarchies for products and locations in the hierarchy master. This data is used to generate the product-location hierarchy. You also must define the PPMs for the product groups and create the PPM hierarchy in the hierarchy master. If you set the Product aggregation indicator in the SNP Optimizer profile, the products are automatically aggregated into the respective groups for planning, and disaggregated after planning has completed.

This essentially allows you to plan a subset of your supply chain. You can limit the products or locations that are taken into account during the optimization run. For example, if optimization is carried out only until the plant level, but forecasts are at the customer level, the optimizer can total the demands at the plant level and use this value during the optimization run. The transportation times, for example, from the plant to the distribution center and to the customer, as well as the duration of the PPMs, are also taken into account.

Note: The optimizer plans all distributed demand for all locations in the distribution network before exploding the BOM and processing dependent demand in the production locations.