3.4.4 Solution techniques
The DSS will allow the logistics system model to interact with the solver technique chosen for the particular model. These solution techniques do not differ from those generally used in operational research and management science. They include heuristics, optimisation and simulation.
Heuristics. Heuristics are known as 'rules of thumb' that direct the solution approach towards the best solution, but do not guarantee that it will be found. To repeat the definition given by Hinkle et al as cited by Ballou (1992):
A heuristic...is a short cut process of reasoning...that searches for a satisfactory rather than an optimal solution. The heuristic which reduces the time spent in the search for the solution of a problem, based upon the analogous human trial and error process of reaching acceptable solutions to problems for which optimising algorithms are not available.
The quality of the solution obtained by using heuristics depends a great deal on the quality of heuristics used. If quality heuristics can be found, a near optimal solution is achievable but finding such heuristics may be elusive. See the examples of heuristics in your text (pp.36 and 37) and note their shortcomings in finding an optimum solution.
Optimising solutions. The optimising algorithms are based on linear programming (LP) techniques or mixed integer linear programming (MIP) techniques. These are powerful mathematical tools and provide an 'optimum solution'. This means that once decisions parameters are selected, the DSS will come out with an optimum solution. This solution may be regarding the location of a facility or the allocation of demand to various facilities. In a survey of computer softwares for facility location in the USA , Ballou(1999) found that LP and MIP remain the dominant method of problem solution (90 percent of the models). It is beyond the scope of this study guide to indulge into the intricacies of LP and MIP techniques; your text provides an example of the mathematical procedures involved in a simple optimisation solution involving LP. The following reading provides examples of optimisation solutions using LP and MIP techniques. These simple problems can be solved, as illustrated, using a standard spreadsheet program, but when you are trying to solve a complex network problem with many variables, the use of a powerful computer becomes essential.
Reading 3.4
Chopra & Meindl (2001) extract from Chapter 11 'Facility decisions: network design in a supply chain', in Supply Chain Management Strategy, Prentice Hall , pp.318-327.
Reading 3.5
Powers, RF (1993) 'Optimization models for logistics decisions', Journal of Business Logistics , vol. 10, no. 1, pp.106-121.
Simulation. Simulation methods are fundamentally different from optimisation techniques. The simulation model of a supply chain network cannot provide an optimum result. It can only provide an output based on input and the mathematical relationship of the various elements of the supply chain as defined by the model. If the designer is concerned with a limited number of options, it is possible to use simulation to generate costs and service outputs of the different networks (including allocations of demands) by manipulating inputs. The comparison of outputs of the various configurations can give the designer the best option.
Since the designer is manipulating the inputs, he or she determines the various locations and demand allocation scenarios. Therefore, the output will only indicate the performance of these particular networks and flow parameters. Whether the optimal answer lies within the experiment or outside is a question which needs to be answered before the results can be accepted. In practical supply chain network design, the options available may be so numerous that it would be virtually impossible for the designer to evaluate all possible scenarios to seek out the optimum configuration.
Nevertheless, simulation is a useful tool because it is dynamic and allows the designer to study variations in the output by manipulating the input. The impact of an individual input element, such as demand variation, transport rate increase or inventory level build up, can be studied. Additionally, future scenarios can be postulated in order to predict the likely effect on the logistics costs of the supply chain.
Though simulation is not very popular in the USA , Gilmour (1993) cites the example of the Mayne Nickless model developed in Australia which is basically a simulation model capable of giving distribution costs as output for various scenarios set by the operator.
You should note that simulation is not an optimisation tool and the output of the simulation model will depend on how the designer configures the system. When you have a very large market and a very large number of possible options regarding allocation and location, it becomes impossible to try all these options one by one in order to seek out the best configuration and the results obtained from simulation are not likely to be optimal. When the options are limited in number, perhaps because the market is small and possible scenarios are limited in number, simulations become an effective tool to seek out the best solution among these different options. The US market is big and large firms are confronted with decisions which cannot be reduced to a manageable number of options for the simulation to provide an optimum solution.
The following reading provides you with account of a practical use of DSS in Australia using simulation modelling.
Reading 3.6
Gilmour, P (1993) extract from Chapter 9 'Planning the logistics system', in Logistics Management: An Australian Framework , Longman Cheshire, pp.254-266.
Activity 3.3
Identify the types of decision support tools used by Timberland and Toyota .
What kind of solution techniques have been used by the various models described by Gilmour? What kind of results can be obtained by the models? Write a short analytical answer of about 500 words.
Which solution technique to use? This is a fundamental question. There are several approaches to this problem. Though conceptually one would think that 'optimisation' is the best answer, there are practical limitations. The optimal finding procedures associated with a problem in a complex supply chain network require very long computational time. This is a special feature of models that use MIP solver techniques. Since it is sometimes impractical to run the computer for such a long time, one solution is to use a hybrid technique. A common approach is to reshape the problem definition in the model so that a solution is found in less time. The most common approach is to use heuristics to find an approximate solution from which further computation proceeds. It is also possible to incorporate all solution techniques within a DSS model. Guedes et al (1993) claims to have developed a decision support system named Stratovision in the UK which has the hybrid environment incorporating heuristics, optimisation and simulation techniques. Ballou's (1999) survey reveals that the majority of the computer models in use in the USA can search for answers for the basic network design problems of finding the most economic number of facilities, their locations and size. The search proceeds from a master list of suggested facilities and is automatic.
Back Page