Optimisation of short term log allocation

Abstract

The objective of the project was to develop an economic model for the management of short term log production and implement it in a prototype computer program. An algorithm was required that could be used in a central system that would control field computers that were running dynamic programming log bucking software used for log production in the forest. The system had to be capable of selecting cutting strategies for crews to follow and provide values for log products that would result in efficient forest resource allocation during harvesting.

The project investigated development of a Dantzig Wolfe linear programming decomposition model that uses Dynamic Programming to generate new columns. The hypothesis tested was that an economic model of log production based on Dantzig Wolfe decomposition and Dynamic Programming column generation would give maximum value log resource allocations that were contingent on company constraints. The Dantzig Wolfe algorithm was successfully implemented using the C programming language. Initial feasible solutions to start the Dantzig Wolfe algorithm were generated by the program using an adjusted Two Phase procedure. The accuracy of the program's calculations was verified by comparison with a problem that was solved using a spreadsheet. The program has been shown to provide feasible maximal solutions to two hypothetical problems. The program has generated new cutting strategies and values, as well as demonstrated iterative progress towards a maximum, in doing so providing validation of the underlying economic model proposed for log production.

The basic decomposition model could be extended to improve its capability and to provide a closer representation of the short term planning problem. Development of the program into a production model would include improvements to the program in the following areas: incorporation of uncertainty, adding restrictions on product combinations and cutting strategies and crew allocation to stands. For larger problems acceleration of computation will be required. This may be achieved by software improvements or multiple processor computing.