Integration of legacy sub-system components into an object-oriented simulation model of a complete pastoral dairy farm

Abstract

An object-oriented (OO) approach to developing a computer simulation of a complete pastoral dairy farm system is outlined. Since component models already existed which encapsulated the functionality of the high level objects in the overall system it was a primary design goal to be able to incorporate them into the OO framework. These components had been developed in procedural programming languages and techniques were investigated to incorporate them with little or no re-coding, and which allow them to be used in an ‘object-like’ way. The techniques involved recompilation of legacy source-code as dynamic link libraries (DLL's), and accessing stand-alone executable programs through inter-process communication (IPC) protocols. A cow metabolism model developed in the proprietary ACSL environment, and a pasture growth model for which FORTRAN source code was available were incorporated into the Smalltalk OO framework using the Microsoft Component Object Model (COM) techniques. Multiple instances of each component (each representing a distinct real farm entity) were required by the OO framework and the COM methodology was able to support this. The integration of these components into the host OO framework involved using proxy objects to provide translation to/from the COM messaging, thus providing the OO framework with consistent inter-object communication.

Introduction

This paper describes techniques for integrating existing software models of real-world entities — legacy component models — into a broader object-oriented (OO) simulation framework. As discussed in a preliminary account of this work (Neil et al., 1997, Sherlock et al., 1997), it was advantageous to do this as part of our on-going development of a simulation model of a complete pastoral dairy farm. The work described here is a further development of those ideas, which are clearly applicable to other systems having the same general characteristics as our specific problem.

Pastoral dairying (the predominant dairy production system in New Zealand and parts of Australia and South America) utilises grazed pasture as the principal feed source (Holmes and Wilson, 1987). The major discrete entities are cows (both lactating and non-lactating) and paddocks (enclosed land areas growing pasture herbage for grazing or, less frequently, crops for feed supplements). The basic management goal is to maximise sustainable economic return, and broadly this requires maximising both production and utilisation of feedstuffs. However, the presence of complex feedback loops in a system of inherently complex components makes a predictive understanding of the overall system behaviour very difficult (e.g. overall pasture production depends on the pattern of harvesting, which depends on the pattern of feed demand, which depends on the calving pattern, which depends on the nutritional state of the cows, which depends on overall pasture production). These difficulties are compounded by uncontrollable, and largely unpredictable climate variables (the dominant system inputs) which make unambiguous farmlet experiments virtually impossible to perform.

Numerous component models which simulate aspects of pastoral dairying do exist. For example, the mechanistic MOLLY bovine metabolism model of Baldwin (1995) is extremely detailed and accurate, and is the result of some 30 years of development. However, there is a clear need for a whole-farm system computational model which would allow agricultural scientists to conduct simulation experiments unbiased by uncontrolled variables. Such a model would also minimise the need for (and focus the direction of) costly field investigations — particularly those which centre on management practices (e.g. stocking rates, grazing patterns, supplementary feeding strategies etc.). This framework would support simulation experimentation beyond current normal farm system operational boundaries and provide a basis for automated searches for optimal conditions (e.g. see Mayer et al., 1996, Hart et al., 1998). The development of a whole farm system simulation model should, however, capitalise on the existing models of system components. A primary design goal should therefore be to provide a framework in which the components can be incorporated in a way consistent with the overall design of the model. In this way the whole-farm model should also constitute a repository for the best current description and understanding (expressed in an operational form) of the system components and processes and will also provide a contextual framework for future component model development and refinement. The development and maintenance of such a research model is a long-term goal of this group; preliminary reports have been given by Sherlock and Penno (1996) and Sherlock et al. (1997).