Abstract:
Optimal control of the greenhouse environment requires a model of the greenhouse-crop system.
Detailed crop models contain many state variables, which makes complete optimization impractical.
We chose to search for a sub-optimal solution, which utilizes a single costate variable.
In this approach the simulation and performance criterion are still based on the complete model, but the control decisions are approximate.
In analogy with a simplified tomato model, normalization of the costate variable with respect to the sunlit leaf-area-index (LAI) was expected to yield a simple sub-optimal control policy.
The normalization is of the form
P = [p + kg] h,
where P is the the normalized costate, p is the (proper) dry-matter production costate (effort to produce the marginal unit of dry matter), k is the price of tomato, g is the fraction of dry matter converted to fruit and h is the sunlit LAI. It is assumed that in vegetative stage g = 0 and in the reproductive stage k, g and h are constant.
Computational experiments with a complex tomato model (TOMGRO) showed that the P is almost constant for a growing season.
|