MIMICKING AN EXPERT GREENHOUSE GROWER WITH A NEURAL-NET POLICY

A neural-net approach to specifying an expert control policy for the g reenhouse environment is proposed. The main idea is to utilize data ro utinely collected in a commercial greenhouse, rather than to interroga te the grower verbally. The expert should, however, specify which inpu t information seems important. Two automatically collected data sets ( Avignon, France, and Lake City, Florida) were used to test the propose d method. Comparison of the temperature operating-point policies for t omato in these two greenhouses showed that the general decision patter n is similar, but differences of several degrees are evident. This met hod is thought to be more objective and less time consuming than the t raditional expert system approach. Ir requires, however, a considerabl e amount of data.