In this approach, the expert is allowed to express his knowledge in a more complex way, as he may declare a relative necessity degree for each elementary premise. In the expert system approach, the relative necessity for a context was defined globally).

For example, the expert could describe his knowledge about a favourable zone for irrigation as being :

"Near hydrographic network : totally necessary and at an altitude less than 300m : quite necessary and near roads : very necessary"

This knowledge can be represented by a production rule as defined above

IF (favourable zone for irrigation) THEN       (near hydrographic network : totally necessary (+1.0)) AND       (altitude less than 300m : quite necessary (+0.4)) AND       (near road : very necessary (+0.6)).

So we propose to solve this kind of problem-solving approach by neural networks techniques. The result is a potentiality map giving a realization degree of the corresponding rule for each point.

A rule corresponds to a particular situation, describes the ideal context for resolution of a particular problem and is represented by a neural network [Zah92]. The inputs of the net are represented by realization degrees of all possible atomic propositions expressed in the conclusion part of the rule , "near a road", "elevation less than 300m", "near hydrographic network", ...

These inputs take values in the interval [-1,+1]:

• the value -1 corresponds to the case of false/absent proposition,
• +1 to the case of true/present information,
• and 0 to the case of unknown/no information.

These values are modulated by the necessity degrees , expressions related to premises, associated to each atomic proposition.