Strategic production planning with neuro-fuzzy systems

Production planning means primarily to sort, e.g., thousands of products with respect to their importance to the company. This helps the management to focus on the relatively small number of strategically important products and plan their production process first. But what if strategic importance, and, for example, cost criteria contradict each other. In this case the effect has to be discussed, if we suppose different product plans. Usually, for simplicity reasons, only one criterion concerning importance is taken into account and a backward reasoning is trivial. We introduce a method based on fuzzy logic that considers a large number of criteria concerning the importance of a product. We use fuzzy rules in a similar way as Pareto's rule to achieve an ordering of the products for each criterion. Afterwards this system is transformed in an equivalent neural network. The described method of internal feedback is a straight forward consequence of backpropagation. The idea is simply to propagate the error signals back to the input layer and to modify the input activations instead of the weights. This results in an approximation of the desired target vector in the output layer, leaving the weights fixed. The differentiable formulation is the foundation for backward reasoning in production planning.