Analysis of demand variability patterns through neural network

The paper illustrates the design and implementation process of a neural network to identify characteristic patterns of variability in a demand signal, extracting peculiar and critical elements. The starting assumption comes out from an analogy with the behaviour of general manufacturing and industrial processes. With this statement, it was possible to develop classical statistical process control tools to compile X-R charts, defining the progress of average and dispersion value of the demand. The introduction of a decision support system, a Multi-Layer Perceptron neural network, helps the operators to identify common and special causes of variability and gives an hint to plan specific market analysis to better understand customer’s behaviour. In particular, after being trained to recognize a set of formal process patterns, the architecture classifies a recognition window of data to identify the existence of dominant or influent effects on the demand signal.