NEURAL NETWORK, NEURAL NETWORK SYSTEM AND NEURAL NETWORK PROCESSING PROGRAM

摘要

PROBLEM TO BE SOLVED: To provide a neural network capable of learning a time series characteristic with a simple structure. SOLUTION: A neural network system is provided with a preprocessing means to which an input vector is inputted and which combines respective elements of the inputted vector to perform nonlinear transformation and outputs the respective elements, a first layer to which a value subjected to nonlinear transformation by the preprocessing means is inputted and which branches the input into a plurality of parts and outputs the branched parts, a second layer to which a value obtained by integrating values obtained by multiplying an output of each unit of the first layer by a weight coefficient acquired by learning is inputted and which converts the value with a predetermined function and outputs the converted value, a third layer to which an output obtained by integrating outputs about a distribution model component corresponding the state (k) of each unit of the second layer is inputted and which multiplies the input by an output ahead by one hour of a state corresponding to a fourth layer in the same event and outputs an output obtained by the multiplication, the fourth layer to which an output about the state of each unit of the third layer is inputted and which divides the input by what is obtained by integrating the state with an output about an event/ class and outputs the result of this division, and a fifth layer to which an output obtained by integrating outputs about the state of each unit of the fourth layer is inputted and which outputs the input as analysis results of the event/class.