Optimal Neural Network Modeling Method Based on Data Noise Information

Aiming at some problems such as ldquoover-fitrdquo with ANN modeling, a new type of combined neural network and corresponding optimal modeling method are proposed in this paper. By this method, expectation error (namely, possible minimum error) is firstly estimated according to data quality; then, optimize network structure and choose optimal training result according to the difference between actual error and expectation error. To evaluate objectively the performance of model, a new evaluation index named error average power (EAP) is introduced. Simulation experimental results show that the methods mentioned above can work with good performance.