Modeling And Application Of A Different-Inputs-Same-Output Combinational Neural Network

Based on the research results so far on Combined Neural Networks and the problems arising from actual manufactures, an innovative neural network model is put forward in this paper. When solving practical problems, different neural network models with different predictions on the same problem can be set up from different affecting factors. This Different - Inputs- Same - Output- Combinational neural network model composes of many different neural network models, integrates these models and jointly uses their predicted data by assembly learning to gain an integrative prediction. The mechanism is explored and compared with the conventional BP models in the domain of freight traffic predicting problem. The results indicate that the predicting precision of this new model exceeds that of traditional BP models.After putting forward of the new neural network model, the following aspects are further studied and the performances of the new model are optimized.1. The optimization of the sub-network.BP neural network is selected as the sub-network of the new model. After analysis on the structure and algorithm of the BP neural network, the advantages and disadvantages are summarized and then the BP Algorithm is optimized, which offers an algorithm with better performance for the combined neural network.2. The optimization of the combined neural network model.The rule of the selecting of sample data, the determination of the number of sample data, and the pretreatment of sample data are analyzed. The creation of the sub-network and output-combination are studied. The researches on the determination of the number of layers and the number of nodes in the hidden layer are carried out.3. The application of the new model.The new model is applied in the forecasting of short-term electric power load, the simulation results indicate that the new model obtains better performance in generalization, learning speed and forecasting precision.