원문정보
초록
영어
In order to overcome the problems which are difficult to be accurately predicted, such as voilent vibration, large amplitude, and pseudoperiod, we put forward a load-classification method and a model-selection method following a multi-model merit. The multi-model merit can be realized by the result of Network training, so we can forecast the load of iron and steel enterprises respectively. In this way, we can avoid the limitations of traditional load forecasting methods which simply depend on sample datas. In the framework of this model, to minimize the load forecast error is the target. On the one hand, it can be convenient to add new models into the framework, so as to improve the accuracy of the prediction, find more characteristics of the load, and better the model. On the other hand, based on the load data, we can also adaptively change the way the model is formed, so as to expand the applicability of prediction methods. By simulating different load management forecasting systems, we confirm that the effectiveness of the proposed method is verified. A software package based on the methods presented in this paper for power systems scheduling is also completed. Some native steel generation corporations have already used the system.
목차
1. Introduction
2. Selecting-best Multi-model Prediction Framework of Adaptive Data Quality
2.1. Model space
2.2 Combination of multi-model
2.3 Frame design
3. Multi-modeling and Simulation Analysis of Daily Load Forecasting
3.1 Combination of moving average and linear regression model
3.2 Artificial neural network [17]
3.3 Support vector regression model
4. Multi-model Selection
5. Conclusion
Acknowledgements
References