A Neural Network Prediction Model On Multiple Error Dimension Integration

Forecasting technique is an important research area in machine learning, and average accuracy is usually recognized as the main target of prediction, such as 2-Norm error. However, this accuracy pursuing forecasting model may not always targets on the solution with optimal impact in further usage of prediction result. Facing this issue, this paper proposes a composite prediction structure model based on Long Short-Term Memory (LSTM), which considers both the mean deviation and the maximum positive bias error between the predicted value and the actual value. A numerical study with practical photovoltaic (PV) data is presented and the result shows that composite prediction structure model can reduce more total cost from prediction error than pure model on 2-Norm error only.