Fig 2 The principle diagram of the niche
ELMAN neural network based string level over prediction error modeling: This study introduces a cascaded lead forecasting error network, anchored in the ELMAN neural network architecture, and consisting of three distinct ELMAN neural networks, each assigned a unique function: the historical data error generation network (Network 1), the lead forecasting error network (Network 2), and the prediction output network (Network 3). The core idea behind this model’s design involves using the error between Network 1’s predicted values and the actual values as the target output for training Network 2. Network 2’s testing set inputs mirror those of Network 3, rendering Network 2’s output the lead forecasting error of Network 3’s output, namely, predicting the error between Network 3’s predicted values and their actual values in advance. Consequently, the subtraction of Network 2’s forecasted error from Network 3’s predicted output is aimed at reducing the actual error. The model’s principle is depicted in Figure 3.