Abstract:Effective flood forecasting can significantly prevent losses related to severe floods. However, nonlinear and dynamic flooding processes are difficult to simulate using traditional static networks, necessitating the use of dynamic networks with feedback connections and time delays. In this paper, we introduced a general framework to uniformly describe different dynamic networks, including the focused time delay network (FTDN), layered recurrent network (LRN), and nonlinear autoregressive network with exogenous inputs (NARX) network, which we used to construct 1080 models with six different lead times, ten neuron numbers, and six different time delays to forecast the reservoir inflow in east China. Three hybrid algorithms were utilized for training the models: steepest descent (SD), Broyden–Fletcher–Goldfarb–Shanno (BFGS), and Levenberg–Marquardt (LM) mixing with real-time recurrent learning (RTRL), respectively, which is an online implementation algorithm suitable for real-time flood forecasting. Early stopping was applied to improve model generalizations. In addition to the number of hidden layer neurons, we analyzed the impact of input time delays on the model error. The numerical and experimental results indicated that the training errors of the models exhibited significant change trends with the number of neurons and time delays. The BFGS algorithm demonstrated better stability and convergence than the SD and LM algorithms. For 1–6 h lead times, the NARX-based real-time flood forecasting model achieved the highest accuracy though this advantage is not obvious.

Modeling and Adaptive Control with Fuzzy Neural Networks-Selected Papers from the 6th International Symposium on Neural Networks

A combination of traditional time series forecasting models with fuzzy learning neural networks

Neural network model based predictive control for multivariable nonlinear systems

A delay damage model selection algorithm for NARX neural networks

Time series prediction based on NARX neural networks: An advanced approach

Time Series Prediction Model Based on Non-Single Fuzzy Regular Network

Control of Nonaffine Nonlinear Discrete-Time Systems Using Reinforcement-Learning-Based Linearly Parameterized Neural Networks

Support Vector Regression Based Nonlinear Model Reference Adaptive Control

A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction

Adaptive Control of Nonlinear Systems Based on Quasi-ARX Multilayer Learning Network Models

Recurrent-Neural-Network-Based Multivariable Adaptive Control for a Class of Nonlinear Dynamic Systems With Time-Varying Delay

Adaptive critic-based tracking control of non-affine nonlinear discrete-time systems with unknown dynamics

Long-term time series prediction with the NARX network: An empirical evaluation

Simple Recurrent Neural Network‐Based Adaptive Predictive Control for Nonlinear Systems

A Predictive Control Method of Nonlinear Systems

Nonlinear System Modeling by Competitive Learning and Adaptive Fuzzy Inference System.

Nonlinear System Identification Based on NARX Network.

Time Series Forecasting Using a Hybrid RBF Neural Network and AR Model Based on Binomial Smoothing

Comparison of NARX Neural Network and Classical Modelling Approaches

Real-time flood forecasting based on a general dynamic neural network framework

A Method to Model Nonlinear Systems by Neural Networks