Xiaoyan He

Improving event-based rainfall-runoff simulation using an ensemble artificial neural network based hybrid data-driven model

An ensemble artificial neural network (ENN) based hybrid function approximator (named PEK), integrating the partial mutual information (PMI) based separate input variable selection (IVS) scheme, ENN-based output estimation, and K-nearest neighbor regression based output error estimation, has been proposed to improve event-based rainfall-runoff (RR) simulation. A hybrid data-driven RR model, named non-updating PEK (NU-PEK), is also developed on the basis of the PEK approximator. The rainfall and simulated antecedent discharges input variables for the NU-PEK model are selected separately by using a PMI-based IVS algorithm. A newly proposed candidate rainfall input set, sliding window cumulative rainfall is also proposed. These two methods are integrated to make a good compromise between the adequacy and parsimony of the input information and make contribution to the understandings of the hydrologic responses to the regional precipitation. The number of component networks and the topology and parameter settings of each component network are optimized simultaneously by using the multi-objective NSGA-II optimization algorithm and the early stopping Levenberg–Marquardt algorithm. The optimal combination weights of the ENN are obtained according to the Akaike information criterions of component networks. By combining all these methods, the simulation accuracy and generalization property of the PEK approximator are much better than traditional artificial neural network. The NU-PEK model is constructed by combining the PEK approximator with a newly proposed non-updating modeling approach to improve event-based RR simulation. The NU-PEK model was applied to three Chinese catchments for RR simulation and compared with two popular RR models, including the conceptual Xinanjiang model and the conceptual-data-driven IHACRES model. The results of simulation and sensitivity analysis indicate that the developed model generally outperforms the other two models. The NU-PEK model is capable of producing high accuracy non-updating RR simulation without the use of the real-time information, e.g. the observed discharges at previous time steps.

A new hybrid data-driven model for event-based rainfall–runoff simulation

Abstract A new hybrid data-driven model named PBK has been proposed to improve the event-based rainfall–runoff simulation. The PBK is developed by coupling partial mutual information-based input variable selection (IVS), ensemble back-propagation neural network (EBPNN)-based discharge forecasting and K-nearest neighbor algorithm-based discharge error forecasting. This model is proposed for solving the hard problem of how to implement non-updating rainfall–runoff simulation by data-driven models. For the purpose of solving the hard problems, the PBK model has the following innovations and improvements: (1) a newly proposed non-updating modeling approach without the using of the real-time information and can obtain higher simulation accuracy; (2) a newly proposed IVS scheme and a newly proposed candidate rainfall input set to ensure the adequacy and parsimony of the rainfall and antecedent discharge input variables; and (3) a newly proposed calibration method for the EBPNN to ensure higher simulation accuracy and better generalization property. This method is a combination of the NGSA-II, Levenberg–Marquardt algorithm, and the AIC-based combination weights generating method. For the purpose of comparing simulation accuracy with traditional non-updating data-driven models, a back-propagation neural network model (PB_R) and a linear model (CLS) were also studied. This study utilized event flood data of Dongwan catchment for intercomparisons between different models. The simulation results indicated that the PBK model outperforms other data-driven models and has higher accuracy and better forecasting capability.

A multi-core CPU and many-core GPU based fast parallel shuffled complex evolution global optimization approach

In the field of hydrological modelling, the global and automatic parameter calibration has been a hot issue for many years. Among automatic parameter optimization algorithms, the shuffled complex evolution developed at the University of Arizona (SCE-UA) is the most successful method for stably and robustly locating the global “best” parameter values. Ever since the invention of the SCE-UA, the profession suddenly has a consistent way to calibrate watershed models. However, the computational efficiency of the SCE-UA significantly deteriorates when coping with big data and complex models. For the purpose of solving the efficiency problem, the recently emerging heterogeneous parallel computing (parallel computing by using the multi-core CPU and many-core GPU) was applied in the parallelization and acceleration of the SCE-UA. The original serial and proposed parallel SCE-UA were compared to test the performance based on the Griewank benchmark function. The comparison results indicated that the parallel SCE-UA converged much faster than the serial version and its optimization accuracy was the same as the serial version. It has a promising application prospect in the field of fast hydrological model parameter optimization.

Accelerating the SCE-UA Global Optimization Method Based on Multi-Core CPU and Many-Core GPU

The famous global optimization SCE-UA method, which has been widely used in the field of environmental model parameter calibration, is an effective and robust method. However, the SCE-UA method has a high computational load which prohibits the application of SCE-UA to high dimensional and complex problems. In recent years, the hardware of computer, such as multi-core CPUs and many-core GPUs, improves significantly. These much more powerful new hardware and their software ecosystems provide an opportunity to accelerate the SCE-UA method. In this paper, we proposed two parallel SCE-UA methods and implemented them on Intel multi-core CPU and NVIDIA many-core GPU by OpenMP and CUDA Fortran, respectively. The Griewank benchmark function was adopted in this paper to test and compare the performances of the serial and parallel SCE-UA methods. According to the results of the comparison, some useful advises were given to direct how to properly use the parallel SCE-UA methods.

A heterogeneous computing accelerated SCE-UA global optimization method using OpenMP, OpenCL, CUDA, and OpenACC

The shuffled complex evolution optimization developed at the University of Arizona (SCE-UA) has been successfully applied in various kinds of scientific and engineering optimization applications, such as hydrological model parameter calibration, for many years. The algorithm possesses good global optimality, convergence stability and robustness. However, benchmark and real-world applications reveal the poor computational efficiency of the SCE-UA. This research aims at the parallelization and acceleration of the SCE-UA method based on powerful heterogeneous computing technology. The parallel SCE-UA is implemented on Intel Xeon multi-core CPU (by using OpenMP and OpenCL) and NVIDIA Tesla many-core GPU (by using OpenCL, CUDA, and OpenACC). The serial and parallel SCE-UA were tested based on the Griewank benchmark function. Comparison results indicate the parallel SCE-UA significantly improves computational efficiency compared to the original serial version. The OpenCL implementation obtains the best overall acceleration results however, with the most complex source code. The parallel SCE-UA has bright prospects to be applied in real-world applications.

A novel hybrid data-driven model for multi-input single-output system simulation

Artificial neural network (ANN)-based data-driven model is an effective and robust tool for multi-input single-output (MISO) system simulation task. However, there are several conundrums which deteriorate the performance of the ANN model. These problems include the hard task of topology design, parameter training, and the balance between simulation accuracy and generalization capability. In order to overcome conundrums mentioned above, a novel hybrid data-driven model named KEK was proposed in this paper. The KEK model was developed by coupling the K-means method for input clustering, ensemble back-propagation (BP) ANN for output estimation, and K-nearest neighbor (KNN) method for output error estimation. A novel calibration method was also proposed for the automatic and global calibration of the KEK model. For the purpose of intercomparison of model performance, the ANN model, KNN model, and proposed KEK model were applied for two applications including the Peak benchmark function simulation and the real-world electricity system daily total load forecasting. The testing results indicated that the KEK model outperformed other two models and showed very good simulation accuracy and generalization capability in the MISO system simulation tasks.

An improved hybrid data-driven model and its application in daily rainfall-runoff simulation

In previous literatures, a coupled data-driven rainfall-runoff (RR) model, NU-PEK, has been proposed and successfully applied in hourly RR simulation task. However, numerical experiments show that its performance for daily RR simulation is unsatisfactory. It is noticed that the poor performance is due to the inability of the original model to capture the much higher non-linear characteristics contained in the daily data. In order to improve the nonlinearity simulation capability of the original model, an improved model named NU-PKEK and its calibration methodology are developed in this paper. The improved model is constituted by adding a K-means clustering module and utilizing multiple NU-PEK modules instead of using only one NU-PEK model. This study applies the improved model, the Xinanjiang model, and the original model for daily RR simulation in Chengcun catchment for intercomparison and verification. The simulation results prove that the NU-PKEK performs best, and has better simulation and forecasting capability.

Fast hydrological model calibration based on the heterogeneous parallel computing accelerated shuffled complex evolution method

ABSTRACT Hydrological model calibration has been a hot issue for decades. The shuffled complex evolution method developed at the University of Arizona (SCE-UA) has been proved to be an effective and robust optimization approach. However, its computational efficiency deteriorates significantly when the amount of hydrometeorological data increases. In recent years, the rise of heterogeneous parallel computing has brought hope for the acceleration of hydrological model calibration. This study proposed a parallel SCE-UA method and applied it to the calibration of a watershed rainfall–runoff model, the Xinanjiang model. The parallel method was implemented on heterogeneous computing systems using OpenMP and CUDA. Performance testing and sensitivity analysis were carried out to verify its correctness and efficiency. Comparison results indicated that heterogeneous parallel computing-accelerated SCE-UA converged much more quickly than the original serial version and possessed satisfactory accuracy and stability for the task of fast hydrological model calibration.

An Improved “Dynamic Control Operation Module” for Cascade Reservoirs

Reservoirs are one of the most efficient projects for water resources management, and also play an important role in flood control and conservation. Dynamic control of the reservoir flood limited water level (FLWL) is considered as an effective factor to ensure safety of the flood control during the flood season. The maximum allowed water level of cascade reservoirs is also a fundamental key element for implementing reservoir water level dynamic control operation. In this paper, we discussed and improved “Dynamic Control Operation Module for Cascade Reservoirs” (DCOMR). Therefore, a set of new formulas and new methodologies were proposed (NDCOMR) which considers intermediate variables forecast information, the release of the upstream reservoir and interval flow forecast information in effective lead times, which is applied to the dynamic operation of the maximum allowed water level of cascade reservoirs. The Bikou and Miaojiaba cascade reservoirs were selected as a case study. Based on numerical experiment results, the maximum allowed water level of cascade reservoirs at current time determined by NDCOMR was much safer than that determined by DCOMR. The NDCOMR is more rational and safer than DCOMR for flood control, without the need of reducing flood control standard.