Research On Intelligent Methods For Mid-and Long-Term Runoff Forecasting And Optimal Generation Scheduling For Hydropower Stations

In China, there has been rapid development of hydropower in recent years. And several cascaded reservoirs, characterized by large quantities and huge installed capacity, have been put into operation. While the successive newly large-scale reservoirs putting into operation, China is facing the challenge that how to optimize these reservoirs. Hydrological forecasting and optimal operation are the two key factors for hydropower system management. Reliable inflow forecasting is fundamental to make high-quality reservoir operation schedules and reasonable scheduling scheme is vital to make full use of hydropower resources. For runoff forecasting, the efficiency and precision of calibration of model parameters are the two main indicators to evaluate model performance and also are difficulties faced by the current runoff forecasting research. Meanwhile, with the increasing number of cascaded reservoirs, the elaborate operation has become more significant for cascaded reservoirs. In addition, the impact of delayed inflow on reservoir operations should not be neglected for long distance watercourses existing between the first upstream reservoir and the most downstream reservoir. Therefore, it is extremely important to research on the improved hydrological inflow forecast methods, effective optimal operation models and rational evaluation about delayed energy to enhance hydropower system management. In this paper, the reservoirs on southern China are selected as the engineering backgrounds. The artificial intelligence technologies with high forecast precision have been studied for mid- and long-term runoff forecasting. Meanwhile, the parallel intelligence algorithm for reservoir operation and the problem of delayed inflow in mid-term reservoir operation have already been studied. Major achievements are listed as follows:(1) The feedforward neural networks usually use traditional gradient-based training algorithms to calibrate model parameters for runoff forecasting, which are often time-consuming and may easily converge to local minimum. In this paper, a conjunction model of wavelet neural networks with extreme learning machine (WNN-ELM) is presented for 1-month ahead discharge forecasting. Due to the powerful mathematical analytical ability of wavelet analysis, the a trous wavelet transform is used to decompose the original discharge time series into several sub-series. The sub-series are then as inputs for single-hidden layer feedforward neural networks (SLFNs) and the output is the next observed discharge. To make full use of the merits of ELM with fast learning speed and good generalization capability, the ELM algorithm is employed to train WNN model. Monthly discharge time series data from two reservoirs in southwestern China are derived for validating the models. For comparison, the SLFNs-ELM and support vector machine (SVM) are also employed. The results indicate that the SLFNs-ELM performs slightly better than the SVM for peak discharge estimation, and the proposed model WNN-ELM provides more accurate forecast precision than SLFNs-ELM and SVM.(2) Due to linear regression algorithm adopted by standard echo state network (ESN) to calibrate model parameters for runoff forecasting, the over-fitting phenomenon may often occurs. To overcome this drawback, a Bayesian echo state network (BESN) model is proposed for daily rainfall-runoff forecasting. The BESN model combined Bayesian theory and ESN obtains the optimal output weights via maximizing posterior probabilistic density and improves its generalization ability. Two Case studies on daily inflow forecasting for Ansha Reservoir and Xinfengjiang Reservoir show that the BESN model is effective and feasible. which can provide better forecast accuracy than the traditional BP neural network and ESN models.(3) The traditional differential evolution (DE) may easily fall into local optimum, when it is used to solve optimization problems for cascaded reservoirs. In this paper, to overcome this disadvantage and enhance computational efficiency, an algorithm named multicore parallel hybrid differential evolution (PCSADE) is proposed to solve long-term operation of cascaded reservoirs. Firstly, to make full use of the merits of strong randomness and ergodicity of chaos theory, the tent map is used to generate the initial population and dynamically adjust the control parameters of DE algorithm. Secondly, the SA algorithm is utilized to replace the selection operation of DE algorithm by Metropolis rule. Finally, the improved algorithm (CSADE) also utilizes the Fork/Join parallel framework to implement parallel computation, of which the performance is tested with different population size. The results show that CSADE has good global search ability and PCSADE can improve the solution quality and efficiency significantly.(4) For the cascaded reservoirs on mid-term operation, long distance existing between the first upstream reservoir and the most downstream reservoir leads to the problem of delayed inflow. In this paper, a storage energy maximization model considering delayed energy is presented and solved by Lagrange relaxation method based on two stage subgradient algorithm to update multipliers. Firstly, the progressive optimality algorithm is employed to solve maximum power generation to determine the total charge process of the operation horizon. Secondly, the storage energy maximization model at the end of operating horizon is built and solved by Lagrange relaxation method according to systemic load demand Meanwhile, the dynamic programming successive approximations algorithm is utilized to solve the dual problem and the multipliers are updated based on two stage subgradient algorithm. A case study of the mid-term optimal operation of cascaded reservoirs on the mainstream of Lancang River middle and lower reaches has been conducted. The comparsion between the proposed model and the model that delayed energy is not considered is carried out. The results show that the mid-term delayed energy impacts the optimal operation plan to some degree, and the effect of which need to be taken into consideration when making the mid-term operation schedules.