An application of ANN in scheduling pumped-storage

Abstract

An artificial neural network (ANN) based optimization method in scheduling pumped-storage is proposed in the paper. Short-term scheduling as well as real-time dispatch of a pumped-storage station is a constrained optimization problem. It becomes more complicated when coordinated with other generation resources. The computation time is often long and the operation conditions may change unpredictably. A fast and practical way is expected. The ANN is used as a signal processing device, which represents mapping functions from input space to output space. Through a training process, multi-layered feedforward and neural networks can be used to approximate the continuous functions with a given accuracy and real-time solution can be achieved. In this paper three layer feedforward ANN and improved BP algorithm are adopted to solve the problem of pumped-storage scheduling. A set of ANN training data are obtained by running an optimization software. The paper describes how to select and organize the input data and how to train the ANN. A work example is presented and a comparison with traditional method is made. It shows that a fast and accurate solution for pumped-storage scheduling can be achieved with ANN.