摘要
潮位预测对于潮汐电站日常运行、优化调度有着非常重要的作用。针对潮位受非周期性因素影响而具有非平稳性特点以及传统人工神经网络对潮位预测存在训练速度慢、收敛精度低、易陷入局部最优等缺陷,提出了一种改进实用的遗传神经网络潮位预测算法,首先通过潮位数据的异常值检测,采用均值替换法克服由观测记录产生的数据误差;然后通过神经网络拓扑结构合理设计、节点优选、遗传算法优化网络权值与阈值等措施,建立了潮位预测模型。通过实际港口潮位预测应用验证了算法的有效性。
For tidal power station, the tidal forecasting plays an important role in the daily operation and optimal scheduling. The paper proposes an improved genetic neural network prediction algorithm to forecast tidal level to tackle the non-stationary characteristics of tides caused by non-cyclical factors and the defects in traditional artificial neural network： such as slow training, low precision and proneness to local optimum. First, the abnormal value in tidal data was detected and the mean substitution method was adopted to overcome data error generated by the ob- servation records. The tidal forecasting model was built after the reasonable design of neural network topology, op- timal node selection, and setting of weights and thresholds to optimize network in genetic algorithms. The applica- tion of the algorithm to actual port tide forecasting demonstrated its effectiveness.
出处
《电力科学与工程》
2015年第8期1-7,共7页
Electric Power Science and Engineering
基金
国家自然科学基金(61273144)
北京市自然科学基金(4122071)
关键词
神经网络
遗传算法
潮位预测
潮汐发电
neural network
genetic algorithm
tidal forecasting
tidal power generation
