基于网格气象数据的LLE-LSTM小水电发电功率超短期预测

Ultra-short-term Prediction of Small Hydropower Generation Power Based on Grid Meteorological Data and LLE-LSTM

考虑小水电区域内降雨量时间分布及空间分布的影响,以网格气象数据为基础,提出LLE-LSTM小水电发电功率超短期预测方法。首先利用相关性分析筛选空间上与小水电发电功率相关的网格区域,而后引入局部线性嵌入算法(LLE)对网格降雨数据进一步降维;最后将其作为输入代入到长短记忆神经网络(LSTM)训练,构建小水电发电功率的超短期预测模型。利用所提方法对湖北省某地区小水电发电功率数据进行仿真检验,结果表明该方法能够较好表征区域内降雨量空间分布对小水电发电功率的影响,且有效避免了因网格气象数据维度过高导致的过拟合问题,显著提高了小水电发电功率超短期预测精度。

Considering the influence of time distribution and spatial distribution of rainfall in small hydropower areas, ultra-short-term LLE-LSTM prediction method of generation power was proposed based grid meteorological data. Firstly, correlation analysis was used to select grid area related to the generation power in space. Then local linear embedding(LLE) was introduce to reduce dimensionality of the temporal and spatial distribution of rainfall in the area. Finally, the dimensionality reduction processed data was used as input to training the long short-term memory(LSTM), and an ultra-short-term prediction model of small hydropower generation power was established. The proposed method was used to simulate the power data of small hydropower generation in a certain area of Hubei Province. The results show that this method can better characterize the influence of the spatial distribution of rainfall in the region on the power generation of small hydropower generation, and effectively avoid the over-fitting problem caused by the excessively high dimension of grid meteorological data as well as greatly improve the prediction accuracy of small hydropower generation power.