基于CNN-LSTM神经网络的电网调度火电厂短期存煤预测

采用传统的回归拟合进行发电厂存煤量单变量单步预测已无法满足电网优化调度的需求,针对该问题,提出卷积神经网络(CNN)与长短期记忆(LSTM)神经网络相结合的CNN-LSTM神经网络预测模型,利用CNN良好的特征提取能力以及LSTM神经网络特殊的记忆预测功能实现对未来电厂存煤量的精准预测。为了使预测结果更符合实际存煤量,在已有预测结果的基础上进行进一步优化。实例验证结果表明,相较于传统差分自回归移动平均(ARIMA)模型和单一LSTM神经网络模型,所提模型取得的效果更好,且经过优化后的预测精度得到了进一步提高。

基于奇异谱分析与长短时记忆神经网络的电厂存煤量短期预测

针对火力发电厂存煤量预测精度不高的问题,提出了一种基于奇异谱分析(SSA)和长短时记忆(LSTM)神经网络的多变量多步预测模型.考虑电厂历史数据和温度的影响,以奇异谱分析或小波分析(wavelet analysis)对数据做平滑处理,剔除数据中的噪声成分,作为LSTM神经网络的输入向量进行仿真测试.结果表明,相比于普通BP(back propagation)神经网络、循环神经网络(RNN)和小波分析与LSTM结合算法,本文采用的基于奇异谱分析和长短时记忆神经网络的多变量多步预测模型精度更高.

Flue Gas CO_2 Emission Reduction Technologies and Applications

The current status and trend of CO2 emission from coal-fired power plants in China are introduced. Main flue gas decarbonization technologies and their prospective of applications in China are discussed in two separate parts-capture and sequestration. It is stated that the selection of CO2 capture and sequestration technologies relates closely with the geographical location of power plants, with the destination of CO2 being the key. Further, it is suggested that industrialized test centers or test platforms of national or industrial level should be set up.