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

Short-term Prediction of Coal Stock in Power Plant Based on Singular Spectrum Analysis and Long Short-term Memory Neural Network

摘要: 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.

Abstract: In view of the low accuracy of coal storage predict, ion in thermal power plants, we propose a multi-variable and multi-step prediction model based on singular spectrum analysis (SSA) and long short-term memory (LSTM) neural network. Considering the influence of historical data and temperature of power plant, we use singular spectrum analysis or wavelet analysis to smooth the data, remove the noise components from the data, and use it as the input vectors of LSTM neural network for simulation test. The results show that compared with the combination of back propagation (BP) neural network, recurrent neural network, wavelet analysis and LSTM, the multi variable and multi-step prediction model based on singular spectrum analysis and long short time memory neural network adopted has higher accuracy.