一种基于小波分析和Elman动态神经网络的中长期电力负荷预测方法

A Mid-term and Long-term Electricity Load Prediction Method Based on Wavelet Analysis and Elman Dynamic Neural Network

针对传统静态神经网络自适应能力差、收敛速度慢、预测精度低的问题,提出了一种基于小波分析和Elman动态神经网络的中长期电力负荷预测方法,该算法通过对原始样本进行小波分解,将分解后的低频趋势信号和高频细节信号分别进行预测,在输出端再进行重构后得到预测曲线;然后就传统负荷预测问题中数据预处理环节的数据校验问题,提出了一种基于小波理论的奇异点检测法,该方法对原始样本进行一维离散小波分解,抽取一层高频细节信号进行分析,根据工程实践中设置的阈值,来检测有可能因为系统故障、人为失误导致的数据记录错误,为准确预测提供了保障。

Due to the poor adaptability, slow convergence rate and low prediction accuracy of the traditional static neural network, a mid-teim and long-term electricity load prediction method has been put forward based on wavelet analysis and Elman dynamic neural network. Through wavelet analysis of primary samples, the prediction curve is created after decomposed low- frequency tendency signal and high-frequency detail signal are predicted separately and reconstructed at the output terminal. As to data check in data pretreatment step of the traditional load prediction, a singular point detecting method based on wavelet theory is suggested. Using this method, a layer of high-frequency detail signal is analyzed after one-dimensional discrete wavelet decomposition of the primary samples. Then, according to the threshold set in the project, the data record mistakes resulted from system faults and human errors can be detected, so as to provide guarantee for accurate predictions.