优化相空间近邻点与递归神经网络融合的短期负荷预测

根据在相空间重构拓扑近邻点的时间演化原理，提出了优化近邻点(optimal neighbor points，ONP)的短期负荷预测(Short-term load forecasting，STLF)法，它可克服伪近邻点在高嵌入维对局域动力学估计的不利影响，以提高预测精度。在此基础上，又提出ONP与递归性时延神经网络(Tune Delay Neural Network，TDNN)模型融合的STLF法，具有动态性能的TDNN是按优化近邻相点的演化轨迹构造，是属于对预测点跟踪的智能辩识动态行为模型。它能增强模型对系统动力学的联想性和泛化能力，使预测精度提高一倍以上。该文经两类不同负荷系统周、日预测仿真测试，证实所研究的预测模型能有效、稳定地提高预测精度，且有高的适应能力，为基于相空间理论预测法用于实际取得有效的进展。

A NEW STLF APPROACH BASED ON THE FUSION OF OPTIMAL NEIGHBOR POINTS IN PHASE SPACE AND THE RECURSIVE NEURAL NETWORK

Based on the time evolution principle of the topological neighbors in the PSR, a Short-Term Load Forecasting (STLF) approach on selecting the optimal neighbor points (ONP) is presented in this paper. It can eliminate some false neighbor points through identifying exponential separating rate of time evolutional trajectory, and can improve the precision of STLF. Further the STLF approach is proposed based on the fusion of ONP and TDNN (time-delay neural network) model with recursiveness. The TDNN model with the performance of intelligent dynamic identifying is constructed by following the time evolutionary track of forecast point, and it can enhance associative memory and generalization ability of forecasting model, so that the precision can be improved further. Two kinds of load systems are used to simulate, and the testing results prove the proposed model and approach can improve effectively and stably the precision of STLF and possess a good adaptability in the weekday and weekend. This research acquires the effective progression and practical significance in the prediction engineering.