基于相空间重构理论和优化递归神经网络结合的短期负荷预测方法

根据电力负荷序列的混沌特性,提出以相空间重构理论和优化递归神经网络结合的电力系统短期负荷预测方法,以相空间重构理论确定递归神经网络输入维数;训练样本集由对应预测相点的最近邻相点集构成,并按预测相点步进动态相轨迹生成;优化递归神经网络是以双重遗传算法来确定递归神经网络的隐层结构和权值,总体寻优性可抑制伪近邻点的影响,保证提高预测精度及其稳定性。对两类不同负荷系统日、周预测仿真测试,证实其比传统神经网络预测模型能有效地提高预测精度0.8%。因此,所研究的预测模型和方法在实际预测领域有较高的实用价值。

A new approach of STLF based on combination of phase space reconstruction theory and optimal recursive neural networks

A new approach of STLF in power systems using PSRT (phase space reconstruction theory) combined with ORNN (optimal recursive neural network) is presented in this paper according to the chaotic character of the load series. The input dimension of ORNN is decided by PSRT. The training samples are formed by the nearest neighbor phase points of forecasting phase point, and they are generated by means of the stepping dynamic phase track of the forecasting phase point. The structure and weighs of ORNN are trained by the DGA (dual-genetic algorithm). So it can enhance associative memory and generalization ability to chaotic dynamics of forecasting model, and it can restrain the disturbance of the false neighbor points. It has global searching optimum ability and can improve effectively and stably the precision of STLF. Two kinds of load systems are used to simulate, and the test results show the proposed model can improve the 0.8% forecasting precision and possesses a good adaptability in the weekday and weekend. This research acquires the effective progression in the practical prediction engineering using the STLF model based on the combination of PSRT and ORNN.