【摘要】 提出了一种利用改进粒子群算法优化广义神经网络的平滑因子,并采用优化后的网络预测系统边际价格的方法,该方法克服了利用梯度下降法优化平滑因子时易陷入局部极值点以及利用遗传算法优化平滑因子时收敛速度慢等缺点。采用该方法利用美国加州电力市场公布的历史数据进行系统边际价格预测,结果表明本文提出的方法比传统的BP网络预测方法更有效。更多还原

【Abstract】 A kind of smoothing factor,which optimizes general regression neural network (GRNN) by improved particle swarm optimization (PSO),is put forward and a method to forecast system marginal price by GRNN with optimized parameters is proposed.In the smoothing factor optimization the proposed method overcomes the defects caused by other methods,e.g.,easy to fall into local extreme points while gradient descent method is used as well as slow convergence speed while genetic algorithm is adopted.Results of case study,in which the historical data published by California electricity market is taken for example,show that comparing with traditional forecasting method based on BP neural network the proposed method is more effective and accurate.更多还原