激励函数可调的混沌神经网络短路负荷预测

Short Term Load Forecasting by Using Neural Networks with Variable Activation Functions and Embedded Chaos Algorithm

为提高BP神经网络的收敛速度并避免陷入局部极小,提出了一种多参数可调激活函数的人工神经网络,推导出相应的BP学习算法并将改进的BP算法和混沌优化算法相结合,使混合算法的学习向全局最优方向进行以不至于陷入局部极小。在非线性函数仿真和短期负荷预测的研究中,该算法和传统BP算法的对比试验显示,改进算法的收敛速度更快、预测精度更高。

To accelerate the convergence speed of the BP neutral network and avoid to getting into the local minimum, a novel variant activation (transform) sigmoid function with three parameters is proposed in this paper, then the improved BP learning algorithm based on it is deduced, furthermore, we combined the new fast BP algorithm and chaos optimization algorithm, the learning of the new algorithm converges fast and globally with no local minimum. The faster convergence speed and higher prediction precision of the improved algorithm compared with the traditional BP ANNs is proved by the results of simulation tests of nonlinear function and the prediction results of short-term load.