论文部分内容阅读
提出一种BP网络的全局优化算法(GOA),在监测环境变化的同时调整整个网络使得网络能够保持较为准确的输出.该算法分为3个子算法——输入属性约简算法RAIA、隐藏层优化算法HLOA、神经网络监测算法NNSA.其中输入属性约简算法对神经网络的输入层进行优化以剔除对输出影响不大的输入,隐藏层优化算法在满足最小精度要求的前提下对神经网络隐藏层进行剪除或添加使其更加高效,神经网络监测子算法能够实时监测整个神经网络的运行状态并随时启动全局优化算法以适应当前环境的变化.该算法能够以较小的开销实时地改变整个神经网络结构,使得整个网络随着环境的稳定而稳定下来直到下次环境发生变化再次对网络进行优化.
A global optimization algorithm (GOA) for BP network is proposed, which adjusts the entire network while monitoring the environment changes so that the network can maintain a more accurate output.The algorithm is divided into three sub-algorithms: input attribute reduction algorithm RAIA, hidden layer optimization Algorithm HLOA, neural network monitoring algorithm NNSA, in which the input attribute reduction algorithm optimizes the input layer of the neural network to eliminate the input that has little effect on the output. The hidden layer optimization algorithm, on the premise of meeting the minimum precision requirements, To cut or add to make it more efficient, the neural network monitoring sub-algorithm can monitor the running status of the whole neural network in real time and start the global optimization algorithm to adapt to the current environment change at any time.The algorithm can change the whole neural network with less overhead in real time Structure, making the entire network stable with the environment stabilized until the next environment changes again to optimize the network.