Ｂ Ｐ 网络是目前应用最为广泛的神经网络，但由于 Ｂ Ｐ 网络采用的是梯度下降法，这就不可避免地会出现网络学习收敛速度慢及容易陷入局部极小等问题。此外，学习因子和惯性因子选取对网络的收敛有较大的影响，但它们只能凭 经验确定。因此， Ｂ Ｐ 网络的有 效应用受到 了一定的限 制。针对 Ｂ Ｐ网络的学习收敛速度慢这一主要缺陷，对改进激励函数、改进 误差函数、改进一般化误差、学 习因子和惯性因子的自适应调整、梯度下降法与直接搜索法相结合、全局 优化、非线性优化、拓仆修正算 法等多种改进方案按改进原理进行了分类综述，并在此基础上，通过解决 Ｘ Ｏ Ｒ 问题的仿真实验对部分改进方案进行实验性评价，分析说明了它们的优劣和特点。 B P network is currently the most widely used neural network, but because B P network uses the gradient descent method, it inevitably brings about such problems as the slow convergence rate of network learning and easy to fall into local minima. In addition, the selection of learning factors and inertia factors has a greater impact on the convergence of the network, but they can only be determined empirically. Therefore, the effective application of B P network has been limited. Aiming at the main disadvantage of slow learning convergence in B P networks, the optimization of the excitation function, the improvement of the error function, the improvement of the generalized error, the adaptive adjustment of the learning factor and the inertia factor, the combination of the gradient descent method and the direct search method, Nonlinear Optimization, Topo Correction Algorithm and many other improvement programs are classified according to the principle of improvement. On the basis of this, some improved solutions are experimentally evaluated by solving the XO R problem, and the results show that their Pros and cons and characteristics.