为使组合预测的组合权重随样本数据不同而改变,从而使误差平方和达到最小,我们使用二次规划对样本每期组合权重进行确定.利用所得权重样本对广义神经网络训练使其对样本外组合权重进行确定.结果表明:这是一种可行得组合预测方法,能够提高预测精度,为变权组合预测增添了一种可行方法. In order to minimize the sum of squared errors, we use the quadratic programming to determine the combined weight of each sample period.Using the weighted sample to train the generalized neural network to the outside of the sample The results show that this is a feasible combined forecasting method, which can improve the forecasting accuracy and add a feasible method for the changeable portfolio forecast.