改进的QPSO-BP算法的铀价格预测模型及应用

铀产品价格的变化直接决定了铀矿项目的价值，铀产品价格的预测，可提高企业的经营决策能力和抗风险能力。为提高预测的精度，采用基于改进的量子粒子群算法优化训练BP神经网络的学习算法，对铀价格进行建模预测。采用改进的QPSO算法优化BP网络的权值与阈值。将通过优化搜索得到的粒子的位置向量解码作为网络的权值与阈值，选择网络结构5．11—1对铀价格进行预测。结果表明：QPSO—BP模型的预测精度（0．15％）高于PSO-BP模型（4．55％）与BP模型（30．86％）。泛化能力指标平均相对变动值为O．0025，预测结果的泛化能力提高。相对误差分布集中，预测结果稳定。说明该模型在铀价格预测中有效，对项目投资决策有一定的参考价值。

Uranium price forecasting model based on BP improved by QPSO and its application

Changes in the price of uranium products directly determine the value of the uranium project. The uranium price fore- casting can improve business decision-making ability and the ability to resist risks. In order to improve the generalization ability of BP network to predict the price of U3O8, a QPSO-BP model is proposed. This model uses the QPSO to optimize the initial value of weights and thresholds of BP network. The position vector of the individual particle searched in global space is encoded as the best optimized value of weights and thresholds used in the 5-11-1 streamlined structure to predict the price of U308. The experi- ments show that the BP network optimized by QPSO can produce a stable prediction result, and its ARV is 0.0025. QPSO-BP model is more stable with its relative error mainly below 1%. Compared with the PSO-BP and BP prediction models, the general- ization ability is better than the first two models, and the prediction accuracy with a least value （0.151% ）. The result indicates that the QPSO-BP model is effective and can be applied in uranium price forecasting, and also provides some reference value for the policy decision for mining project investment.