基于BP神经网络的铀尾矿砂氡射气系数预测

以中国南方某铀尾矿砂为研究对象,在实测150组不同环境温度、湿度和尾矿砂粒径下的射气系数数据的基础上,基于BP神经网络预测理论,将环境温度、湿度和尾矿砂粒径作为BP神经网络的输入元,射气系数作为输出元,建立了颗粒堆积型介质射气系数的BP神经网络预测模型。将130组实测数据作为预测模型的训练样本,经过7 974次训练后精度满足要求,训练后的预测模型所得射气系数预测值与实测值的最大相对误差为2.68%,利用颗粒堆积型介质射气系数的BP神经网络预测模型得到的预测值与实测值吻合较好,预测模型可用于分析环境温度、湿度和介质粒径对颗粒堆积型介质射气系数的影响规律。

Forecast for Radon Emanation Coefficient of Uranium Tailings Based on BP Neural Network

Taking a uranium tailings sand in southern China as the research object, under the measurement of 150 sets of emanation coefficient data with different environmental temperatures, humidity, and particle sizes of tailings and based on the BP neural network prediction theory, the environmental temperature, humidity, and particle size of the tailings sand are taken as the input elements of the BP neural network and the emanation coefficient is taken as the output element, the BP neural network prediction model of particle-packing media emanation coefficient is established. The 130 sets of measured data were used as training samples for the prediction model, the accuracy meets the requirement after 7 974 trainings. The maximum relative error between the predictive value and the measured value of the emanation coefficient obtained after training is 2.68%. The predicted values obtained by the BP neural network prediction model for the emanation coefficient of the particle-packing media were found to be in good agreement with the measured values, and the values can be used to analyze the influence rule of environmental temperature, humidity and media particle size on the emanation coefficient of particle-packing media.