A variety of wastewaters can be generated in the coal chemical industry, and their treatment processes are complicated and have difficulty meeting standards. Using wastewater to prepare coal water slurry is an efficient and convenient new approach. The concentration of coal wastewater slurry is related to the content of the main wastewater components. A backpropagation neural network is developed to predict the maximum slurry concentration and analyze the mechanism at the data level according to the main component indicators, and a particle swarm algorithm is used to improve the neural network. The results are as follows: (a) it is feasible to predict the maximum concentration of coal wastewater slurry by a neural network, and a particle swarm algorithm can effectively improve the prediction accuracy in different models, reducing mean absolute error by up to 0.44%; (b) different input factors have different impacts on model prediction results—organic matter, ammonia nitrogen, and monovalent metal ions content as input factors to predict the maximum slurry concentration can get the most accurate results, obtaining a mean absolute error of 0.16% for the optimized backpropagation neural network and the lowest mean square error; and (c) divalent metal ions and phenols content are not suitable as input factors for predicting, as they all cause an increase in model error due to their weak or complex effects on the slurryability.

中文翻译：

---

基于废水成分数据的优化神经网络预测煤废水泥浆的最大固体浓度

煤化工会产生多种废水，处理工艺复杂，达标难度大。利用废水制备水煤浆是一种高效便捷的新方法。煤废水浆液浓度与废水主要成分的含量有关。开发了反向传播神经网络，根据主要成分指标在数据层面预测最大矿浆浓度并分析其机理，并采用粒子群算法对神经网络进行改进。结果如下：（a）用神经网络预测煤废水浆液的最大浓度是可行的，粒子群算法可以有效提高不同模型下的预测精度，将平均绝对误差降低多达 0.44%；(b) 不同的输入因素对模型预测结果的影响不同——有机物、氨氮、单价金属离子含量作为输入因素预测最大浆液浓度可以得到最准确的结果，平均绝对误差为0.16%对于优化的反向传播神经网络和最低的均方误差；(c) 二价金属离子和酚类含量不适合作为预测的输入因素，因为它们对可浆化性的影响较弱或复杂，都会导致模型误差增加。将单价金属离子含量作为输入因素预测最大浆液浓度可以得到最准确的结果，优化后的反向传播神经网络的平均绝对误差为0.16%，均方误差最小；(c) 二价金属离子和酚类含量不适合作为预测的输入因素，因为它们对可浆化性的影响较弱或复杂，都会导致模型误差增加。将单价金属离子含量作为输入因素预测最大浆液浓度可以得到最准确的结果，优化后的反向传播神经网络的平均绝对误差为0.16%，均方误差最小；(c) 二价金属离子和酚类含量不适合作为预测的输入因素，因为它们对可浆化性的影响较弱或复杂，都会导致模型误差增加。