基于SSA-MIC-SMBO-ESN的水质预测模型

Water quality prediction model based on SSA-MIC-SMBO-ESN

水污染事件威胁人类生产和生活安全,提前预测水质变化对水污染的快速反应具有重要意义。基于水质数据的时序性,引入泄漏积分型回声状态网络(ESN),以莆田市东圳水库水质监测站的10种水质指标数据作为样本,分别构建DO、COD_(Mn)、TP的水质预测模型。首先,在用邻近点线性趋势法对缺失值进行填充,用Z-score法和邻近点线性趋势法对异常值进行检测修正的基础上,用奇异谱分析(SSA)算法对水质数据进行平滑降噪处理;然后,采用最大互信息系数(MIC)衡量水质指标之间的相关度,选取相关系数较大的水质指标作为待预测水质指标的输入特征;最后,利用ESN构建基于多特征的水质预测模型,其中采用序列模型优化(SMBO)算法对模型的超参数进行优化。试验结果表明,构建的DO、COD_(Mn)和TP的SSA-MIC-SMBO-ESN水质预测模型都具有较高的预测精度,适合实际应用。

Water pollution threatens the safety of human production and life.Predicting the change of water quality in advance is of great significance to the rapid response to water pollution.Based on the temporality of water quality data,the echo state networks(ESN)of leakage integral type was introduced to construct the water quality prediction models of DO,COD_(Mn) and TP with the 10 water quality indicators data of Dongzhen monitoring station in Putian city as samples.Firstly,on the basis of filling the missing value with the Z-score method and adjacent point linear trend method,the water quality data was smoothed and denoised by the singular spectrum analysis algorithm(SSA).Then,the maximum information coefficient(MIC)was used to measure the correlation among water quality indicators,and the water quality indicators with large correlation coefficient were selected as the input characteristics for the water quality indicators to be predicted.Finally,the water quality prediction model based on multi-characteristics was constructed by ESN,in which the super parameters of the model were optimized by sequential model optimization algorithm(SMBO).The experimental results showed that,the SSA-MIC-SMBO-ESN prediction models of DO,COD_(Mn) and TP had high prediction accuracy and could be applied in practice.