年最大洪峰流量群居蜘蛛优化投影寻踪预测模型

年最大洪峰流量预测,受较多的复杂因素的影响,不确定性较强,用常规统计方法做出准确预报具有较大困难。从水文序列本身出发,提出将投影回归模型应用于年最大洪峰流量预测,为了更好获得投影寻踪模型参数和预测精度,提出了运用延迟相关系数法确定回归预测因子、群居蜘蛛算法优化投影寻踪模型最佳投影方向参数a、利用最小二乘法确定多项式的权系数c、岭函数个数M的群居蜘蛛优化投影寻踪年最大洪峰流量预测模型,结合长江宜昌站(1882年-2004年)的年最大洪峰流量资料进行实例预测,训练阶段平均绝对相对误差为8.61%,预测阶段平均绝对相对误差为10.51%,该模型预测效果较好,模型结果稳定,可有效应用于年最大洪峰流量预测。

Social spider optimization-based projection pursuit model to predict annual maximum flood peak flow

The prediction of annual maximum flood peak flow is affected by many complicated factors and highly indeterminate, making it difficult to deliver accurate forecasts by conventional statistical methods. In this paper, based on hydrological sequence itself, we proposed that the projection regression model be used to predict the annual maximum flood peak flow. In order to obtain the optimal parameters for the projection pursuit model and improve prediction accuracy, we proposed the hybrid intelligence-based prediction model for annual maximum flood peak flow, in which the delay correlation coefficient method was used to determine the regression prediction factor, the social spider optimization algorithm to optimize the optimal projection direction parameter a of the projection pursuit model, the least square method to determine the weight coefficient c of the polynomial, and qualified rate to control the number of parameter M. The annual maximum flood peak flow data of Yichang Gauging Station of Yangtze River (1882 - 2004) were used to test the proposed model. The results indicated that the model can obtain very good prediction results, with a mean absolute relative error of 8.61% in the training phase, and a mean absolute relative error of 10.51% in the testing phase. The model can produce stable results and can be effectively applied to the prediction of annual maximum flood peak flow.