
基于HS算法优化的EEMD-RNN混凝土坝位移预测模型
范鹏飞, 祝福源
基于HS算法优化的EEMD-RNN混凝土坝位移预测模型
Displacement Prediction Model of EEMD-RNN Concrete Dam Based on HS Algorithm Optimization
由于混凝土坝位移的监测过程中存在噪声误差,使得智能算法预测位移时容易出现过拟合、局部最小值及收敛速度慢等缺点。为了解决上述存在的问题,首先采用集合经验模态分解法(EEMD)对位移实测值进行分解,得到不同频次的分量IMF和剩余量R,然后通过递归神经网络(RNN)对各分量和剩余项进行训练,得到环境量和时效与IMF及R的映射关系,最后通过模型得到的映射关系可求各分量和剩余项的响应预测值,并进行等权求和可得位移的预测值。在RNN训练过程中引入和声搜索算法(HS)对其进行优化,系统性去噪,以一定的概率限差函数进行扰动以得到最优解,从而优化了RNN的权重及阈值,提高了模型的健壮性。以某混凝土坝为计算案例,结果表明,HS-EEMD-RNN模型的整体拟合和预测的精度较高、过拟合程度小,与EEMD-RNN模型对比可知,HS算法可以明显提高其模型精度,并且对位移过程线的突跳值预测的精度高。
Due to the noise error in the monitoring process of concrete dam displacement, the intelligent algorithm is prone to overall fitting, local minimum value and slow convergence speed. In order to solve the above problems, this paper uses the ensemble empirical mode decomposition (EEMD) method to decompose the measured displacement values to obtain the components IMF and residual R with different frequencies. Then, the components and residual terms are trained by recurrent neural network (RNN) to obtain the mapping relationship between environmental variables and time effect and IMF and R. Finally, the predicted response values of each component and residual term can be obtained by the mapping relation obtained from the model, and the predicted value of displacement can be obtained by summation with equal weight. In the process of RNN training, harmony search algorithm (HS) is introduced to optimize the RNN, systematically de-noising, and a certain probability limit function is used to disturb to get the optimal solution. Thus, the weights and thresholds of RNN are optimized and the robustness of the model is improved. Taking a concrete dam as an example, the results show that HS-EEMD-RNN model has higher accuracy of overall fitting and prediction, and less overfitting degree. Compared with EEMD-RNN model, HS algorithm can significantly improve the accuracy of its model, and has high precision in predicting the sudden jump value of displacement hydrograph.
位移预测 / 混凝土坝 / 集合经验模态分解 / 和声搜索法 / 递归神经网络 {{custom_keyword}} /
displacement prediction / concrete dam / ensemble empirical mode decomposition (EEMD) / harmony search method (HS) / recurrent neural network (RNN) {{custom_keyword}} /
表1 两种模型的拟合精度对比Tab.1 Comparison of fitting accuracy of two models |
模型 | R | MAE/mm | RMSE/mm | MAPE |
---|---|---|---|---|
EEMD-RNN | 0.986 5 | 0.197 | 0.207 | 0.107 |
HS-EEMD-RNN | 0.991 8 | 0.074 | 0.086 | 0.041 |
图6 EEMD-RNN模型和HS-EEMD-RNN模型对总位移的预测Fig.6 Prediction of total displacement by EEMD-RNN model and HS-EEMD-RNN model |
表2 EEMD-RNN和HS-EEMD-RNN模型预测值误差对比Tab.2 Comparison of prediction error between EEMD-RNN model and HS-EEMD-RNN model |
模型 | 平均绝对误差/mm | 均方根误差 /mm | 平均绝对 百分比误差 |
---|---|---|---|
EEMD-RNN | 0.221 | 0.250 | 0.123 |
HS-EEMD-RNN | 0.088 | 0.101 | 0.048 |
表3 位移突跳点预测值对比Tab.3 Comparison of predicted values of displacement jump points |
日期 | 位移实测值/ mm | 位移预测值/mm | 绝对百分比误差/% | ||
---|---|---|---|---|---|
HS-EEMD-RNN | EEMD-RNN | HS-EEMD-RNN | EEMD-RNN | ||
2017-07-07 | -1.353 | -1.348 28 | -1.408 28 | 0.34 | 4.10 |
2017-07-09 | -1.280 | -1.385 30 | -1.585 30 | 8.20 | 23.82 |
2017-07-16 | -1.381 | -1.406 81 | -1.466 81 | 1.87 | 6.22 |
2017-07-24 | -2.516 | -2.610 65 | -2.310 65 | 3.76 | 8.17 |
2017-07-28 | -3.090 | -3.054 94 | -2.614 94 | 1.12 | 15.36 |
2017-07-30 | -2.907 | -2.801 51 | -2.651 51 | 3.63 | 8.79 |
2017-08-03 | -0.941 | -1.098 20 | -1.338 20 | 16.71 | 42.22 |
2017-08-10 | -3.176 | -3.105 14 | -2.985 14 | 2.24 | 6.02 |
2017-08-12 | -2.439 | -2.583 02 | -2.683 02 | 5.91 | 10.01 |
2017-08-19 | -1.868 | -1.970 15 | -2.320 15 | 5.48 | 24.22 |
平均值 | - | - | - | 4.93 | 14.89 |
1 |
吴中如. 水工建筑物安全监控理论及其应用[M].北京:高等教育出版社,2003:2-4.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
2 |
周健,仲静文,郝利朋,等.基于马尔科夫链残差修正的EMD-MELM混凝土坝位移预测模型[J].水利水电技术,2020,51(6):40-48.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
3 |
彭圣军. 混凝土坝安全监控模型数值优化及变位预警指标研究[D].南昌:南昌大学,2014.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
4 |
顾冲时,吴中如.混凝土坝空间位移场的确定性模型反演分析法[J].河海大学学报,1994(2):82-87.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
5 |
吴中如. 混凝土坝安全监控的确定性模型及混合模型[J]. 水利学报, 1989(5):64-70.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
6 |
徐朗,蔡德所.改进的PSO-SVM模型在大坝变形预测中的应用[J].中国农村水利水电,2018(3):120-123,128.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
7 |
郭志扬,王建,黄庆.基于卡尔曼滤波的GA-BP模型在大坝变形预测中的应用[J].中国农村水利水电,2016(12):113-116.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
8 |
何启,戴波.基于灰色神经网络-加权马尔可夫链的大坝变形监控模型及预报研究[J].中国农村水利水电,2016(10):146-150,155.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
9 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
10 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
11 |
刘淑琼,高鑫,李长春.EEMD-RBF神经网络的电离层TEC预报模型[J].测绘工程,2020,29(3):15-19.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
12 |
李华,刘韬,伍星,等.EEMD和优化的频带熵应用于轴承故障特征提取[J].振动工程学报,2020,33(2):414-423.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
13 |
谭冬梅,刘晓飞,姚欢,等.基于EEMD-JADE的桥梁挠度监测中温度效应的分离[J].土木与环境工程学报(中英文),2020,42(3):90-99.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
14 |
周暄焯. 基于RNN及其融合方法的DNA甲基化预测模型研究[D].成都:电子科技大学,2020.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
15 |
芦效峰,张胜飞,伊胜伟. 基于CNN和RNN的自由文本击键模式持续身份认证[J].清华大学学报(自然科学版), 2018,58(12): 1 072-1 078.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
16 |
范竣翔,李琦,朱亚杰,等.基于RNN的空气污染时空预报模型研究[J].测绘科学,2017,42(7):76-83,120.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
17 |
张磊,丁香乾,宫会丽,等.改进和声搜索算法的近红外光谱特征变量选择[J].光谱学与光谱分析,2020,40(6):1 869-1 875.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
18 |
李明,胡江平.基于和声搜索算法的无线传感器网络多重连通覆盖[J].传感技术学报,2020,33(2):272-278.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
19 |
王志刚,郭钰卓,周启昕,等.基于和声搜索算法的RGV动态调度优化模型[J].海南大学学报(自然科学版),2019,37(2):112-118.
{{custom_citation.content}}
{{custom_citation.annotation}}
|
20 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
21 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
22 |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
23 |
DAS S,
{{custom_citation.content}}
{{custom_citation.annotation}}
|
{{custom_ref.label}} |
{{custom_citation.content}}
{{custom_citation.annotation}}
|
/
〈 |
|
〉 |