Forecast Method of Soil Moisture Based on Improved BP Neural Network and Support Vector Machine
-
摘要: 为提高土壤墒情预测精度,提出了一种基于遗传算法(GA)、改进粒子群算法(IPSO)、误差反向传播(BP)神经网络和支持向量机(SVM)的土壤墒情组合预测模型(GA_IPSO_BP-SVM)。该模型首先在BP神经网络的权阈值选择中同时引入GA和IPSO构成GA_IPSO_BP模型,然后对GA_IPSO_BP和SVM模型分别进行训练和数据仿真,最后利用建立的加权模型对GA_IPSO_BP和SVM模型的土壤墒情预测结果进行组合。以安庆市8个监测站某时段内农田土壤墒情数据为例,分别按隔日、两日后和三日后三种时间跨度进行土壤墒情预测,并对照BP、GA-BP、PSO-BP、IPSO-BP、GA_IPSO_BP和SVM模型,验证和比较提出的GA_IPSO_BP-SVM模型的土壤墒情预测精度。结果表明,GA_IPSO_BP-SVM模型的土壤含水量预测相对误差平均值最小。GA_IPSO_BP与SVM模型组合的GA_IPSO_BP-SVM模型提高了土壤墒情的预测精度,更适合于土壤墒情的短期预测,该方法可为农业节水灌溉方案的制定提供技术支撑。Abstract: In order to improve the forecast accuracy of soil moisture, a combined forecasting model GA_IPSO_BP-SVM for soil moisture is proposed based on genetic algorithm (GA), improved particle swarm optimization (IPSO), BP neural network and support vector machine (SVM). The model introduced GA and IPSO into the weight threshold selection of BP neural network to form a GA_IPSO_BP model, and then the GA_IPSO_BP and SVM models were trained and simulated separately. Finally, an established weighted model was used to combine the soil moisture forecast results of the GA_IPSO_BP and SVM models. Taking the farmland soil moisture data of 8 monitoring stations in Anqing city within a certain period as an example, the soil moisture was predicted in three time spans of after one day, after two days and after three days separately, and the forecast accuracy of soil moisture between the proposed GA_IPSO_BP-SVM model and the comparison models BP, GA-BP, PSO-BP, IPSO-BP, GA_IPSO_BP and SVM were verified and compared. The comparison results showed that the average value of relative error of soil water content forecast accuracy of the proposed GA_IPSO_BP-SVM model was the smallest. The GA_IPSO_BP-SVM model based on the combination of GA_IPSO_BP and SVM model improves the forecast accuracy of soil moisture, and is more suitable for short-term forecast of soil moisture. The proposed method could provide technical support for the formulation of water-saving irrigation schemes in agriculture.
-
表 1 不同影响因素组合预测精度对比表
Table 1. Forecast accuracy comparison of different combined influencing factors
影响因素
Influencing factor相对误差平均值(%)
Average value of relative error隔日
After one day两日后
After two days三日后
After three days日平均气温 + 平均土壤湿度 + 降雨量 4.86 5.77 7.53 日平均气温 + 平均土壤湿度 + 土壤含水量 4.55 5.60 6.82 日平均气温 + 降雨量 + 土壤含水量 3.74 4.45 6.04 平均土壤湿度 + 降雨量 + 土壤含水量 4.38 5.30 6.75 日平均气温 + 平均土壤湿度 + 降雨量 + 土壤含水量 3.22 4.06 5.84 表 2 各模型预测精度对比表
Table 2. Forecast accuracy comparison of all used models
预测模型
Forecasting model相对误差平均值(%)
Average value of relative error隔日
After one day两日后
After two days三日后
After three daysBP 5.74 6.18 8.25 GA-BP 4.95 5.17 7.32 PSO-BP 4.88 5.24 7.54 IPSO-BP 4.17 4.91 6.89 GA_IPSO_BP 3.63 4.58 6.37 SVM 3.89 4.83 6.16 GA_IPSO_BP-SVM 3.22 4.06 5.84 -
[1] 邹文安, 马俊英, 孙晓梅. 基于Excel + VBA技术条件的土壤含水量预测模型[J]. 中国防汛抗旱, 2013, 23(2): 49 − 52. doi: 10.3969/j.issn.1673-9264.2013.02.020 [2] 蓝 盈. 基于VI-LST特征空间反演土壤湿度的应用研究[D]. 南京: 南京信息工程大学, 2016. [3] 李新振. 土壤墒情预报预警系统的设计与实现[D]. 哈尔滨: 东北农业大学, 2017. [4] 尚松浩, 雷志栋, 杨诗秀. 冬小麦田间墒情预报的经验模型[J]. 农业工程学报, 2000, 16(5): 31 − 33. doi: 10.3321/j.issn:1002-6819.2000.05.008 [5] 李 波, 王溥泽, 迟道才, 等. 彭曼模型在给定土壤含水率条件下预测玉米腾发量研究[J]. 沈阳农业大学学报, 2014, 45(6): 719 − 25. doi: 10.3969/j.issn.1000-1700.2014.06.014 [6] 蒲胜海, 何新林, 何春燕, 等. 新疆棉花膜下滴灌基于消退指数的土壤水分预报方法研究[J]. 节水灌溉, 2008, (8): 5 − 7 + 11. doi: 10.3969/j.issn.1007-4929.2008.08.002 [7] 董勤各, 许 迪, 章少辉, 等. 基于高精度数值解法的畦灌一维土壤水动力学模型[J]. 灌溉排水学报, 2013, 32(5): 1 − 6. [8] Ashis B, Amalendu G. Regression model for prediction of optimum moisture content and maximum dry unit weight of fine grained soil[J]. International Journal of Geotechnical Engineering, 2011, 5(3): 297 − 305. doi: 10.3328/IJGE.2011.05.03.297-305 [9] Sanghyun K. Time series modeling of soil moisture dynamics on a steep mountainous hillside[J]. Journal of Hydrology, 2016, 536: 37 − 49. doi: 10.1016/j.jhydrol.2016.02.027 [10] Huang P, Su R, Xu W, et al. Multi-source data-based method for retrieval of soil moisture in grassland[J]. The Journal of Engineering, 2019, 19: 5717 − 5721. [11] 李小刚, 王 静, 马友华, 等. 基于BP神经网络的土壤墒情预测精度研究−以肥东县为例[J]. 土壤通报, 2017, 48(2): 292 − 297. [12] 李 宁, 张 琪, 杨福兴, 等. 土壤墒情预测自适应遗传神经网络算法研究[J]. 计算机工程与应用, 2018, 54(1): 54 − 59. doi: 10.3778/j.issn.1002-8331.1608-0261 [13] 冀荣华, 张舒蕾, 郑立华, 等. 基于多值神经元复数神经网络的土壤墒情预测[J]. 农业工程学报, 2017, 33(S1): 126 − 131. [14] Zhang Z W, Liu X T. Prediction of long-term gas load based on particle swarm optimization and gray neural network model[J]. Advances in Mechanical Engineering, 2017, 9(7): 1 − 8. [15] Cheng Y F, Han J S, Liu B, et al. A prediction method for galloping of transmission lines based on improved neural network[J]. International Journal of Mechanical Engineering and Applications, 2018, 4(4): 126 − 133. [16] 任利强, 张立民, 王海鹏, 等. 基于IPSO-GPR的短期负荷区间预测[J]. 计算机工程与设计, 2019, 40(10): 3002 − 3008. [17] 谭 峰, 殷 鸣, 彭 骥, 等. 基于集成BP神经网络的数控机床主轴热误差建模[J]. 计算机集成制造系统, 2018, 24(6): 1383 − 1390. -