Research on the optimization processing of greenhouse environmental data based on EEMD-WPT
-
摘要:目的
解决温室系统中的数据采集传感器容易受到多种环境因素的干扰，从而导致数据中存在噪声的问题。
方法提出一种集合经验模态分解(Ensemble empirical mode decomposition，EEMD)与小波包自适应阈值 (Wavelet packet adaptive threshold，WPT) 算法联合的数据降噪处理方法，并采用卡尔曼滤波与自适应加权平均算法对降噪后的数据进行融合。
结果将EEMD-WPT算法应用于含噪温、湿度数据的降噪处理，相较于降噪前的数据，信噪比提升了73.08%。该算法相较于传统WPT算法具有更好的降噪效果，处理后的数据信噪比提升了40.31%，均方根误差降低了84.75%。
结论该算法能解决数据跳动、冗余和丢失等问题，并为温室控制系统提供了有效的参数，具有较大的实际应用价值。
Abstract:ObjectiveTo address the problem that the data acquisition sensors in greenhouse system are easily disturbed by various environmental factors, leading to the presence of noise in the data.
MethodThis study proposed a data noise reduction processing method combining ensemble empirical mode decomposition (EEMD) and wavelet packet adaptive threshold (WPT) algorithm, and the Kalman filter and adaptive weighted average algorithm were used to fuse the noise-reduced data.
ResultAfter applying the EEMD-WPT algorithm to the noise reduction processing of the noise-containing temperature and humidity data, the signal-to-noise ratio was improved by 73.08% compared with the data before noise reduction. The EEMD-WPT algorithm had better noise reduction effect compared with the traditional WPT algorithm, and the signal-to-noise ratio of the processed data was improved by 40.31% and the root mean square error reduced by 84.75%.
ConclusionThe algorithm can solve the problems of data skipping, redundancy and loss, and provides effective parameters for the greenhouse control system, making it highly practical and valuable for application.
-
Keywords:
- EEMD /
- Wavelet packet /
- Adaptive threshold /
- Noise reduction /
- Greenhouse /
- Data fusion
-
-
![]()
图 2 传感器节点在温室中的位置
红色数字表示节点编号，括号中为节点位置
Figure 2. Positions of sensor nodes in the greenhouse
The red number represents the node number, and the node position is indicated in parentheses
![]()
图 4 纯净温度数据(A)与含噪温度数据(B)
Figure 4. Pure temperature data (A) and temperature data with noise (B)
![]()
图 5 不同算法对温度数据降噪处理的仿真结果
Figure 5. Simulation results of different algorithms for noise reduction processing of temperature data
![]()
图 6 温度(A)和相对湿度(B)误差补偿模型
Figure 6. Temperature (A) and relative humidity (B) error compensation models
![]()
图 7 温度(A)和相对湿度(B)误差分布直方图
Figure 7. Temperature (A) and relative humidity (B) error distribution histogram
![]()
图 10 温度数据不同阈值降噪效果对比
Figure 10. Comparison of noise reduction effects of different thresholds for temperature data
![]()
图 11 相对湿度(RH)数据不同阈值降噪效果对比
Figure 11. Comparison of noise reduction effect of different thresholds for realative humidity (RH) data
![]()
图 12 不同算法对温度数据降噪处理结果
Figure 12. Results of noise reduction processing of temperature data by different algorithms
![]()
图 13 不同算法对相对湿度(RH)数据降噪处理结果
Figure 13. Results of noise reduction processing of relative humidity (RH) data by different algorithms
表 1 模型和训练的超参数设置
Table 1 Hyperparameter settings for models and training
参数类型
Parameter type参数
Parameter参数值
Parameter value模型参数
Model parameter全连接神经网络的层数
The number of layers of a fully connected neural network3 输入维度 Input dimension 2 隐藏层维度
Hidden layer dimension10 输出维度 Output dimension 2 训练参数
Training parameter学习率 Learning rate 0.001 训练轮次 Training round 100 批量训练的批大小
Batch size for batch training32 交叉验证的 K-fold
Cross-validated K-fold5 优化器 Optimizer Adam 
下载: 导出CSV
表 2 不同算法温度数据输出结果的RMSE、SNR和R
Table 2 The RMSE, SNR and R of the temperature data output result of each algorithm
算法 Algorithm RMSE SNR/dB R LMS 0.219 41.20 0.999 FIR 0.112 47.05 0.999 IIR 0.189 42.52 0.998 VMD-WT 0.136 45.33 0.999 VMD 0.155 44.23 0.999 WPT 0.236 40.56 0.997 EMD-WPT 0.047 54.61 1.000 EEMD-WPT 0.036 56.91 1.000 CEEMDAN-WPT 0.056 53.10 1.000
下载: 导出CSV
表 3 不同阈值去噪方法温度数据输出结果的RMSE、SNR和R
Table 3 RMSE, SNR and R of temperature data output results of different threshold denoising methods
去噪方法 Denoising method RMSE SNR/dB R 硬阈值去噪 Hard threshold denoising 0.224 40.69 0.996 软阈值去噪 Soft threshold denoising 0.224 40.70 0.995 固定阈值去噪 Fixed threshold denoising 0.224 40.70 0.996 自适应阈值去噪 Adaptive threshold denoising 0.169 43.13 0.998
下载: 导出CSV
表 4 不同阈值去噪方法相对湿度数据输出结果的RMSE、SNR和R
Table 4 RMSE, SNR and R of relative humidity data output results of different threshold denoising methods
去噪方法 Denoising method RMSE SNR/dB R 硬阈值去噪 Hard threshold denoising 0.335 46.45 0.951 软阈值去噪 Soft threshold denoising 0.336 46.45 0.950 固定阈值去噪 Fixed threshold denoising 0.162 52.76 0.989 自适应阈值去噪 Adaptive threshold denoising 0.120 55.40 0.994
下载: 导出CSV
-
[1] 杜肖鹏, 李恺, 王春辉, 等. 国内温室空气温湿度检测及传输技术研究进展[J]. 农业工程技术, 2022, 42(34): 28-34. [2] 张敏, 方健, 王勇, 等. 基于ZigBee无线通信的变电站端子箱温湿度传感器系统设计[J]. 制造业自动化, 2023, 45(2): 169-174. doi: 10.3969/j.issn.1009-0134.2023.02.034 [3] 于玉婷, 柳平增, 王秀丽, 等. 北方番茄日光温室温度传感器布设位置研究[J]. 中国农业科技导报, 2023, 25(3): 107-118. [4] MANOHARAN D S. Supervised learning for microclimatic parameter estimation in a greenhouse environment for productive agronomics[J]. Journal of Artificial Intelligence and Capsule Networks, 2020, 2(3): 170-176. doi: 10.36548/jaicn.2020.3.004
[5] LIU F, LI J, LIU L, et al. Application of the EEMD method for distinction and suppression of motion-induced noise in grounded electrical source airborne TEM system[J]. Journal of Applied Geophysics, 2017, 139: 109-116. doi: 10.1016/j.jappgeo.2017.02.013
[6] THUC V C, LEE H S. Partial discharge (PD) signal detection and isolation on high voltage equipment using improved complete EEMD method[J]. Energies, 2022, 15(16): 5819. doi: 10.3390/en15165819
[7] VANUS J, KUBICEK J, VILIMEK D, et al. A innovative wavelet transformation method optimization in the noise-canceling application within intelligent building occupancy detection monitoring[J]. Heliyon, 2023, 9(5): e16114. doi: 10.1016/j.heliyon.2023.e16114
[8] 熊春宝, 于丽娜, 常翔宇. 基于EEMD−小波阈值去噪的桥梁结构模态参数识别[J]. 天津大学学报(自然科学与工程技术版), 2020, 53(4): 378-385. [9] XIA S, NAN X, CAI X, et al. Data fusion based wireless temperature monitoring system applied to intelligent greenhouse[J]. Computers and Electronics in Agriculture, 2022, 192: 106576. doi: 10.1016/j.compag.2021.106576
[10] WANG C, WANG Z, ZHU Y, et al. The application of data-level fusion algorithm based on adaptive-weighted and support degree in intelligent household greenhouse[M]//Innovative Techniques and Applications of Modelling, Identification and Control. Singapore: Springer, 2018: 93-108.
[11] 王子乐, 王子谋, 蔡莹, 等. 基于长短期记忆神经网络组合算法的短期电力负荷预测[J]. 现代电力, 2023, 40(2): 201-209. [12] 肖旰, 周莉, 李敬兆, 等. 基于EEMD融合BAS-CNN的高压电缆故障诊断[J]. 电子测量技术, 2022, 45(4): 160-167. [13] VENKATESAN C, KARTHIGAIKUMAR P, VARATHARAJAN R. A novel LMS algorithm for ECG signal preprocessing and KNN classifier based abnormality detection[J]. Multimedia Tools and Applications, 2018, 77(8): 10365-10374. doi: 10.1007/s11042-018-5762-6
[14] ALMALCHY M T, CIOBANU V, POPESCU N. Noise removal from ecg signal based on filtering techniques[C]//2019 22nd International Conference on Control Systems and Computer Science (CSCS). Bucharest, Romania: IEEE, 2019: 176-181.
[15] APPATHURAI A, JERUSALIN CAROL J, RAJA C, et al. A study on ECG signal characterization and practical implementation of some ECG characterization techniques[J]. Measurement, 2019, 147: 106384. doi: 10.1016/j.measurement.2019.02.040
[16] 陈龙, 史文库, 陈志勇, 等. 基于小波−VMD联合降噪的传动系扭振信号分析[J]. 振动. 测试与诊断, 2022, 42(4): 703-709,825. [17] 刘伯相, 张远民, 江竹. 基于变分模态分解的管道泄漏信号降噪方法[J]. 振动. 测试与诊断, 2023, 43(2): 397-403. [18] 杨洪涛. 样本熵改进小波包阈值去噪的轴承故障诊断[J]. 组合机床与自动化加工技术, 2020(1): 79-82. [19] KE L. Denoising GPS-based structure monitoring data using hybrid EMD and wavelet packet[J]. Mathematical Problems in Engineering, 2017, 2017: 4920809.
[20] 李瑞, 谷立臣, 赵鹏军, 等. 应用EEMD和小波包分解的压力脉动信号时域特征提取方法[J]. 现代制造工程, 2018(7): 1-6. [21] 栾孝驰, 李彦徵, 徐石, 等. 基于小波包变换与CEEMDAN的滚动轴承故障诊断方法[J]. 航空动力学报, 2024, 39(5): 20220473.
计量
- 文章访问数: 261
- HTML全文浏览量: 35
- PDF下载量: 35
