#### 期刊菜单

Prediction of Power Plant Sensor Point Based on LSTM Model
DOI: 10.12677/MOS.2024.131049, PDF, HTML, XML, 下载: 180  浏览: 243

Abstract: With the development of big data, many power plants in China have also begun to improve their in-telligence. The real-time value of sensor points is an important indicator reflecting the current health status of the generator set, which has strong tracking significance. Studying and predicting the real-time value changes of sensor points in a short period of time can help power plant moni-toring personnel understand the current health status of the generator set and respond to unex-pected situations in advance. By modeling and analyzing the time series data of sensor point values, the LSTM model was used to make short-term predictions for subsequent sensor point values, and good fitting prediction results were obtained, fully demonstrating the feasibility of applying the time series model to short-term prediction of sensor point values. With the deepening of intelli-gence in power plants, the accuracy of sensor point prediction should be further improved to help power plant monitoring personnel better cope with abnormal conditions of generator units.

1. 引言

2. 研究方法及模型构建

2.1. LSTM模型

LSTM模型全称长短期记忆网络模型，是在循环神经网络(RNN)的基础上变化而来。

Figure 1. Recurrent neural network model

${h}_{t}={f}_{a}\left({W}_{xh}+{W}_{hh}{h}_{t-1}+{b}_{h}\right)$ (1)

${y}_{t}={W}_{hy}{h}_{t}+{b}_{y}$ (2)

Figure 2. Long short-term memory neural network model

${i}_{t}=\sigma \left({W}_{xi}{x}_{t}+{W}_{hi}{h}_{t-1}+{W}_{ci}{c}_{t-1}+{b}_{i}\right)$ (3)

${f}_{t}=\sigma \left({W}_{xf}{x}_{t}+{W}_{hf}{h}_{t-1}+{W}_{cf}{c}_{t-1}+{b}_{f}\right)$ (4)

${c}_{t}={f}_{t}{c}_{t-1}+{i}_{t-1}\mathrm{tanh}\left({W}_{xc}{x}_{t}+{W}_{hc}{h}_{t-1}+{b}_{c}\right)$ (5)

${o}_{t}=\sigma \left({W}_{xo}{x}_{t}+{W}_{ho}{h}_{t-1}+{W}_{co}{c}_{t}+{b}_{o}\right)$ (6)

${h}_{t}={o}_{t}\mathrm{tanh}\left({c}_{t}\right)$ (7)

2.2. 电厂传感器点位预测模型

Figure 3. Sensor point prediction model based on LSTM

3. 实验与结果分析

3.1. 实验环境以及参数设置

3.2. 数据来源及数据预处理

1) 剔除空行缺失值。

2) 对传感器点位数据的异常值进行过滤。在电厂发电机组实际运行过程中，时常会对其进行停机维护和调整，于是传感器会获取到错误的点位数据。由于每个传感器点位的数值在不同的范围进行波动，因此需要对每一个传感器点位进行异常数据的识别与删除。

3) 将传感器点位进行划分为多个传感器组。在发电机组运行的过程中，传感器组的整体情况是对发电机组健康状况进行判断的重要标准，当一个传感器组的多个点位出现异常时，就会认定其出现了特定的问题，并且会有对应的解决方案。

Table 1. Time series of 5 sensor points in a sensor group

$\frac{{x}_{i}-\mu }{\sigma }$ (8)

3.3. 模型评价指标

MSE (Mean Square Error, MSE)均方误差，反映估计量与被估计量之间差异程度的一种度量：

$loss\left(x,y\right)=\frac{1}{N}\underset{t=1}{\overset{N}{\sum }}{|x-y|}^{2}$ (9)

MAE (Mean Absolute Error)平均绝对误差，预测值和真实值之间的绝对误差的平均数：

$loss\left(x,y\right)=\frac{1}{N}\underset{t=1}{\overset{N}{\sum }}|x-y|$ (10)

SmoothL1Loss，当误差在 $\left(-1,1\right)$ 上是平方误差，其他情况是L1损失，分段使用均方误差和平均绝对误差，用于回归模型：

$loss\left(x,y\right)=\frac{1}{N}\left\{\begin{array}{l}\frac{1}{2}{\left({x}_{i}-{y}_{i}\right)}^{2}\cdots \cdots if\text{}|{x}_{i}-{y}_{i}|<1\\ |{x}_{i}-{y}_{i}|-\frac{1}{2}\cdots \cdots other\end{array}$ (11)

3.4. 实验结果

Figure 4. Comparison of predicted and true values (after standardization) at five points

3.5. 模型评价指标效果

Figure 5. Training-set MSE and test-set MAE, MSE, SmoothL1Loss

4. 结语

NOTES

*通讯作者。

 [1] Fordal, J.M., Schjølberg, P., Helgetun, H., et al. (2023) Application of Sensor Data Based Predictive Maintenance and Artificial Neural Networks to Enable Industry 4.0. Advances in Manufacturing, 11, 248-263. https://doi.org/10.1007/s40436-022-00433-x [2] Box, G.E.P., Jenkins, G.M., Reinsel, G.C., et al. (2015) Time Series Analysis: Forecasting and Control. John Wiley & Sons, New York. [3] Zhang, F., Deb, C., Lee, S.E., et al. (2016) Time Se-ries Forecasting for Building Energy Consumption Using Weighted Support Vector Regression with Differential Evolution Op-timization Technique. Energy and Buildings, 126, 94-103. https://doi.org/10.1016/j.enbuild.2016.05.028 [4] Wong, F.S. (1991) Time Series Forecasting Using Backpropagation Neural Networks. Neurocomputing, 2, 147-159. https://doi.org/10.1016/0925-2312(91)90045-D [5] Wang, L., Zeng, Y.I. and Chen, T. (2015) Back Propagation Neural Network with Adaptive Differential Evolution Algorithm for Time Series Forecasting. Expert Systems with Applications, 42, 855-863. https://doi.org/10.1016/j.eswa.2014.08.018 [6] Connor, J. and Atlas, L. (1991) Recurrent Neural Networks and Time Se-ries Prediction. IUJCNN-91-Seattle International Joint Conference on Neural Networks, Seattle, 08-12 July 1991, 301-306. https://doi.org/10.1109/IJCNN.1991.155194 [7] Hochreiter, S. and Schmidhuber, J. (1997) Long Short-Term Memory. Neural Computation, 9, 1735-1780. https://doi.org/10.1162/neco.1997.9.8.1735 [8] 杨青, 王晨蔚. 基于深度学习LSTM神经网络的全球股票指数预测研究[J]. 统计研究, 2019, 36(3): 65-77. https://doi.org/10.19343/j.cnki.11-1302/c.2019.03.006 [9] 陈亮, 王震, 王刚. 深度学习框架下LSTM网络在短期电力负荷预测中的应用[J]. 电力信息与通信技术, 2017, 15(5): 8-11. https://doi.org/10.16543/j.2095-641x.electric.power.ict.2017.05.002 [10] 季学武, 费聪, 何祥坤, 等. 基于LSTM网络的驾驶意图识别及车辆轨迹预测[J]. 中国公路学报, 2019, 32(6): 34-42. https://doi.org/10.19721/j.cnki.1001-7372.2019.06.003 [11] Greff, K., Srivastava, R.K., Koutnik, J., et al. (2016) LSTM: A Search Space Odyssey. IEEE Transactions on Neural Networks & Learning Systems, 28, 2222-2232. https://doi.org/10.1109/TNNLS.2016.2582924 [12] Kingma, D.P. and Ba, J. (2015) Adam: A Method for Stochastic Op-timization. Computer Science. https://doi.org/10.48550/arXiv.1412.6980