基于物联网的传感器校正方法研究

发布时间：2018-04-14 10:31

本文选题：空气质量监测系统 + 气体交叉干扰　；参考：《北京交通大学》2017年硕士论文

【摘要】：近年来,空气污染日益严重,提升空气质量是民众的迫切期盼,因此空气质量监测系统的建设也成为各地环境保护局和众多环境保护企业所关注的焦点领域。建设空气质量监测系统首先要在需要进行监测的位置布置气体传感器,然后将所有气体传感器连接组成一个传感器网络。由于气体传感器的气敏特性,气体传感器在监测混合气体污染物时会受到交叉干扰,导致监测不准确。本文主要针对电化学气体传感器在工业园区周围监测无机气体污染物时产生交叉干扰的问题进行具体的校正方法研究。目前,主流的校正方法是基于气体传感器集群形成的物联网,利用神经网络对上传的数据进行训练学习,构建传感器校正模型。因此,本文对基于误差反向传播(BP)神经网络的传感器校正模型,进行了算法优化以及模型改进。论文的主要工作包括:首先对常规的基于BP神经网络的校正模型使用的BP算法进行分析,指出BP算法由于其理论缺陷,在训练校正模型阶段易于陷入局部最优的误区,并提出将粒子群(PSO)算法与BP算法结合的优化算法。该优化算法从优化网络初始权重的角度,充分发挥PSO算法全局寻优的优势,结合BP算法局部最优的特点,有效避免了校正模型在训练过程中陷入局部极小的情况,并加快了校正模型训练的收敛速度。然后对校正模型的工作方式以及实际气体传感器的监测数据进行分析,从信息利用以及校正模型实际应用所处环境的角度,指出常规的基于BP神经网络的校正模型不能充分利用气体传感器监测浓度的变化信息,以致校正模型的校正精度受限,并提出融合了长短期记忆(LSTM)神经网络和BP神经网络的改进校正模型。该改进模型首先利用气体浓度变化在时间上连续的特点,通过LSTM网络消除了环境中未知杂气对气体传感器的影响,然后级联BP神经网络,实现改进模型的构建,提高了校正模型的工作性能。论文最后从数据验证的角度,结合仿真建模得到的实验数据,进一步分析证明了文中提出的优化算法较之BP算法在校正模型训练过程中的优越性,以及改进的校正模型在实际应用环境中比常规校正模型拥有更高的校正精度。
[Abstract]:In recent years, the increasingly serious air pollution, improve air quality is urgently looking forward to the public, so the construction of air quality monitoring system has become the focus of the field of environmental protection and Environmental Protection Bureau around many of the concerns of the enterprise. Firstly, the construction of air quality monitoring system should be located in the monitoring of the gas sensor, and then all the gas sensor connection a sensor network. Because of the characteristics of the gas sensors, gas sensors will be cross interference in the monitoring of mixed gas pollutants, resulting in inaccurate monitoring. This paper mainly studies the concrete correction method for electrochemical gas sensor cross interference in the industrial park around the problem of monitoring of inorganic gas pollutants. At present, the mainstream correction method the network is based on gas sensor clusters, using neural network to upload the number According to the learning and training, build the sensor calibration model. Therefore, this paper based on back propagation (BP) neural network sensor calibration model, the optimization algorithm and the model is improved. The main work includes: firstly, the conventional BP neural network calibration models using BP algorithm based on the analysis, pointed out that the BP algorithm because of the defects in the training theory, correction model is easy to fall into local optimum phase errors, and puts forward the particle swarm optimization (PSO) algorithm and BP algorithm. The optimization algorithm in optimizing the initial weights of the network perspective, give full play to the advantages of global optimization PSO algorithm, combined with the characteristics of BP local optimal algorithm, effective to avoid the calibration model in the process of training into the local minimum, and speed up the convergence correction model training. Then the correction model works and actual gas The monitoring data of the sensor is analyzed from the practical application of the model and the correction of information using the environment point of view, pointed out that the conventional BP neural network model based on the correction cannot make full use of the change information monitoring the concentration of gas sensors, so that the correction model correction accuracy is limited, and puts forward the integration of long short term memory (LSTM) neural network and improved BP neural network calibration model. The improved model using gas concentration change in time is continuous, the influence of environment on the unknown impurity gas gas sensor is eliminated by LSTM network, and then cascaded BP neural network, realize the construction of the improved model, improve the working performance of calibration model. Finally, from the data validation point of view, combined with the experimental data obtained by simulation modeling, further analysis shows that the optimization algorithm proposed in this paper is compared with the BP algorithm in training model The superiority of the process and the improved correction model have higher correction precision in the practical application environment than the conventional correction model.

【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212;TP391.44;TN929.5

【参考文献】