基于IPSO优化BP神经网络的蛋鸡舍有害气体监测系统

为实现蛋鸡养殖过程有害气体浓度监测，改善复杂环境下常用气体传感器之间因存在交叉敏感性而导致测量数据不准确的问题，设计了基于IPSO优化BP神经网络模型的有害气体监测系统。选用无线ZigBee模块、传感器模块和STM32模块，搭建了蛋鸡舍各点数据采集硬件平台，利用GPRS远程通信模块将平台采集到的数据传输至服务器，同时开发手机APP软件，对有害气体进行实时监测。利用权重线性递减及改进学习因子的IPSO算法，对BP神经网络进行优化，利用优化后的网络对气体传感器采集到的数据进行处理，有效提高了有害气体的数据精度。利用该系统对河北省保定市某鸡舍有害气体进行测试实验，将传感器测量值与真实值进行对比分析，验证了利用IPSO优化BP神经网络模型的有效性。测试表明，SGP30型二氧化碳传感器测量精度由81.75%提升到94.69%，MQ135型氨气传感器由61.83%提升到91.23%，MQ137型氨气传感器由70.18%提升到91.23%，MQ136型硫化氢传感器由62.35%提升到92.80%，TGS2602型硫化氢传感器由62.97%提升到92.80%。本研究为蛋鸡养殖过程中有害气体的精确监测提供了新方法。

Monitoring System of Harmful Gas in Layer House Based on Improved Particle Swarm Optimization BP Neural Network

In order to monitor the concentration and improve the accuracy of harmful gases during layer breeding, the monitoring system based on improved particle swarm optimization back propagation (BP) algorithm was developed. Wireless ZigBee module, sensor module and STM32 module were used to construct the data collection hardware platform at each point of the layer house, the general packet radio service remote communication module was used to transmit the data to the server, the mobile application (APP) software platform was developed to monitor the layer house in real-time. Based on the linearly decreasing weight and the improved learning factor strategy, the particle swarm optimization BP pattern recognition algorithm was used to process the data. Because of the cross-sensitivity caused by common gas sensors in complex environments, the data was not accurate, to improve the accuracy of harmful gas, improved particle swarm optimization optimized BP neural network model was developed. The environmental monitoring data of a chicken house in Baoding, Hebei Province was analyzed, and the effectiveness of the improved particle swarm optimization BP neural network model algorithm was verified by comparing the measured value with the real value of the sensor. The measurement accuracy of the SGP30 carbon dioxide was increased from 81.75% to 94.69%, the measurement accuracy of the MQ135 ammonia was increased from 61.83% to 91.23%, that of the MQ137 ammonia was increased from 70.18% to 91.23%, that of the MQ136 hydrogen sulfide was increased from 62.35% to 92.80%, and that of TGS2602 hydrogen sulfide was increased from 62.97% to 92.80%. The design process of terminal collection node, server and mobile phone APP in layer house environment was given. The functions of the system were verified by experiments.