基于GWQPSO-SVM的水产养殖水总磷软测量系统设计与试验

Design and test of GWQPSO-SVM based soft sensing system for total phosphorus in aquaculture water

[目的]总磷是影响水产养殖水质质量的重要参数之一,而目前市场上总磷测量装置价格昂贵,实时监测成本高。为实现低成本水产养殖水质监测,设计了基于灰狼量子粒子群-支持向量机(grey wolf quantum particle swarm optimization-support vector machine,GWQPSO-SVM)的水产养殖水质监测系统。[方法]首先,选用传感器组、STM32F103单片机、ESP8266WIFI无线通信模块搭建水质监测系统数据处理模块;分析与水质总磷含量相关性强的水质参数,据此确定系统传感器的选型,设计水质监测系统服务器交互端,开发水质监测小程序对水质等级进行实时监测。其次,提出GWQPSO算法,对SVM进行优化,据此提出GWQPSO-SVM总磷软测量模型。最后,采用南京通威水产科技有限公司的养殖水塘的80组历史水质数据作为训练数据,实时采集55组水质数据作为测试数据对GWQPSO-SVM总磷软预测模型进行试验验证。[结果]试验结果表明,GWQPSO-SVM的5种误差分别为2.2970、0.0418、0.2747、0.0036、0.0599,相较于SVM模型,分别下降了65.70%、65.68%、61.85%、88.16%、65.63%;GWQPSO收敛时的迭代次数、最终收敛适应度分别为74、0.0812,相较于PSO算法,分别下降了87.84%、10.47%。[结论]本文研究可为水产养殖业提供一种高精度、低成本的水质参数测量技术方案。

[Objectives]Total phosphorus is one of the important parameters affecting the quality of aquaculture water.At present,the total phosphorus measuring device on the market is expensive and the cost of real-time monitoring is high.In order to realize low-cost water quality monitoring,the research designed a water quality monitoring system for aquaculture based on GWQPSO-SVM(grey wolf quantum particle swarm optimization-support vector machine).[Methods]Firstly,data processing module of water quality monitoring system was built by selecting sensor group,STM32F103 single-chip computer and ESP8266WIFI wireless communication module.Based on the analysis of water quality parameters with strong correlation to total phosphorus content in water quality,the selection of system sensors was determined,the server interactive terminal of water quality monitoring system was designed,and a small water quality monitoring program was developed to monitor the water quality level in real time.Secondly,a GWQPSO algorithm was proposed to optimize the SVM,and a GWQPSO-SVM total phosphorus soft measurement model was proposed accordingly.Finally,the GWQPSO-SVM soft prediction model of total phosphorus was tested and validated by using 80 sets of historical water quality data as training data and 55 sets of real-time water quality data as test data in the aquaculture pond of Nanjing Tongwei Aquaculture Technology Co.,Ltd.[Results]The test results showed that the five errors of GWQPSO-SVM were 2.2970,0.0418,0.2747,0.0036 and 0.0599,respectively.Compared with the SVM model,these errors decreased 65.70%,65.68%,61.85%,88.16%and 65.63%,respectively.The number of iterations and the final convergence adaptability of GWQPSO at convergence were 74 and 0.0812 respectively,which were 87.84%and 10.47%lower than that of PSO algorithm.[Conclusions]This research can provide a high-precision and low-cost water quality parameter measurement technical scheme for aquaculture industry.