The Research On Real-time Detection Of CO₂Concentration And Its Realization

In recent years, with the rapid development of industry, emission of carbondioxide rises sharply, and then the greenhouse effect has threat to the survival of thehuman. Therefore,“low-carbon economy” is addressed, and the reduction andrecycling of carbon dioxide become a hot issue in the twenty-first century. At thesame time, it is necessary to monitoring the emissions of carbon dioxide. Someexperts and scholars have researched a number of detecting methods and some relatedequipments, in which optical absorption spectroscopy has achieves on-line inspectionof carbon dioxide gas concentration. The advantage of this method is of high accuracyand good real-time, while the stability and robustness is poor. Considering theproblem that measurement result is apt to be disturbed under actual circumstance, thisdissertation makes the following research:Firstly, by analyzing the theory of spectral absorption, we know that there are afew limitations at detecting concentration of carbon dioxide, such as the test result isdirectly related to temperature and pressure, and has uncertain relation to humidityand similar gases under actual circumstance. On this basis, a detected method forcorrecting error by multi-sensor data fusion technology is proposed in this paper. Themethod selects Extreme Learning Machine model as the fusion mode, which isapplicable to real-time detection.Then, the following four factors are selected: temperature, pressure, humidity,similar gas, and simulation experiment is designed, including the hardware part andsoftware part. Based on the experiment, we analyze influence characteristic of singlevariable by changing the variable under the overall environment variable. Sample datais got by changing overall variable, and every variable should change one cycleperiod.Next, normalization method is used to deal with the sample data, and severalfusion models are used to analysis results. The first model is directly use ELMnetwork. The second one use K-means method for data classification, and thenimprove ELM network. The third one improves the ELM model by applyingasynchronous multi-sensor information fusion algorithm based on missingmeasurements theory.Finally, the results of direct measurement of TDLAS and fusion modified data of fusion model are compared from precision, robustness and controllability. The resultshow that environment variable affect the detecting data largely when circumstanceschange a lot, and the outcome modified by improved ELM model has higher precision,more robust, wider adaptable than the outcome detected by TDLAS.