Method for quantitatively predicting microorganisms in automobile air conditioner by using electronic nose

Abstract

The invention discloses a method for quantitatively predicting microorganisms in an automobile air conditioner by using an electronic nose, and belongs to the technical field of microorganism detection. The method comprises the following steps: collecting and treating a microorganism sample in the automobile air conditioner; acquiring smell sample information by using an electronic nose; selecting a data feature extraction method; and constructing a microbe quantity prediction model. The electronic nose is used for detecting microorganisms in the automobile air conditioner and comprises the steps that the electronic nose is arranged at an air outlet of the automobile air conditioner; an electronic nose system is used for feeding back the number of microorganisms in real time to decide whether the cabin environment is seriously polluted by microorganisms at the moment. By utilizing the electronic nose, nondestructive, rapid, easy-to-operate and low-cost quantitative detection of microorganisms in the automobile air conditioner can be realized, and the detection result is fed back in real time, so that the problems of high detection cost, complicated operation, need of professional sample treatment operation and the like in the existing microorganism detection technology are solved; and a more convenient and rapid method can be provided for detecting the microorganisms in the automobile air conditioner.

Description

Technical Field

[0001] The invention belongs to the technical field of microorganism detection, and in particular relates to a method for quantitatively predicting microorganisms in automobile air conditioners by using an electronic nose. Background Art

[0002] Microbial metabolism produces a variety of volatile gases such as aldehydes, alcohols, ketones, nitrogen-containing, oxygen-containing or sulfur-containing compounds, esters, volatile phenols and hydrocarbons. The presence of a large number of microorganisms in the automotive air-conditioning system is one of the main causes of odor in the vehicle cabin and cabin environmental pollution. At the same time, there are a large number of microbial communities in the air-conditioning filter, including more than 400 species of bacteria and fungi such as Cladosporium, Penicillium, and Aspergillus, making the automotive air-conditioning system the main source of microbial hazards in the car. Microbial contamination of the car cabin environment can cause serious health threats to drivers and passengers. For example, inhalation of bacteria or fungi can cause diseases such as asthma, pneumonia, rhinitis, and even some bacteria are directly related to cancer. Therefore, rapid quantitative prediction of microorganisms in the car air-conditioning is conducive to timely reminding drivers to clean the air-conditioning system and disinfect the cabin to improve the air quality in the cabin.

[0003] The number of microorganisms is directly related to the degree of microbial contamination. When the number of microorganisms exceeds 200 CFU / cm2, the environment at this time will be seriously polluted. Therefore, the number of microorganisms in the air-conditioning system can be used to assess its degree of contamination. At present, there are many methods to detect the number of microorganisms. The standard microbial detection method is the plate count method, which can accurately provide the number of microorganisms, but this method is time-consuming and cannot provide real-time information on microorganisms; polymerase chain reaction (PCR), enzyme-linked immunosorbent assay (ELISA) and adenosine triphosphate bioluminescence assay (ATP) are widely studied rapid microbial detection methods, but the detection cost is high, the operation is complicated, and professional sample processing operations are required. In addition, the sampling of these detection methods is destructive and is not suitable for in situ and rapid prediction of microorganisms in automobile air-conditioning systems.

[0004] Microbial metabolism produces a variety of volatile gases such as aldehydes, alcohols, ketones, nitrogen-, oxygen- or sulfur-containing compounds, esters, volatile phenols and hydrocarbons. The number of microorganisms can be predicted by analyzing the gas blown out of the air conditioning system. The electronic nose is an intelligent detection device that imitates the biological olfactory system. It obtains the odor information of the target volatile compounds through a sensor array, and then uses a machine learning algorithm to analyze the data and generate decision results. Compared with other gas detection equipment, the electronic nose has the advantages of low cost, fast detection speed and small size. Given the advantages of the electronic nose, it is suitable for in-depth analysis of the odor information produced by microbial metabolism and its application in the prediction of microorganisms in automobile air conditioners. Summary of the invention

[0005] The purpose of the present invention is to provide a method for quickly predicting microorganisms in automobile air conditioners using an electronic nose, which can solve the problems of the prior art, such as complex operation, long time consumption, and influence on the normal use of automobile air conditioners.

[0006] The method for quantitatively predicting microorganisms in automobile air conditioners using an electronic nose of the present invention comprises the following steps:

[0007] 1) Sample preparation: Collect used automobile air conditioning filters as microbial detection samples for subsequent microbial analysis and electronic nose data collection;

[0008] 2) calibrating the number of microorganisms in the microbial test sample obtained in step 1) by plate counting method, comprising the following steps:

[0009] 2.1 Determination of the total number of microorganisms by plate count method: According to the procedures specified in "WS 394-2012 Hygienic Standards for Central Air Conditioning and Ventilation Systems in Public Places", the reference value of the amount of microorganisms in automobile air conditioning filters is determined by plate count method;

[0010] 2.2 According to the number of microorganisms, the microbial test samples are divided into: Microbial number <200CFU / cm 2 Low microbial contamination and microbial count ≥ 200 CFU / cm 2 High microbial contamination;

[0011] 3) Use electronic nose to obtain odor information of microbial samples: The sensor array in the electronic nose is composed of different types of MOS sensors;

[0012] 3.1 Before the electronic nose test, the microbial test sample of the air conditioning filter was placed in a square container and incubated for 30 minutes in a sealed container to produce headspace volatiles;

[0013] 3.2 The sampling frequency and sampling time of the electronic nose are set to f(50≤f≤100)Hz and t1(30≤f≤120)s respectively;

[0014] 3.3 Use an air pump to pump the headspace gas into the electronic nose to react with the sensor array;

[0015] 3.4 Before the next measurement, purge the electronic nose with clean air, and the purge time is set to t2 (120 ≤ f ≤ 300) s;

[0016] 4) Preprocessing the response data of the electronic nose includes the following steps:

[0017] 4.1 Feature extraction: Use maximum value--MAX, average value--Mean, wavelet transform--WT, Fourier transform--FT, principal component analysis--PCA for feature extraction;

[0018] 4.2 Feature selection: Compare the prediction accuracy of the feature extraction method in step 4.1 through the machine learning model and select the best features for subsequent data analysis;

[0019] 4.3 Normalize the extracted features;

[0020] 5) Constructing an electronic nose data processing model, including the following steps:

[0021] 5.1 A qualitative-quantitative two-stage hybrid model is proposed. The qualitative model provides category labels for microbial samples, and the quantitative model predicts the number of microorganisms. Specifically, it includes the following steps:

[0022] 5.1.1 According to whether the number of microorganisms in the air conditioning filter exceeds 200 CFU / cm2, the microbial samples are divided into high microbial contamination samples and low microbial contamination samples;

[0023] 5.1.2 According to the sample category label, conduct microbial quantitative analysis on each type of microbial sample to provide accurate information on the number of microorganisms;

[0024] 5.2 Select the qualitative analysis model: Use the back propagation network (BPNN), support vector machine model (SVM), random forest (RF), radial basis neural network (RBF) to classify the samples, compare the recognition accuracy of each classifier, and select the optimal classifier as the qualitative analysis model;

[0025] 5.3 Select the quantitative prediction model: Use artificial neural network model-ANN, support vector machine model-SVM, random forest-RF, extreme learning machine model-ELM to predict the samples, compare the prediction accuracy of each regression model, and select the optimal regression model as the quantitative prediction model;

[0026] 6) Model performance evaluation: Evaluate the prediction accuracy of the final qualitative-quantitative hybrid model;

[0027] 7) The electronic nose is placed at the air outlet of the car air conditioner for on-board detection, and real-time feedback is given on the number of microorganisms in the car air conditioning system. If the number of microorganisms is predicted to be excessive, the driver is reminded to replace the air conditioning filter or disinfect the cabin.

[0028] The beneficial effects of the present invention are:

[0029] 1. The quantitative prediction method of microorganisms in automobile air conditioners based on a qualitative-quantitative two-stage model proposed in this paper is different from the traditional mature

[0030] Compared with qualitative and quantitative methods, it has the characteristics of non-destructive, easy operation, fast and time-saving, real-time detection, and low cost. It can quickly and reliably determine the microbial contamination of odors in cars and car air-conditioning systems.

[0031] 2. The method of the present invention can perform real-time dynamic monitoring of the number of microorganisms in automobile air conditioners and has high prediction accuracy.

[0032] And timely

[0033] Remind drivers to take measures to improve the air quality in the car and effectively protect passengers from health threats posed by the car environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The prediction accuracy results for different features;

[0035] Figure 2 For in-car odor identification based on electronic nose system and microbial characterization process of air conditioning system;

[0036] Figure 3 Comparison between the prediction results of the validation set samples by the BPNN-SVM hybrid model and the true values ​​of the samples. DETAILED DESCRIPTION

[0037] The present invention is a qualitative-quantitative two-stage model design for quantitatively predicting microorganisms in automobile air conditioners using electronic nose technology. The present invention is described below by taking the detection of volatile odors and the number of microorganisms in three old air conditioner filters collected from three different cars as an example.

[0038] 1) Sample preparation: Three old air conditioning filters were collected from three different cars. The collected air conditioning filters were placed in a glass box with a small hole on the top. The three air conditioning filters were placed in a high humidity environment (80%-90% relative humidity) for 0.5-20 days and normal humidity (40%-50%) for 0.5-40 days. 34 samples with different numbers of microorganisms were obtained as microbial detection samples for subsequent microbial analysis and electronic nose data collection.

[0039] 2) calibrating the number of microorganisms in the microbial test sample obtained in step 1) by plate counting method, comprising the following steps:

[0040] 2.1 Determination of the total number of microorganisms by plate counting method: According to the procedures specified in "WS 394-2012 Hygiene Standards for Central Air Conditioning and Ventilation Systems in Public Places", the reference value of the amount of microorganisms in the air conditioning filter element is determined by plate counting method. 2The air conditioner filter element is sampled, and the collected samples are appropriately diluted and placed on plate technology agar for incubation (bacteria: 35-37℃ for 48h, fungi: 28℃ for 5d). The reference value of the number of microorganisms in the sample is calculated based on the bacterial count results and the dilution ratio.

[0041] 2.2 The total number of microorganisms in the prepared test samples is between 17 and 391 CFU / cm 2 The samples were divided into two categories: low and high microbial contamination. The samples were divided into: microbial count <200 CFU / cm 2 Low microbial contamination and microbial count ≥ 200 CFU / cm 2 High microbial contamination.

[0042] 3) Use an electronic nose to obtain odor information of microbial samples: The sensor array in the electronic nose used consists of 25 MOS sensors, including: TGS2612, TGS2611, TGS2620, TGS2603, TGS2602, TGS2610, TGS2600, GSBT11, MS1100, MP135, MP901, MP-9, MP-3B, MP-4, MP-5, MP-2, MP503, MP801, MP905, WSP1110, WSP2110, WSP7110, MP-7, MP-702 and TGS2618.

[0043] 3.1 Before the electronic nose test, the air conditioning filter sample was placed in a square container and incubated for 30 minutes in a sealed container to produce headspace volatiles.

[0044] 3.2 The sampling frequency and sampling time of the electronic nose are set to 100 Hz and 60 s respectively.

[0045] 3.3 Use an air pump to pump the headspace gas into the electronic nose to react with the sensor array.

[0046] 3.4 Before the next measurement, purge the electronic nose with clean air and set the purge time to 5 minutes.

[0047] 4) The 34 samples were tested 10 times using an electronic nose, resulting in a total of 340 sample data sets, each of which consisted of a matrix of 6000 rows (100 Hz × 60 s) and 25 columns (25 MOS sensor arrays). The 340 data sets were divided into 308 sets as training subsets for model training and testing, and 32 sets as validation subsets for evaluating the prediction accuracy of the model. The electronic nose response data was preprocessed, including the following steps:

[0048] 4.1 Feature extraction: Different methods are used for feature extraction, including maximum value (MAX), average value (Mean), wavelet transform (WT), Fourier transform (FT), principal component analysis (PCA), etc.

[0049] 4.2 Feature Selection: The prediction accuracy of the above feature extraction methods is compared by using the support vector machine model. Figure 1 Therefore, the average value feature is used as the feature extraction method of the electronic nose for subsequent data analysis.

[0050] 4.3 Normalize the extracted features, the formula is as follows:

[0051]

[0052] Among them, x i is the input data, x mean is the mean value of the data series, x stan is the standard deviation of the data series.

[0053] To build an electronic nose data processing model, the steps are as follows:

[0054] 5) Build a qualitative-quantitative two-stage hybrid model framework, as shown in the following figure: Figure 2 shown.

[0055] 5.1 Select the qualitative analysis model: Four classifiers are selected, namely, back propagation network (BPNN), support vector machine (SVM), random forest (RF) and radial basis neural network (RBF). The accuracy of the classifiers for the sample data set is shown in Table 1. The back propagation neural network (BPNN) with the highest accuracy is selected as the qualitative analysis model.

[0056] Table 1 Accuracy of different classification models for microbial datasets

[0057]

[0058] 5.2 Select the quantitative prediction model: Four regression models are selected, including artificial neural network model (ANN), support vector machine model (SVM), random forest (RF) and extreme learning machine model (ELM). The prediction accuracy of each regression model is shown in Table 2. The support vector machine model (SVM) with the highest accuracy is selected as the quantitative prediction model.

[0059] Table 2 Prediction accuracy of each model for different data sets

[0060]

[0061] 6) Model performance evaluation: The final BPNN-SVM two-stage hybrid model was evaluated for accuracy using the validation set. The R2 of BPNN-SVM on the validation set was 0.94 and the RMSE was 28.293. Figure 3 The difference between the predicted values ​​and true values ​​of the BPNN-SVM model for all validation set samples is shown. The TBC of most samples can be accurately identified. In the validation data set, the test sample with a TBC of 40 CFU / cm2 generated a relatively reliable predicted TBC value (40.3414 CFU / cm2).

[0062] 7) The electronic nose is placed at the air outlet of the car air conditioner for on-board detection, and real-time feedback is given on the number of microorganisms in the car air conditioning system. If the predicted number of microorganisms is too high, the driver is prompted to replace the air conditioning filter or disinfect the cabin.

Claims

1. A method for quantitatively predicting microorganisms in automobile air conditioners using an electronic nose, characterized in that The following steps are involved: 1) Sample preparation: Collect used automobile air conditioning filters as microbial detection samples for subsequent microbial analysis and electronic nose data collection; 2) calibrating the number of microorganisms in the microbial test sample obtained in step 1) by plate counting method, comprising the following steps: 2.1 Determination of the total number of microorganisms by plate count method: According to the procedures specified in "WS 394-2012 Hygienic Standards for Central Air Conditioning and Ventilation Systems in Public Places", the reference value of the amount of microorganisms in automobile air conditioning filters is determined by plate count method; 2.2 According to the number of microorganisms, the microbial test samples are divided into: Microbial number <200CFU / cm 2 Low microbial contamination and microbial count ≥ 200 CFU / cm 2 High microbial contamination; 3) Use electronic nose to obtain odor information of microbial samples: The sensor array in the electronic nose is composed of different types of MOS sensors; 3.1 Before the electronic nose test, the microbial test sample of the air conditioning filter was placed in a square container and incubated for 30 minutes in a sealed container to produce headspace volatiles; 3.2 The sampling frequency and sampling time of the electronic nose are set to f(50≤f≤100)Hz and t1(30≤f≤120)s respectively; 3.3 Use an air pump to pump the headspace gas into the electronic nose to react with the sensor array; 3.4 Before the next measurement, purge the electronic nose with clean air, and the purge time is set to t2 (120 ≤ f ≤ 300) s; 4) Preprocessing the response data of the electronic nose includes the following steps: 4.1 Feature extraction: Use maximum value--MAX, average value--Mean, wavelet transform--WT, Fourier transform--FT, principal component analysis--PCA for feature extraction; 4.2 Feature selection: Compare the prediction accuracy of the feature extraction method in step 4.1 through the machine learning model and select the best features for subsequent data analysis; 4.3 Normalize the extracted features; 5) Constructing an electronic nose data processing model, including the following steps: 5.1 A qualitative-quantitative two-stage hybrid model is proposed. The qualitative model provides category labels for microbial samples, and the quantitative model predicts the number of microorganisms, including the following steps: 5.1.1 Whether the number of microorganisms in the air conditioning filter exceeds 200 CFU / cm 2 , the microbial samples were divided into samples with high microbial contamination and samples with low microbial contamination; 5.1.2 According to the sample category label, conduct microbial quantitative analysis on each type of microbial sample to provide accurate information on the number of microorganisms; 5.2 Select the qualitative analysis model: Use the back propagation network (BPNN), support vector machine model (SVM), random forest (RF), radial basis neural network (RBF) to classify the samples, compare the recognition accuracy of each classifier, and select the optimal classifier as the qualitative analysis model; 5.3 Select the quantitative prediction model: Use artificial neural network model-ANN, support vector machine model-SVM, random forest-RF, extreme learning machine model-ELM to predict the samples, compare the prediction accuracy of each regression model, and select the optimal regression model as the quantitative prediction model; 6) Model performance evaluation: Evaluate the prediction accuracy of the final qualitative-quantitative hybrid model; 7) The electronic nose is placed at the air outlet of the car air conditioner for on-board detection, and real-time feedback is given on the number of microorganisms in the car air conditioning system. If the number of microorganisms is predicted to be excessive, the driver is reminded to replace the air conditioning filter or disinfect the cabin.

Images