CN-122020250-A - Seasonal river water environment quality evaluation method based on chemical research service data

CN122020250ACN 122020250 ACN122020250 ACN 122020250ACN-122020250-A

Abstract

The invention discloses a seasonal river water environment quality evaluation method based on chemical research service data, which comprises the following steps of 1, deploying a low-field nuclear magnetic resonance on-line detection device on each sampling section, applying radio frequency excitation pulses to a water sample entering a detection cavity, collecting free induction attenuation signals, collecting one-dimensional nuclear magnetic resonance spectrum vectors of each sampling section according to seasonal division to form a nuclear magnetic resonance spectrum data set, 2, constructing a one-dimensional convolutional neural network model to form a pollutant characteristic matrix corresponding to each season of each sampling section, and 3, establishing classical domain matters and section matters corresponding to each water quality grade according to water environment quality standards, and summarizing water environment quality evaluation grades of each season of each sampling section to form a seasonal river water environment quality comprehensive evaluation result. The invention realizes the on-site rapid detection and automatic spectrogram analysis of the organic pollutants, and solves the problem of poor timeliness of the traditional off-line detection.

Inventors

ZHANG MIN
QIAN YUTING
JIANG NA
CHEN XIUNA
WANG CHAO

Assignees

山东省科霖检测有限公司
山东省聊城生态环境监测中心

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. The seasonal river water environment quality evaluation method based on the chemical research service data is characterized by comprising the following steps of: The method comprises the steps of 1, deploying a low-field nuclear magnetic resonance online detection device on each sampling section, applying radio frequency excitation pulses to a water sample entering a detection cavity, collecting free induction attenuation signals, sequentially performing Fourier transform, phase correction and baseline correction on the free induction attenuation signals to obtain nuclear magnetic resonance spectrograms, discretizing the nuclear magnetic resonance spectrograms to obtain one-dimensional nuclear magnetic resonance spectrum vectors, and collecting the one-dimensional nuclear magnetic resonance spectrum vectors of each sampling section according to seasonal division to form a nuclear magnetic resonance spectrum data set; Step 2, nuclear magnetic resonance spectrum analysis and pollutant characteristic extraction based on a convolutional neural network are carried out, wherein the one-dimensional convolutional neural network model comprises a convolutional layer, a batch normalization layer, an activation layer, a pooling layer, a global average pooling layer, a full connection layer and an output layer which are sequentially connected, the one-dimensional convolutional neural network model is trained by utilizing water sample detection data provided by a chemical research service institution, and one-dimensional nuclear magnetic resonance spectrum vectors in the nuclear magnetic resonance spectrum data set are input into the trained one-dimensional convolutional neural network model to obtain pollutant characteristic vectors, so that pollutant characteristic matrixes corresponding to seasons of sampling sections are formed; and 3, comprehensively evaluating the quality of the quaternary water environment based on the extension of the physical elements, namely establishing classical domain physical elements and festival domain physical elements corresponding to each water quality grade according to the quality standard of the water environment, calculating the comprehensive relevance of each pollutant characteristic vector in a pollutant characteristic matrix on each water quality grade, determining the quality evaluation grade of the water environment according to the comprehensive relevance, and summarizing the quality evaluation grade of the water environment in each season of each sampling section to form a seasonal river water environment quality comprehensive evaluation result.
2. The method according to claim 1, wherein in step 1, the magnetic field strength of the low-field nuclear magnetic resonance online detection device is 0.5 tesla, a broadband radio frequency transmitting and receiving coil is configured, the radio frequency excitation pulse is a 90-degree radio frequency excitation pulse, the duration is 10 microseconds, the acquisition duration of the free induction decay signal is 2048 milliseconds, and the sampling frequency is 10000 hertz, so that a free induction decay signal sequence comprising 20480 sampling points is obtained.
3. The method of claim 2, wherein in the step 1, the specific processes of performing fourier transform, phase correction and baseline correction processing on the free induction attenuated signal sequence are that the front 16384 sampling points of the free induction attenuated signal sequence are intercepted, 16384 zero value points are supplemented at the tail end to form a spread sequence of 32768 points, fast fourier transform is performed on the spread sequence to obtain a frequency domain complex spectrum, the real part and the imaginary part of the frequency domain complex spectrum are respectively extracted, the real part is used as an abscissa, the imaginary part is used as an ordinate to draw a phase track curve, a principal axis and a real axis coincide to determine a zero-order phase correction angle by rotating the phase track curve, a linearly increasing phase offset is applied to each frequency point along the direction of a chemical displacement axis to eliminate first-order phase distortion, the real part of the corrected frequency domain complex spectrum is used as an absorption mode spectrum, the 2 areas of minus 1 to 0 function value and 9 to 10 are selected on the chemical displacement axis to be used as baseline sampling areas, the spectrum intensity value corresponding to each frequency point in the baseline sampling area is extracted, the least square fitting is performed by using 3-order polynomials to obtain a baseline function curve, and the spectrum obtained after subtracting the intensity value of the absorption mode spectrum at each frequency point from the corresponding to the baseline nuclear magnetic resonance point is corrected.
4. The method of claim 1, wherein in step 1, the nmr spectrum is uniformly discretized within a range of 0 to 9 chemical shifts, 1 spectrum intensity value is taken at intervals of 0.01 chemical shifts to obtain a one-dimensional nmr spectrum vector containing 900 intensity data points, and the one-dimensional nmr spectrum vectors are collected according to a division manner of 3 months to 5 months in spring, 6 months to 8 months in summer, 9 months to 11 months in autumn, and 12 months to 2 months in winter to form a nmr spectrum data set.
5. The method of claim 1, wherein in step 2, the structure of the one-dimensional convolutional neural network model comprises an input layer, a1 st convolutional layer, a1 st normalized layer, a1 st active layer, a1 st pooling layer, a2 nd convolutional layer, a2 nd normalized layer, a2 nd active layer, a2 nd pooling layer, a 3 rd convolutional layer, a 3 rd normalized layer, a 3 rd active layer, a global average pooling layer, a full connection layer and an output layer in sequence, and the input dimension of the input layer is 900.
6. The method of claim 5, wherein the 1 st convolution layer comprises 32 convolution kernels, the convolution kernel length is 15, the convolution step size is 1, the packing number is 7, the 1 st convolution layer has a pooling window length of 4, the pooling step size is 4, the 2 nd convolution layer comprises 64 convolution kernels, the convolution kernel length is 11, the convolution step size is 1, the packing number is 5, the 2 nd convolution layer has a pooling window length of 5, the pooling step size is 5, the 3 rd convolution layer comprises 128 convolution kernels, the convolution kernel length is 7, the convolution step size is 1, and the packing number is 3.
7. The method of claim 5, wherein the 1 st normalization layer, the 2 nd normalization layer, and the 3 rd normalization layer perform normalization processing on the input sequence by calculating a mean and variance of all elements of the input sequence, dividing each element of the input sequence by a square root of the variance, adding a sum of 0.00001, multiplying the square root of the variance by a learnable scaling parameter, and adding a learnable offset parameter, wherein the 1 st activation layer, the 2 nd activation layer, and the 3 rd activation layer perform a modified linear element activation operation, wherein elements greater than zero remain unchanged, wherein elements less than or equal to zero are set to zero, wherein the 1 st pooling layer and the 2 nd pooling layer perform a maximum pooling operation, and wherein a maximum value within each pooling window is taken as an output.
8. The method according to claim 5, wherein the global averaging pooling layer calculates arithmetic average values of all elements in each sequence to obtain feature vectors containing 128 elements for 128 sequences output by the 3 rd activation layer, the full connection layer maps the feature vectors to output vectors containing 6 elements through a connection matrix containing 128 rows and 6 columns and a bias vector containing 6 elements, and the 6 elements output by the output layer correspond to benzene organic pollutant concentration indication values, phenol organic pollutant concentration indication values, polycyclic aromatic hydrocarbon organic pollutant concentration indication values, organochlorine pollutant concentration indication values, petroleum hydrocarbon pollutant concentration indication values and soluble organic carbon total amount indication values respectively.
9. The method according to claim 1, wherein in the step 3, specific processes of establishing classical domain matter elements and node domain matter elements are that for each pollutant concentration indicated value in a pollutant characteristic vector, an upper limit and a lower limit of a value range of each pollutant concentration indicated value under class I water quality, class II water quality, class III water quality, class IV water quality and class V water quality are respectively determined, a matter with a water quality grade as a matter element, each pollutant concentration indicated value as a characteristic of the matter element, a value range of each characteristic under a corresponding grade as a magnitude range of the corresponding characteristic are formed, 5 classical domain matter elements corresponding to 5 water quality grades are formed, and a value range of each characteristic in all grade ranges is taken as a magnitude range of the corresponding characteristic to form the node domain matter element.
10. The method according to claim 9, wherein in the step 3, the specific process of calculating the comprehensive relevance and determining the water environment quality evaluation level includes calculating, for each pollutant concentration indicated value in the pollutant characteristic vector to be evaluated, a classical domain distance between each pollutant concentration indicated value and a corresponding characteristic value interval in a classical domain object element and a node domain distance between each pollutant concentration indicated value and a corresponding characteristic value interval in a node domain object element, dividing the opposite number of the single index relevance equal to the classical domain distance by half the length of the corresponding characteristic value interval in the classical domain object element if the classical domain distance is smaller than or equal to zero, dividing the single index relevance equal to the classical domain distance by the difference value between the node domain distance and the classical domain distance if the classical domain distance is larger than zero, obtaining the comprehensive relevance by dividing the number of the pollutant concentration indicated value after summing the single index relevance of each pollutant concentration indicated value, and taking the water quality level corresponding to the largest comprehensive relevance value as the water environment quality evaluation level.

Description

Seasonal river water environment quality evaluation method based on chemical research service data Technical Field The invention relates to the field of water quality monitoring by using a deep learning technology, in particular to a seasonal river water environment quality evaluation method based on chemical research service data. Background At present, water environment quality detection mainly depends on two technical routes. The first method is to send the water sample to a chemical research service organization for offline analysis, and perform qualitative and quantitative detection on organic pollutants in the water by adopting an instrument analysis method such as gas chromatography, liquid chromatography, gas chromatography-mass spectrometry and the like. The method has the advantages of high detection precision and reliable result, but has the limitations of long detection period, high cost, incapability of reflecting the dynamic change of water quality in real time and the like. The time from sampling to obtaining a detection report usually needs days or even weeks, and the requirements of water environment emergency monitoring and real-time early warning are difficult to meet. And secondly, on-line monitoring equipment is deployed on a monitoring section to continuously and automatically monitor conventional physicochemical indexes such as pH value, dissolved oxygen, conductivity, turbidity and the like. The online monitoring has the advantages of strong real-time performance and low labor cost, but the existing online monitoring technology is mainly aimed at inorganic indexes and few simple organic matters, and has limited detection capability on complex organic pollutants such as benzene series, polycyclic aromatic hydrocarbon, organic chloride and the like. Nuclear magnetic resonance spectroscopy is an important means for analyzing molecular structures of organic compounds, and can obtain structural and composition information of molecules by detecting resonance absorption signals of nuclei in a sample under the action of an external magnetic field. In recent years, with the progress of permanent magnet technology and radio frequency electronics technology, low-field nuclear magnetic resonance equipment gradually develops toward miniaturization and portability, and conditions are created for the application of nuclear magnetic resonance technology in the field of field detection. However, the application of nuclear magnetic resonance technology to online monitoring of water environments still faces many challenges. The water sample has complex components, resonance peaks of various organic matters in the nuclear magnetic resonance spectrogram are overlapped with each other, the traditional manual spectrogram analysis method has low efficiency and high requirements on professional knowledge, and automatic online detection is difficult to realize. The rise of the deep learning technology provides a new thought for the automatic analysis of complex spectrograms. The convolutional neural network can automatically learn characteristic representation from original data, and has been remarkably successful in the fields of image recognition, voice recognition and the like. The convolutional neural network is applied to spectral data analysis, but a deep learning analysis method aiming at a water sample nuclear magnetic resonance spectrogram is not mature. The existing research is focused on detection of single type of pollutants, and a comprehensive analytical model for simultaneously identifying multiple types of organic pollutants is lacking. In addition, how to deploy a deep learning model to an edge computing device to implement online reasoning is also an engineering problem to be solved. Disclosure of Invention The invention aims to provide a seasonal river water environment quality evaluation method based on chemical research service data, which realizes on-site rapid detection and automatic spectrogram analysis of organic pollutants, solves the problem of poor timeliness of traditional off-line detection, can reveal seasonal variation rules of water quality, and provides scientific basis for river basin water environment management. In order to solve the technical problems, the invention provides a seasonal river water environment quality evaluation method based on chemical research service data, which comprises the following steps: The method comprises the steps of 1, deploying a low-field nuclear magnetic resonance online detection device on each sampling section, applying radio frequency excitation pulses to a water sample entering a detection cavity, collecting free induction attenuation signals, sequentially performing Fourier transform, phase correction and baseline correction on the free induction attenuation signals to obtain nuclear magnetic resonance spectrograms, discretizing the nuclear magnetic resonance spectrograms to obtain one-dimensional nuclear magnetic resonance spectrum ve