CN113155767A - Distributed water quality detection system based on ultraviolet spectroscopy and water quality evaluation method - Google Patents

Abstract

The invention discloses a distributed water quality detection system and a water quality evaluation method based on ultraviolet spectroscopy. The system includes: a water quality sensor based on ultraviolet spectroscopy, a spectrum acquisition module, a wireless sensor network, a detection terminal water quality evaluation model, and a water quality sensor. The water samples to be tested are irradiated with ultraviolet spectrum to separate different spectra. The spectrum acquisition module collects the spectral data of the CCD detector, and the spectral data is sent to the terminal through the wireless sensor network. BP neural network was used for optimization, and a water quality evaluation model was established. Both particle swarm optimization and AdaBoost algorithm can effectively solve the problem of unstable evaluation results caused by random initialization parameters of BP neural network. But under the same conditions, AdaBoost algorithm has higher prediction accuracy and shorter training time than PSO algorithm, and has better application value in water quality evaluation.

Description

Distributed water quality detection system and water quality evaluation method based on ultraviolet spectroscopy

technical field

The invention relates to the technical field of water quality detection, in particular to a distributed water quality detection system and a water quality evaluation method based on ultraviolet spectroscopy.

Background technique

With the rapid development of social economy, environmental pollution is increasingly threatening people's life and health. Water pollution has become one of the very serious problems faced by all countries in the world today. It is of great significance to establish a real-time online water quality monitoring system for serious water pollution problems. Based on the UV-Vis spectrum water quality monitoring technology, multi-parameter measurement of water quality can be quickly realized. It has the advantages of simple operation, low cost, secondary pollution, online and on-site measurement, etc., and has become an important development direction of water quality monitoring instruments. Especially when modern water quality monitoring technology puts forward the requirements of micro-portable, low-cost, real-time online, on-site multi-parameter measurement for instruments and equipment, the design and manufacture of such detection systems have become the key technology for spectroscopic water quality measurement.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a distributed water quality detection system and water quality evaluation method based on ultraviolet spectroscopy.

To achieve the above object, the present invention is implemented according to the following technical solutions:

A distributed water quality detection system based on ultraviolet spectroscopy. The ultraviolet light source emits an ultraviolet spectrum that is transmitted through the water source to be tested to become signal light, and then enters the spectrum analyzer to obtain the spectrum arranged in different wavelengths. The array CCD detector converts the spectrum data. The optical signal is converted into an electrical signal, and the signal acquisition and processing module collects the output spectral data of the array CCD detector. The spectral data is uploaded to the water quality monitoring terminal through the wireless sensor network for water quality evaluation, and the composition and content information of the measured water quality are obtained.

Further, the water quality monitoring terminal is a mobile phone or a tablet computer or a PC terminal.

Further, the power supply is a 12V battery, the battery is connected with a photovoltaic charging device, the photovoltaic charging device includes a solar panel, and there is a first ADC conversion circuit between the solar panel and the DC-DC of the battery, which is There is also a second ADC conversion circuit between the DC-DC and the battery. When the voltage measured by the first ADC conversion circuit is too low, disconnect the power connection between the solar panel and the DC-DC; When there is a gap between the voltage value measured by the ADC conversion circuit and the predetermined 24V, the switching PWM duty cycle of the MOSFET should be controlled.

In addition, the present invention also provides a water quality evaluation method, the specific steps are as follows:

S1. Establish the AdaBoost-BP water quality evaluation model in the water quality monitoring terminal,

S2. Import data samples, determine training samples and test samples, and initialize the weights of training sample data. The calculation formula is:

Among them: D _i is the initialization weight, i=1,2,...,m; m is the number of training samples;

S3. Set the number and network structure of the BP weak classification, and use the BP weak predictor to train and predict the training samples, and then obtain the weak classifier C _t (x);

S4. Calculate the classification error of C _t (x):

S5. Calculate the weight of the C _t (x) classifier:

S6. Update the weights of the training data:

where: g _t is the normalization factor, and _yi is the data label;

S7, the final strong classifier:

The water quality evaluation level involves 5 standards, and 5 BP neural networks are used to form a weak classifier group. On the basis of the BP neural network, the strong classifier designed by AdaBoost can effectively improve the instability of a single BP neural network and improve the evaluation. The accuracy and generalization ability of the model;

S8. Use the spectral data of the signal acquisition and processing module as the input of the AdaBoost-BP water quality evaluation model to obtain the water quality evaluation results.

Compared with the prior art, the present invention mainly detects water quality indicators such as COD (chemical oxygen demand), ZD (turbidity), TOC (total organic carbon), nitrate nitrogen (NO3-N) in water by ultraviolet spectroscopy. , the water samples to be tested are irradiated with the ultraviolet spectrum by the water quality sensor to split different spectra, the spectrum acquisition module collects the spectral data of the CCD detector, and the spectral data is sent to the terminal through the wireless sensor network; and then uses the particle swarm algorithm (PSO) and The AdaBoost algorithm optimizes the BP neural network and establishes a water quality evaluation model. Both particle swarm optimization and AdaBoost algorithm can effectively solve the problem of unstable evaluation results caused by random initialization parameters of BP neural network. Under the same conditions, the AdaBoost algorithm has higher prediction accuracy and shorter training time than the PSO algorithm, and has good application value in water quality evaluation.

Description of drawings

Fig. 1 is the overall picture of the water quality detection system based on ultraviolet spectrum;

Figure 2 is a structural diagram of a water quality sensor based on ultraviolet spectrum;

Fig. 3 is the structure of the signal acquisition system;

Fig. 4 is the implementation structure of the water quality monitoring system based on ultraviolet spectrum;

Fig. 5 Water quality evaluation model based on AdaBoost-BP neural network;

Figure 6 shows the results of the AdaBoost-BP water quality evaluation model;

Figure 7 is a comparison chart of the results of the BP, PSO-BP and AdaBoost-BP evaluation models;

Figure 8 is a comparison chart of the error rate of the BP, PSO-BP and AdaBoost-BP evaluation models;

Figure 9 Dimensional drawing of the water quality sensor;

FIG. 10 is a structural diagram of a solar power supply system.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments. The specific embodiments described herein are only used to explain the present invention, but not to limit the invention.

As shown in Figure 1, the water quality sensor irradiates the water sample to be tested by ultraviolet spectrum to split different spectra, the spectrum acquisition module collects the spectral data of the CCD detector, and the spectral data is sent to the terminal through the wireless sensor network; Algorithm (PSO) and AdaBoost algorithm are used to optimize the BP neural network and establish a water quality evaluation model.

As shown in Figure 2, it is the structure diagram of the water quality sensor based on ultraviolet spectrum, the resolution is within 0.5nm and it works stably. The light is incident from the slit to the spherical collimating mirror M ₁ , and the parallel light collimated by the spherical mirror is incident on the grating. On G, light with different wavelengths is diffracted by the grating and then incident on the spherical focusing mirror M ₂ according to an angle, and the focused light is finally imaged on the surface of the CCD.

In the spectrum analyzer system, the parameters of each optical element are mutually restricted by the spatial position of the element and the aberration, so each parameter needs to be determined according to the spatial relationship of each element. Diffraction grating is the core device in the spectrum analyzer. The parameters of the grating directly affect the response wavelength and resolution of the optical system spectrum analyzer. The measuring range of the ultraviolet spectrum analyzer designed in this patent is 200-500nm. It is known that:

i+θ=Φ(1)

In the system, the incident ray and the diffracted ray are located on both sides of the normal, and the grating equation is:

d(sini-sinθ)=mλ (2)

In the formula, d=1/n is the grating constant, n is the grating line density, m is the diffraction order, and the first-order diffraction spectrum of the grating is taken, that is, m=1, λ is the wavelength, and the above formulas are combined to obtain:

The linear dispersion of the grating characterizes the distances separated by different wavelengths of light on the image plane after imaging, and is an important indicator for judging the resolution of the system in a spectrum analyzer system. The linear dispersion rate formula is:

为像面倾角。对上式在波长范围λ₁-λ₂积分，得到：where f ₂ is the focal length of the focusing mirror M ₁ ,

is the inclination angle of the image plane. Integrating the above formula over the wavelength range λ ₁ -λ ₂ , we get:

In the formula l, it is the effective length of the image plane CCD, and the effective length of the CCD selected by this system is 28.6mm.

Resolution is also an important indicator for judging system resolution.

Since both the collimating mirror and the focusing mirror are used off-axis, a certain coma aberration will be generated, which will reduce the resolution of the system and affect the imaging quality. Therefore, the aberration must be eliminated as much as possible. The system should meet the following requirements:

In the design of the whole system, in addition to the influence of coma aberration, spherical aberration will also reduce the resolution of the system. The system uses a concave spherical mirror as the collimating mirror. The focal length of the straight mirror should satisfy:

f ₁ ≤256λ(F#) ⁴ (7)

In the formula, F#=f ₁ /D ₁ is the object space F number of the spectrum analyzer, D ₁ is the aperture of the collimating mirror, and the aperture of the focusing mirror should satisfy:

In the above formula, θ ₂ is the diffraction angle of the stop wavelength, and θ ₂ can be obtained from the grating equation (2).

The length, incident angle and collimator aperture of the system grating need to meet:

In order to make the image quality good, the slit is placed at the meridional focus of the collimating mirror, which can be obtained from the geometric position:

As shown in Figure 3, a spectral data acquisition system based on STM32 is designed in the signal acquisition part. STM32F103 generates the CCD timing sequence to drive the CCD to work. After the CCD output is reversed and amplified by the signal conditioning circuit, the voltage signal output by the CCD is converted to 0 The voltage signal between ~ V _REF is stored in the memory after AD conversion of each pixel through A/D. When all the 3648 data in one frame are collected, it is transmitted to the host computer through the serial port for data processing, and the relative spectral intensity distribution curve is drawn.

In order to realize long-term online monitoring of water quality parameters, the present invention proposes an improved routing protocol algorithm based on LEACH. Improvements are proposed mainly in the cluster head selection stage: considering the average distance of the cluster head (CH) to all adjacent nodes, the distance between the base station and the CH, and the parameters of prolonging the transmission period, by selecting the one with larger residual energy and relatively shorter distance. The sensor nodes act as CH nodes.

In view of the shortcomings of the traditional LEACH protocol, the energy of each sensor node, the average distance between the node and all adjacent nodes, and the relative distance between the node and the base station are taken into account, and the consumption of control messages in the broadcast and scheduling stages is reduced. The formula _Topt (n) is improved. At this stage, for each sensor point, it generates a random number between 0 and 1. For this random number is less than the threshold _Topt (n), this node becomes a candidate cluster head.

r是已完成轮数；α是轮数延长参数，通过延长传输轮，我们通过减少广播和TDMA调度的消息次数来节省大量能量； G是最近1/p轮没有被选为簇头的集合；E_cur是当前传感器节点功率；E_avg是当前一轮网络的平均能量；d_toCH是簇头与簇内所有节点的平均距离；d_avg是特定节点与所有相邻节点的平均距离；d_toBS是传感器节点到BS的平均距离；d(n,BS)是给定节点n到BS之间的距离；d_max是最大距离。in,

r is the number of completed rounds; α is the round number extension parameter, by extending the transmission round, we save a lot of energy by reducing the number of messages for broadcast and TDMA scheduling; G is the set of recent 1/p rounds that have not been selected as cluster heads; E _cur is the current sensor node power; E _avg is the average energy of the current round of the network; d _toCH is the average distance between the cluster head and all nodes in the cluster; d _avg is the average distance between a specific node and all adjacent nodes; d _toBS is Average distance from sensor node to BS; d(n, BS) is the distance from a given node n to BS; _dmax is the maximum distance.

p represents the proportion of cluster heads, which is given by the following formula:

where k _opt is the optimal number of clusters and N is the number of sensor nodes in the network; where k _opt is given by:

The improved algorithm redefines d _toCH , d _toBS and d _max as follows [7]:

d _avg is calculated by considering the radius of the sensor node's neighborhood, nodes within this radius are considered to be the node's neighborhood, and then their distance from the node is calculated, d _avg is the average distance of all these neighborhood nodes :

where N′ is given by the set of neighboring sensor nodes for a given sensor node, and the neighborhood radius (R _ch ) is defined as:

Among them, M×M is the deployment area of sensor nodes. In addition, the algorithm selects the current energy of the sensor node as the CH node, which means that the greater the current node energy, the more likely it is to become the CH node in this round.

The implementation of a distributed water quality monitoring system based on ultraviolet spectrum provided by the present invention is shown in FIG. 4 . The system includes any of the above-mentioned water quality monitoring instruments based on ultraviolet spectrum, namely node 1 and node 2 in FIG. 4 until Node n, the wireless communication component within the wireless communication range with the water quality monitoring instrument, that is, the "wireless network" in Figure 4, and the water quality monitoring terminal (the personal computer and mobile phone on the right in Figure 4), where the wireless The communication part is used to receive the water quality data transmitted by the water quality monitoring instrument and send it to the water quality monitoring terminal. There is no limit to the number of such water quality monitoring instruments based on multi-wavelength absorbance, generally there are more than one. Such instruments can be set up in water bodies in different places as monitoring nodes in different locations, which can obtain real-time water quality data in different places. , The water quality monitoring instrument can use the 4G communication module to transmit the obtained water quality information, and then the water quality information is transmitted to the water quality monitoring terminal by the wireless network, and the energy-saving wireless sensor network routing algorithm proposed by the present invention can be used. Online monitoring, the monitoring personnel at the water quality monitoring terminal can control the water quality information of multiple locations in real time, so as to quickly analyze the water quality changes in different locations and make relevant decisions. The water quality monitoring terminal can be a mobile terminal, such as a mobile phone or A tablet computer, etc., can also be a PC terminal, etc., which is not limited here. Using a plurality of the above water quality monitoring instruments as monitoring nodes, it is possible to establish a node-based distributed monitoring system that can reflect the water pollution status online in real time, and can feedback the COD information of the water body online in real time, and can combine this water quality information with geographic location data. , The structure is simple and the hardware cost is reduced.

In order to facilitate the rating of water quality, the present invention regards the BP neural network as a weak classifier on the basis of the BP neural network, and uses AdaBoost to establish a strong classifier to establish a water quality evaluation model. It effectively solves the problem of unstable evaluation system caused by random initialization parameters of BP neural network.

According to the basic principle of AdaBoost algorithm, firstly, BP neural network is used to train and predict the training samples. When the output prediction error is greater than the predetermined error range, this sample is regarded as a sample that needs to be strengthened, and the weight of the training sample is adjusted to calculate the weight of the t-th BP weak predictor. After multiple trainings, the weights of T weak BP predictors are obtained. According to the weight distribution combination of each BP weak predictor, a strong predictor is formed, and the strong predictor is used for prediction, and the final prediction result is output.

As shown in Figure 5, the specific construction process of the AdaBoost-BP water quality evaluation model is as follows:

(1) Import data samples, determine training samples and test samples, and initialize the weights of training sample data. The calculation formula is:

where D _i is the initialization weight, i=1,2,...,m; m is the number of training samples.

(2) Set the number and network structure of BP weak classification, and use BP weak predictor to train and predict the training samples. Then get the weak classifier C _t (x).

(3) Calculate the classification error of C _t (x):

(4) Calculate the weights of C _t (x) weak classifiers:

(5) Update the training data weights:

where _gt is the normalization factor and _yi is the data label.

(6) The final strong classifier:

Water quality evaluation involves 5 criteria, and 5 BP neural networks are used to form a weak classification group. On the basis of BP neural network, the strong classifier designed by AdaBoost can effectively improve the stability of a single BP neural network evaluation system, and improve its classification accuracy and generalization ability. Figure 5 is a schematic diagram of the AdaBoost evaluation model.

Comparing the results of the BP neural network with the results of the AdaBoost-BP water quality evaluation model, it can be seen from Figure 6 that on the whole, the recognition rate of the BP neural network is significantly lower than that of the AdaBoost-BP neural network. The BP neural network is affected by random initialization parameters. It can be seen from Figure 7 that the recognition error of the AdaBoost-BP neural network is significantly smaller than that of the traditional BP neural network, and the AdaBoost-BP system is more stable. Therefore, the BP neural network water quality evaluation model optimized by AdaBoost can meet the requirements of practical applications. Then the experimental results of the BP neural network water quality evaluation model and the PSO-BP neural network water quality evaluation model are compared. It can be seen from Figure 6 and Figure 7 that the experimental results of the PSO-BP neural network are better than the experimental results of the BP neural network.

By calculating the error mean and variance of the AdaBoost-BP evaluation model and the PSO-BP evaluation model, it is found that the mean and variance of the errors of the AdaBoost-BP model are 1.62% and 0.0081, respectively, while the mean and variance of the errors of the PSO-BP model are 3.44%, respectively. , 0.0112. Compared with the PSO-BP water quality evaluation model, the AdaBoost-BP water quality evaluation model has higher accuracy and a more stable system, which can better meet the requirements of practical applications.

The water quality evaluation device consists of a combination unit and an acquisition unit. The combining unit is used for combining the obtained various water quality index values into comprehensive index values using the principal component analysis method; the obtaining unit 2 is used for obtaining water quality evaluation results according to the evaluation method AdaBoost-BP water quality evaluation model.

Specifically, the water quality index includes physical properties and chemical properties of water quality, and the water quality index value may be the value of the water quality index collected in a preset period of time with a preset period of time, for example, a period of every month within four years of collection 23 water quality index values. Considering the relationship between various water quality indicators, the principal component analysis method is used to combine various water quality indicators into comprehensive indicators, and the obtained values of various water quality indicators are combined into comprehensive indicators. Principal component analysis is a multivariate mathematical statistical method mainly used for dimensionality reduction, which converts multiple water quality indicators into a few comprehensive indicators. There is a correlation between the water quality indicators, and the principal component analysis method is used to convert a group of related water quality indicators into another group of irrelevant indicators through linear change, and the total variance of the water quality indicators is preserved during the conversion process. Change. Arrange the transformed water quality indexes in the order of decreasing variance, wherein the first transformed water quality index has the largest variance, which is called the first principal component, the second principal component has the second largest variance, and is the same as the first principal component. The principal components are irrelevant, and so on.

The spectral data of the signal acquisition and processing module is used as the input of the AdaBoost-BP water quality evaluation model to obtain the water quality evaluation results. The AdaBoost-BP algorithm combines the advantages of the BP neural network and the AdaBoost algorithm to improve the accuracy and speed up the training. The present invention utilizes the AdaBoost algorithm to construct a strong classifier through the BP neural network, avoids the problems of slow convergence speed and premature maturity of the BP neural network, effectively improves the defects of the BP neural network, and improves the global search ability. By comparing the prediction results of PSO-BP neural network and AdaBoost-BP neural network, it is found that the AdaBoost-BP neural network evaluation model has higher accuracy, stability and better application value.

Fig. 9 is a dimension drawing of the water quality sensor. The internal structure of the water quality sensor is mainly a miniature ultraviolet spectrum analyzer, and it is a spectrum analyzer with high spectral resolution and absolute value measurement function of light intensity. Its key technologies are as follows:

(1) Optical-mechanical structure design. The spectrum analyzer adopts new devices such as optical fiber and CCD, which improves the flexibility of spectrum analyzer application, and can realize real-time multi-channel array photoelectric reception. Its structure is designed with a symmetrical Czemy-Tumer optical structure. The incident light is coupled to a standard SMA905 interface by an optical fiber and enters the optical system. It is collimated by a spherical mirror, and then split by a plane grating. on the detector array.

(2) Selection of grating. The grating not only determines the working wavelength band of the spectrum analyzer. And it directly affects the spectral resolution of the system. The selection of the grating should be determined according to the working band of the spectrum analyzer. Under the spectral resolution required by the design, select the appropriate grating constant. According to the overall design requirements, the possible working band of the miniature spectrum analyzer is 230-400nm, covering the ultraviolet visible and near-infrared spectral regions.

(3) Determination of collimating lens parameters. When using a concave spherical mirror as an objective lens, the spherical aberration must be controlled within the aberration tolerance to ensure the spectral resolution of the system. In the initial design, the Rayleigh criterion, in which the wave aberration generated by spherical aberration is smaller than that, can be used to determine its focal length and allowable system F-number. The wave aberration generated by the spherical aberration of the collimating mirror should be comprehensively considered when determining the relative aperture of the system and the parameters of the collimating mirror according to the Rayleigh criterion. system to adjust.

(4) Determination of incident fiber and slit. When selecting the incident fiber, the following parameters and factors need to be considered: the interface type of the incident fiber, the fiber material, and the size of the fiber core diameter. There are many commonly used fiber interfaces, such as FC, SC, ST, and SMA. The size of the fiber core determines the incident light intensity of the system. In a system with a slit, the spectral resolution of the system is determined by the width of the incident slit. At this time, if the size of the fiber core diameter is increased within the aberration tolerance, it will not only not affect the resolution of the system, but also increase the size of the optical fiber. The height of the effective slit increases the incident light intensity of the system. The numerical aperture of the optical fiber selected for this sensor is 0.22.

FIG. 10 is a structural diagram of a solar power supply system. The invention designs a solar power supply system for the water quality sensor, which can ensure that each electric component of the water quality detection system can operate lastingly and effectively.

(1) Power supply system. The power supply of each equipment of the water quality detection system is DC power supply, and the electric energy stored in the battery by photovoltaic cells is also DC power. Therefore, in this system, the battery can be directly connected to the load for power supply. If the equipment is damaged, the equipment needs to operate at the rated voltage during normal use, so a battery needs to be equipped between the solar cell and the equipment. Here we choose valve lead-acid batteries. The valve-regulated lead-acid battery is relatively cost-effective, saving the computing cost of the system, and does not require human control in function, which is very convenient for intelligent operation. However, due to the requirements of the internal environment of the valve lead-acid battery, its charging requirements are very high, otherwise its service life will be seriously reduced. Therefore, it is necessary to design a complete intelligent charging and discharging system to ensure that the battery can be quickly charged under the allowable voltage and current level. Since the rated voltage of the solenoid valve and the signal lamp is 12-24V, the voltage level of the battery used in this system is 12V and the capacity is 65AH, and the battery can directly supply power to it. The rated voltage of the water quality sensor is 24V, so a booster module should be added between the battery and the sensor for boosting.

(2) Energy conversion. The energy conversion circuit is the main functional module of the system. In this energy conversion circuit, the voltage of the solar panel needs to be stabilized within a certain range by bucking or boosting, and in this system, the voltage is stabilized at 24V. The main body of energy conversion is a buck-boost buck-boost chopper circuit. The input end of the circuit is from the solar panel, and the output end of the circuit is connected to the battery, which is equivalent to a load.

There is an ADC acquisition signal between the solar panel and the DC-DC, and there is also an ADC acquisition point between the DC-DC and the battery. When the voltage measured by the first ADC is too low, disconnect the power connection between the solar panel and the DC-DC. When there is a gap between the voltage value measured by the second ADC and the predetermined 24V, the on-off PWM duty cycle of the MOSFET is controlled.

To sum up, the above solution provided by the present invention does not need to add chemical reagents, has no secondary pollution, has fast detection speed, low power consumption, and can realize the drop-in online measurement function, and the monitoring node has a simple structure and strong scalability. Follow-up functions such as power, obstacle avoidance, and data inspection can be expanded. The instrument can feed back the pollution degree and geographic location information of the water body in real time, and can form a node-type real-time water quality monitoring system to achieve the effect of water pollution early warning; the COD measurement pollution optical window compensation is adopted. The method is simple and easy to operate, and the measurement results after compensation are highly accurate; the above-mentioned optical window pollution compensation method has good applicability under turbidity conditions and is suitable for various types of water environments, and this optical window pollution compensation method for COD measurement In the subsequent measurement process, there is no need to re-invest in manpower and material resources, which can effectively reduce the maintenance cost of the instrument, so long-term real-time measurement of COD can be realized.

The technical solutions of the present invention are not limited to the limitations of the above-mentioned specific embodiments, and all technical modifications made according to the technical solutions of the present invention fall within the protection scope of the present invention.

Claims (5)

1. a distributed water quality detection system based on ultraviolet spectroscopy, is characterized in that, comprises several water quality sensors based on ultraviolet spectrum, the described water quality sensor based on ultraviolet spectrum is connected with water quality monitoring terminal by wireless sensor network; The water quality sensor of ultraviolet spectrum includes a light source that emits ultraviolet spectrum, a spectrum analyzer, an array CCD detector, a signal acquisition and processing module, and a power supply; the ultraviolet spectrum emitted by the light source is transmitted through the water source to be tested to become signal light, and enters the spectrum analyzer to obtain Spectra arranged in order of different wavelengths, the array CCD detector converts the optical signal irradiated on the detector into an electrical signal, the signal acquisition and processing module collects the signal of the array CCD detector, and the spectral data passes through the wireless sensor network. Upload the water quality monitoring terminal for water quality evaluation, and obtain the composition and content information of the measured water quality. 2. The distributed water quality detection system based on ultraviolet spectroscopy according to claim 1, characterized in that: the signal acquisition and processing module generates CCD timing sequence by STM32F103 single-chip microcomputer, drives the array CCD detector to work, and the array CCD detector outputs the After the signal conditioning circuit is reversed and amplified, the voltage signal output by the array CCD detector is converted into a voltage signal between 0 and V _REF , and each pixel is AD converted by the ADC conversion circuit and stored in the memory. After all the data is collected, it is transmitted to the water quality monitoring terminal through the wireless network. 3. The distributed water quality detection system based on ultraviolet spectroscopy according to claim 1 or 2, wherein the water quality monitoring terminal is a mobile phone or a tablet computer or a PC terminal. 4 . The distributed water quality detection system based on ultraviolet spectroscopy according to claim 1 , wherein the power source is a 12V battery, the battery is connected with a photovoltaic charging device, and the photovoltaic charging device comprises a solar panel. 5 . , There is a first ADC conversion circuit between the DC-DC of the solar panel and the battery, and a second ADC conversion circuit between the DC-DC and the battery, when the voltage measured by the first ADC conversion circuit is too low , then disconnect the power connection between the solar panel and the DC-DC; when there is a gap between the voltage value measured by the second ADC conversion circuit and the predetermined 24V, it is necessary to control the switching PWM duty cycle of the MOSFET . 5. a water quality assessment classification method, is characterized in that, using the distributed water quality detection system based on ultraviolet spectroscopy as claimed in claim 1, concrete steps are as follows: S1. Establish the AdaBoost-BP water quality evaluation model in the water quality monitoring terminal, S2. Import data samples, determine training samples and test samples, and initialize the weight of training sample data. The initial weight formula is:

Among them: D _i is the initialization weight, i=1,2,...,m; m is the number of training samples; S3. Set the number and network structure of the BP weak classification, and use the BP weak predictor to train and predict the training samples, and then obtain the weak classifier C _t (x); S4. Calculate the classification error C _t (x):

S5. Calculate the weight of the C _t (x) weak classifier:

S6. Update the training data set weights:

where: g _t is the normalization factor, and _yi is the data label; S7, the final strong classifier:

Water quality evaluation involves 5 criteria, and 5 BP neural networks are used to form a weak classification group. On the basis of BP neural network, the strong classifier designed by AdaBoost can effectively enhance the stability of a single BP neural network and improve the BP neural network evaluation model. accuracy and generalization ability; S8. Use the spectral data output by the signal acquisition and processing module as the input of the AdaBoost-BP water quality evaluation model to predict the water quality grade evaluation results.

Images