Multi-point water quality intelligent monitoring method and big data Internet of Things system

Abstract

The invention relates to the technical field of multi-point water quality automatic monitoring, and discloses a multi-point water quality intelligent monitoring method and a big data Internet of Things system, a plurality of temperature sensors, a plurality of TDS sensors, a plurality of turbidity sensors and a plurality of PH value sensors are installed at multiple points of a water quality area to be detected, and a plurality of sensor values are obtained respectively; after the SOM neural network models 1-4 are used for processing, the water quality information detection module is used for correcting and predicting, finally, the multi-point water quality monitoring subsystem is used for intelligently predicting the water quality of a detected object, and information interaction among the water quality measuring end, the water quality remote monitoring end and the water quality monitoring mobile phone APP end is realized through the water quality monitoring cloud platform. And the water quality monitoring mobile phone APP end and the water quality remote monitoring end access the water quality monitoring cloud platform through a 5G network to realize water quality supervision. Compared with the prior art, intelligent detection and management of water quality are realized, and the modernization level of water quality monitoring and management can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-point water quality automatic monitoring, and particularly to a multi-point water quality intelligent monitoring method and a big data Internet of Things system. Background Art

[0002] Water is the source of life and the main material basis for the structural composition and life activities of all organisms. Water resources are irreplaceable resources for human survival and social development. Strengthening the real-time monitoring of water quality and enhancing the precise monitoring of the ecological environment are of utmost importance. However, in the existing technology, the accuracy and intelligence level of rapid water quality determination devices are low. When water quality parameters are too high or too low, it seriously affects product quality and economic efficiency improvement. The multi-point water quality intelligent monitoring method and its big data Internet of Things system developed in this patent can upload the output data of multi-point water quality detection sensors to the Alibaba Cloud, and perform real-time refined management of water quality on the Alibaba water quality monitoring cloud platform. Summary of the Invention

[0003] Object of the Invention: Apply Internet of Things technology, intelligent control technology, and water quality monitoring cloud platform to water quality management, construct a multi-point water quality intelligent monitoring method and a big data Internet of Things system to realize intelligent detection and management of water quality, and effectively improve the modernization level of water quality monitoring and management.

[0004] Technical Solution: The present invention discloses a multi-point water quality intelligent monitoring method, which includes the following steps:

[0005] Step 1: Install multiple temperature sensors, multiple TDS sensors, multiple turbidity sensors, and multiple pH value sensors at multiple points in the water quality area to be detected, and respectively obtain multiple sensor values;

[0006] Step 2: Input the detection values of the multiple temperature sensors, multiple TDS sensors, multiple turbidity sensors, and multiple pH value sensors into SOM neural network models 1 to 4 respectively to obtain corresponding 1-C types of temperature sensor values, 1-C types of TDS sensor values, 1-C types of turbidity sensor values, and 1-C types of pH value sensor values;

[0007] Step 3: Respectively input the 1-C types of temperature sensor values, 1-C types of TDS sensor values, 1-C types of turbidity sensor values, and 1-C types of pH value sensor values obtained in Step 2 into the water quality information detection module for sensor value correction prediction to obtain the correction prediction values of the temperature sensor, TDS sensor, turbidity sensor, and pH value sensor;

[0008] Step 4: The corrected predicted values of multiple temperature sensors, the corrected predicted values of multiple TDS sensors, the corrected predicted values of multiple turbidity sensors, and the corrected predicted values of multiple pH sensors are respectively input into the ESN neural network models 1-4 of the Transformer model-WOA for data processing;

[0009] Step 5: The output values of the ESN neural network models 1-4 of the Transformer model-WOA are used as the input of the ESN neural network model 2 of the AANN neural network model-WOA with interval dynamic fuzzy numbers, and the output values of the ESN neural network models 1-4 of the Transformer model-WOA are respectively used as the input of the other three ESN neural network models of the Transformer model-WOA for decoupling calibration;

[0010] Step 6: The output of the ESN neural network model 2 of the AANN neural network model-WOA with interval dynamic fuzzy numbers is fed back to the ESN neural network models 1-4 of the Transformer model-WOA after passing through the TDL clock delay device. The output of the ESN neural network model 2 of the AANN neural network model-WOA with interval dynamic fuzzy numbers outputs 5 interval dynamic fuzzy numbers to divide the water quality of the detected area into five levels.

[0011] Furthermore, there are 4C water quality information detection modules. Among them, C water quality information detection modules are used to process 1-C types of temperature sensor values output by the SOM neural network model 1, C water quality information detection modules are used to process 1-C types of TDS sensor values output by the SOM neural network model 2, C water quality information detection modules are used to process 1-C types of turbidity sensor values output by the SOM neural network model 3, and C water quality information detection modules are used to process 1-C types of pH sensor values output by the SOM neural network model 4.

[0012] Furthermore, the water quality information detection module includes the BiGRU neural network model of the AANN neural network model-WOA, the TCN neural network model of the Transformer model-WOA, the BiGRU neural network model of the Transformer model-WOA, the Transformer-ARIMA model, the ESN neural network model-WOA's BiGRU neural network models 1-4, and the ESN neural network model 1 of the AANN neural network model-WOA with interval dynamic fuzzy numbers;

[0013] The outputs of the parameter sensors are respectively used as the inputs of the BiGRU neural network model of the AANN neural network model - WOA, the TCN neural network model of the Transformer model - WOA, the BiGRU neural network model of the Transformer model - WOA, and the Transformer model - ARIMA model. The differences between the outputs of the TCN neural network model of the Transformer model - WOA and the outputs of the ESN neural network model of WOA - BiGRU neural network model 4 are respectively used as the inputs of the ESN neural network model of WOA - BiGRU neural network model 1 and the AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 1 of WOA. The sums of the outputs of the ESN neural network model of WOA - BiGRU neural network model 1 and the outputs of the ESN neural network model of WOA - BiGRU neural network model 4 are respectively used as the inputs of the TCN neural network model of the Transformer model - WOA and the AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 1 of WOA;

[0014] The differences between the outputs of the BiGRU neural network model of the Transformer model - WOA and the outputs of the ESN neural network model of WOA - BiGRU neural network model 4 are respectively used as the inputs of the ESN neural network model of WOA - BiGRU neural network model 2 and the AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 1 of WOA. The sums of the outputs of the ESN neural network model of WOA - BiGRU neural network model 2 and the outputs of the ESN neural network model of WOA - BiGRU neural network model 4 are respectively used as the inputs of the BiGRU neural network model of the Transformer model - WOA and the AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 1 of WOA;

[0015] The differences between the outputs of the Transformer model - ARIMA model and the outputs of the ESN neural network model of WOA - BiGRU neural network model 4 are respectively used as the inputs of the ESN neural network model of WOA - BiGRU neural network model 3 and the AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 1 of WOA. The sums of the outputs of the ESN neural network model of WOA - BiGRU neural network model 3 and the outputs of the ESN neural network model of WOA - BiGRU neural network model 4 are respectively used as the inputs of the Transformer model - ARIMA model and the AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 1 of WOA;

[0016] The AANN neural network model of interval dynamic fuzzy numbers - the output of the ESN neural network model 1 of WOA and the output of the BiGRU neural network model of the AANN neural network model - WOA are used as the inputs of the ESN neural network model - BiGRU neural network model 4 of WOA. The five parameters a, b, u, v, and q of the output of the ESN neural network model 1 of the AANN neural network model of interval dynamic fuzzy numbers form the interval dynamic fuzzy numbers of the parameters to be detected as [(a, b); u; (v, q)], where a, b, u, v, and q respectively represent the minimum value, minimum extreme value, mean value, maximum extreme value, and maximum value of the predicted value of the parameter sensor output.

[0017] Furthermore, in the step 5, the connection relationship between the Transformer model - ESN neural network model 1 - 4 of WOA and the AANN neural network model - ESN neural network model 2 of interval dynamic fuzzy numbers is as follows:

[0018] The output of the Transformer model - ESN neural network model 1 of WOA is used as the input of the AANN neural network model - ESN neural network model 2 of interval dynamic fuzzy numbers, the Transformer model - ESN neural network model 2 of WOA, the Transformer model - ESN neural network model 3 of WOA, and the Transformer model - ESN neural network model 4 of WOA;

[0019] The output of the Transformer model - ESN neural network model 2 of WOA is used as the input of the AANN neural network model - ESN neural network model 2 of interval dynamic fuzzy numbers, the Transformer model - ESN neural network model 1 of WOA, the Transformer model - ESN neural network model 3 of WOA, and the Transformer model - ESN neural network model 4 of WOA;

[0020] The output of the Transformer model - ESN neural network model 3 of WOA is used as the input of the AANN neural network model - ESN neural network model 2 of interval dynamic fuzzy numbers, the Transformer model - ESN neural network model 1 of WOA, the Transformer model - ESN neural network model 2 of WOA, and the Transformer model - ESN neural network model 4 of WOA;

[0021] The output of the ESN neural network model 4 of the Transformer model - WOA serves as the input to the AANN neural network model - WOA of the ESN neural network models 2, 1, 2, and 3 of the Transformer model - WOA.

[0022] Furthermore, in step 6, the output of the AANN neural network model - WOA of the ESN neural network model 2 of the interval dynamic fuzzy number serves as the input to the TDL beat delay device. The outputs of the TDL beat delay device respectively serve as the corresponding inputs to the ESN neural network models 1, 2, 3, and 4 of the Transformer model - WOA. The output of the AANN neural network model - WOA of the ESN neural network model 2 of the interval dynamic fuzzy number outputs 5 interval dynamic fuzzy numbers to classify the water quality of the detected area into five grades.

[0023] Furthermore, the relationship between the interval dynamic fuzzy number in the water quality monitoring result output by the AANN neural network model - WOA of the ESN neural network model 2 and the safety level is as follows:

[0024] Serial number Safety level Interval dynamic fuzzy number 1 Three categories [(0.45，0.55)；0.7；(0.8，0.95)] 2 Two categories [(0.55，0.75)；0.9；(0.95，1)] 3 One category [(0.85，0.95)；1；(1，1)] 4 Four categories [(0.25，0.35)；0.5；(0.65，0.75)] 5 Five categories [(0，0.15)；0.3；(0.45，0.55)] 。

[0025] Furthermore, the AANN neural network model - WOA's ESN neural network model 1 and the AANN neural network model - WOA's ESN neural network model 2 of the interval dynamic fuzzy numbers have the same structure, both of which are the series connection of the Transformer model and the WOA's ESN neural network model; the AANN neural network model - WOA's BiGRU neural network model is the series connection of the AANN neural network model and the WOA's BiGRU neural network model; the Transformer model - ARIMA model is the series connection of the Transformer model and the ARIMA model; the Transformer model - WOA's BiGRU neural network model is the series connection of the Transformer model and the WOA's BiGRU neural network model; the Transformer model - WOA's TCN neural network model is the series connection of the Transformer model and the WOA's TCN neural network model; the WOA's ESN neural network model - WOA's BiGRU neural network models 1 - 4 have the same structure, all of which are the series connection of the WOA's ESN neural network model and the WOA's BiGRU neural network model.

[0026] Furthermore, the Transformer model - WOA's ESN neural network models 1 - 4 have the same structure, all of which are the series connection of the Transformer model and the WOA's ESN neural network model.

[0027] The present invention also discloses a multi - point water quality intelligent monitoring big data Internet of Things system, including a water quality measurement end, a water quality monitoring cloud platform, a water quality remote monitoring end, and a water quality monitoring mobile APP end. The water quality measurement end realizes real - time detection of the water quality parameters to be detected. The water quality remote monitoring end is equipped with an intelligent multi - point water quality monitoring subsystem. The intelligent multi - point water quality monitoring subsystem realizes the above - mentioned multi - point water quality intelligent monitoring method, realizes intelligent prediction of the water quality of the object to be detected, and realizes information interaction among the water quality measurement end, the water quality remote monitoring end, and the water quality monitoring mobile APP end through the water quality monitoring cloud platform. The water quality monitoring mobile APP end and the water quality remote monitoring end access the water quality monitoring cloud platform through the 5G network to realize water quality supervision.

[0028] Compared with the prior art, the present invention has the following obvious advantages:

[0029] 1. The SOM neural network model designed by the present invention is a classification method for water quality parameters. Its purpose is to divide a set of data in the water quality parameter data space output by the water quality parameter sensor into several subsets according to the similarity criterion, so that each subset of the water quality parameter sensor output value represents a certain feature of the data sample set of the entire water quality parameter. Establishing the SOM neural network model to classify the output value of the water quality parameter sensor is to find a reasonable sample subset division of the water quality parameter, and input the characteristics of different subsets of water quality parameters into the corresponding water quality information detection module to improve the accuracy of water quality parameter prediction.

[0030] 2. ESN neural network model 1 to 4 of the Transformer model-WOA of the present invention designs the network hidden layer into a sparse network of water quality parameters composed of many neurons, and achieves the function of memorizing water quality parameters by adjusting the characteristics of the internal weights of the network. Its internal dynamic reserve pool contains a large number of sparsely connected neurons of water quality parameters, which contain the operating status of the system and have the function of short-term memory of water quality parameters. The stability of the recursive network inside the reserve pool is guaranteed by presetting the spectral radius of the internal connection weight matrix of the ESN neural network model, thereby improving the stability and accuracy of predicting water quality parameters.

[0031] 3. The current state of the reserve pool of the ESN neural network model of the present invention inherits the state of the previous moment, and has a short memory characteristic for the historical data of water quality parameters. The research results show that the ESN neural network model with historical memory has a better prediction effect on water quality parameters. The ESN neural network model has the characteristics of high precision, high accuracy, high timeliness and stability of water quality parameters, and can be used as a means of quickly and effectively predicting water quality parameters; as a new type of dynamic recursive neural network, the ESN neural network model uses a linear regression method to establish a model, avoiding the problems of slow convergence speed and easy to fall into local minima of traditional neural networks, simplifying the complexity of the training process, and achieving the purpose of efficiently predicting water quality parameters.

[0032] IV. The AANN neural network model - WOA BiGRU neural network model designed in the present invention outputs a series connection of the AANN auto - associative neural network model and the WOA BiGRU neural network model. The AANN auto - associative neural network model compresses and reconstructs the output data of the water quality parameter sensor non - linearly. The AANN auto - associative neural network model will learn the non - linear relationship between the output variables of the water quality parameter sensor, solidify the characteristics and associations among the output data of the water quality parameter sensor into the model, and seek the network parameters of the optimal AANN auto - associative neural network model with the goal of minimizing the input - output error, so as to make the output restore the output data of the water quality parameter sensor as much as possible and filter out the interference in the output of the water quality parameter sensor. The WOA BiGRU neural network model outputs the time - series values of water quality parameters. This AANN neural network model - WOA BiGRU neural network model improves the accuracy of water quality parameter prediction and the characteristics of time - series changes.

[0033] V. The Transformer model - ARIMA model integrates factors such as the trend factor, cycle factor, and random error of water quality parameters, transforms the non - stationary water quality parameter sequence into a zero - mean stationary random sequence through methods such as differential water quality parameter conversion, and conducts fitting and prediction of water quality parameters through repeated identification and model diagnosis comparison. This ARIMA model combines the advantages of autoregressive and moving average methods, has the characteristics of being not restricted by the type of water quality parameters and having strong applicability, and is a model with good trend prediction effect for water quality parameters.

[0034] VI. In view of the uncertainty and randomness existing in the measurement process of water quality parameters due to problems such as the accuracy error, interference, and abnormal measurement temperature value of the water quality parameter sensor, the present invention patent converts the measured value of the water quality parameter sensor into an interval intuitionistic number form through the water quality information detection module, effectively deals with the fuzziness and uncertainty of the measured parameters of the water quality parameter sensor, and improves the objectivity and credibility of the water quality parameter sensor value.

[0035] VII. The water quality information detection module extracts the spatial eigenvalue, time series value, and trend value output by the water quality parameter sensor through the TCN neural network model of Transformer-WOA, the BiGRU neural network model of Transformer-WOA, and the Transformer-ARIMA model respectively. The outputs of the TCN neural network model of Transformer-WOA, the BiGRU neural network model of Transformer-WOA, and the Transformer-ARIMA model are subtracted from the output of the ESN neural network model of WOA - the BiGRU neural network model 4 of WOA to obtain the spatial change amount, time series change amount, and trend change amount of the water quality parameter sensor output. The ESN neural network model of WOA - the BiGRU neural network models 1-3 of WOA respectively predict the spatial change amount, time series change amount, and trend change amount. The outputs of the ESN neural network model of WOA - the BiGRU neural network models 1-3 of WOA are added to the output of the ESN neural network model of WOA - the BiGRU neural network model 4 to obtain the spatial feature prediction value, time series prediction value, and trend prediction value of the water quality parameter sensor output over a period of time. The input of the AANN neural network model of interval dynamic fuzzy number - the ESN neural network model 1 of WOA is the spatial change amount, time series change amount, and trend change amount of the water quality parameter sensor output and the spatial feature prediction value, time series prediction value, and trend prediction value, which perform multi-angle prediction on the output of the temperature sensor, improving the comprehensiveness, accuracy, dynamics, and robustness of predicting the output parameters of the water quality parameter sensor.

[0036] VIII. The intelligent multi-point water quality monitoring subsystem classifies the outputs of 4 types of water quality parameter sensors through the SOM neural network models 1-4 respectively. The outputs of different types of water quality parameter sensors are used as the inputs of the corresponding water quality information detection modules, improving the accuracy of the water quality parameter sensor output. The first dynamic decoupling calibration prediction is performed on the water quality parameters through the ESN neural network models 1-4 of Transformer-WOA. The second dynamic feedback evaluation of the water quality level is performed through the AANN neural network model of interval dynamic fuzzy number - the ESN neural network model 2 of WOA. The secondary dynamic prediction and evaluation improve the dynamic performance, robustness, and accuracy of the detected water quality. Description of the Drawings

[0037] Figure 1 It is the structural block diagram of the water quality information detection module of the present invention;

[0038] Figure 2 It is the structural block diagram of the intelligent multi-point water quality monitoring subsystem of the present invention;

[0039] Figure 3Block diagram of the multi-point water quality intelligent monitoring big data Internet of Things system of the present invention;

[0040] Figure 4 Structural diagram of the water quality measurement end of the present invention;

[0041] Figure 5 Software of the water quality remote monitoring end of the present invention. Specific implementation manners

[0042] Combined with the attached Figures 1 - 5 , the technical solution of the present invention is further described:

[0043] I. Design of the water quality information detection module

[0044] The water quality information detection module includes the BiGRU neural network model of the AANN neural network model - WOA, the TCN neural network model of the Transformer model - WOA, the BiGRU neural network model of the Transformer model - WOA, the Transformer model - ARIMA model, the ESN neural network model of WOA - the BiGRU neural network model 1 - 4, and the AANN neural network model of interval dynamic fuzzy numbers - the ESN neural network model 1 of WOA. The structure and function of the water quality information detection module are shown in Figure 1 Shown. Among them: the TCN neural network model of WOA, the BiGRU neural network model of WOA, and the ESN neural network model of WOA are respectively the TCN neural network model, the BiGRU neural network model, and the ESN neural network model optimized by the whale optimization algorithm (WOA).

[0045] 1. Design of the BiGRU neural network model of the AANN neural network model - WOA

[0046] AANN neural network model - WOA's BiGRU neural network model is a series connection of AANN neural network model and WOA's BiGRU neural network model. AANN neural network model is an auto-associative neural network model. In AANN neural network model, the process of water quality parameter sensor output passing through input layer, mapping layer to bottleneck layer belongs to the encoding process of water quality parameters, and then passing through bottleneck layer to mapping layer and finally to output layer belongs to the decoding process of water quality parameters. First, the high-dimensional water quality data output by water quality parameter sensor is nonlinearly mapped to bottleneck layer through input layer, so as to realize the compression of high-dimensional water quality data and extract the feature dimension of effective water quality parameters. The hidden layer is mapped to the output layer with the same water quality parameter dimension as input layer through nonlinear function by bottleneck layer to realize the decoding of water quality parameter sensor output, that is, the reconstruction of water quality data input by water quality parameter sensor. In this way, through the nonlinear compression and reconstruction of the water quality parameter sensor output data, the AANN neural network model will learn the nonlinear relationship between the various water quality parameter variables output by the water quality parameter sensor. For the input of the water quality parameter sensor output data, the AANN auto-associative neural network model solidifies the characteristics and associations between the water quality parameter sensor output data into the model, with the goal of minimizing the input and output errors, and seeks the optimal water quality network parameters of the AANN auto-associative neural network model, so that the output can restore the water quality parameter sensor output data as much as possible and filter out the interference of the water quality parameter sensor output.

[0047] The BiGRU neural network model is composed of two positive and negative gated recurrent unit GRU neural network models. The GRU neural network model is a simplified version of the LSTM neural network model. The LSTM neural network model has an input gate, a forget gate, and an output gate, while the GRU neural network model only has an update gate and a reset gate. Therefore, it has a faster water quality parameter training speed. The GRU neural network model can solve the gradient explosion and gradient disappearance problems of water quality parameters. Therefore, using the GRU neural network model to extract the time series features of water quality parameters is a more common way at present. The GRU neural network model belongs to a unidirectional water quality parameter prediction model. The water quality parameter information is input from front to back, which often misses the information of many long-correlated water quality parameter points. The BiGRU neural network model contains two GRU neural network models. One GRU neural network model scans the entire water quality parameter time series from front to back, and the other GRU neural network model scans the water quality parameter time series from back to front. The complementary positive and negative GRU neural network models can establish the connection between the current input water quality parameters and the water quality parameters of the previous and next states, so as to more accurately capture the characteristic information in the input water quality sequence data and improve the water quality parameter prediction performance of the BiGRU neural network model.

[0048] The BiGRU neural network model is a neural network structure for processing time series data of water quality parameters. First, the time series sensor output of the water quality parameters output by the AANN autoassociative neural network model is input into the BiGRU neural network model. The BiGRU neural network model processes the spatial features of the water quality parameters and models the time series values of the spatial features output by the water quality parameter sensor from both forward and backward perspectives, to more comprehensively understand the dependence relationship between the spatial features output by the water quality parameter sensor, thereby improving the accuracy of the prediction of the water quality parameter sensor output; the BiGRU neural network model processes the time series data of the water quality parameters output by the water quality parameter sensor, captures long-term dependence relationships and patterns, and can adapt to different lengths of time series of water quality parameters, providing a reliable guarantee for processing the time series data of the water quality parameter sensor output and better predicting the water quality parameter sensor output.

[0049] 2. Transformer Model - ARIMA Model Design

[0050] Transformer Model - ARIMA Model: The Transformer model and the ARIMA model are connected in series. The Transformer model is mainly composed of the self - attention mechanism for water quality parameters, layer normalization, residual connection, and multi - layer perceptron. The core of the Transformer model is the self - attention mechanism for water quality parameters, which globally encodes the features of the input water quality parameters, enabling the model to complete the global feature modeling of water quality parameters and improving the comprehensiveness of extracting water quality parameter feature values. The introduction of the multi - layer perceptron enables the Transformer model to have a stronger non - linear feature extraction ability for the output values of water quality parameter sensors. The residual connection makes the mapping function of water quality parameters easier to fit when the number of model layers increases, greatly alleviating the problems of gradient disappearance and gradient explosion of water quality parameters when the number of model layers increases. Layer normalization normalizes the features of water quality parameters, accelerating the convergence speed of the model. The input of the ARIMA model is the output of the Transformer model. Due to the comprehensive influence of factors such as trends, seasons, and random fluctuations in water quality parameter time - series data, the non - stationary water quality parameter time - series data is made stationary. The dependent variable of water quality parameters, its lag values, and the random error term of water quality parameters are used for regression processing, and finally, the ARIMA model is established. The modeling steps mainly include the processing of the water quality time - series data output by the Transformer model, model identification and order determination, and model testing. The specific content is as follows: Step 1: Judgment and processing of the water quality parameter sequence. According to the water quality data sequence values, the stationarity of the water quality parameter time - series output by the Transformer model is tested. If the time - series is a stationary sequence, it meets the direct modeling requirements of ARIMA. If the time - series is non - stationary, the water quality parameter values output by the Transformer model need to be differenced until the autocorrelation function values and partial autocorrelation function values of the processed water quality parameters are not significantly different from zero. Step 2: Model identification and order determination. The ARIMA model is generally denoted as ARIMA(p, d, q), where p represents the number of autoregressive terms, d represents the number of differencing times, and q represents the number of moving average terms. Step 3: Testing the ARIMA model. It is required that there is no autocorrelation in the residuals of the water quality parameters of the model. Therefore, a hypothesis test needs to be conducted on the model to diagnose whether the residual sequence of water quality parameters is white noise. If the test is qualified, the model can be used to predict the output of the Transformer model.

[0051] 3. Design of the BiGRU Neural Network Model of Transformer Model - WOA

[0052] The Transformer model - the BiGRU neural network model of WOA is the series connection of the Transformer model and the BiGRU neural network model of WOA. The design methods of the Transformer model and the BiGRU neural network model of WOA refer to other design steps of the present invention.

[0053] 4. Design of the Transformer model - the TCN neural network model of WOA

[0054] The Transformer model - the TCN neural network model of WOA is the series connection of the Transformer model and the TCN neural network model of WOA. The output of the Transformer model serves as the input of the TCN neural network model of WOA.

[0055] The TCN neural network model is a convolutional network generated by one - dimensional convolution combined with various convolution methods, which is more suitable for time - series data of water quality parameters. It mainly "temporalizes" one - dimensional convolution and makes improvements on the basis of the residual network, which can effectively avoid problems such as gradient disappearance or gradient explosion of water quality parameters. The TCN neural network model includes a fully convolutional network of one - dimensional water quality parameters, causal convolution, dilated convolution, and residual connection. It has the advantages of parallel computing of water quality parameters, flexible receptive field, stable gradient, and lower memory. The TCN neural network model is mainly stacked by several residual modules of water quality parameters. Each residual module contains two layers of dilated causal convolution to increase the direct mapping of the input information of water quality parameters in the TCN neural network model and speed up the calculation. Causal convolution makes the convolution have a causal relationship, and the current convolution result of water quality parameters only depends on the current and previous water quality input parameters, thus improving the prediction accuracy of water quality parameters in the TCN neural network model; the one - dimensional full convolution of water quality parameters generates an output sequence with the same length as the input water quality parameter sequence of the TCN neural network model to retain the information of the input water quality data; dilated convolution enables the TCN neural network model to obtain a larger receptive field of water quality parameters by increasing the interval number of points in the convolution kernel, so that the feature of the input water quality parameter information can be extracted faster; the residual connection is an effective method for training deep networks, which can enable the network to transmit water quality parameter information across layers. The design methods of the Transformer model and the TCN neural network model of WOA refer to other design steps of the present invention.

[0056] 5. Design of the ESN neural network model of WOA - the BiGRU neural network models 1 - 4 of WOA

[0057] The ESN neural network model of WOA - The structure of the BiGRU neural network models 1 - 4 of WOA is the same as that of the ESN neural network model of WOA - The BiGRU neural network model of WOA. The ESN neural network model of WOA - The BiGRU neural network model of WOA is the series connection of the ESN neural network model of WOA and the BiGRU neural network model of WOA. The design methods of the ESN neural network model of WOA and the BiGRU neural network model of WOA refer to the relevant design steps of this patent.

[0058] 4. Design of the AANN neural network model of interval dynamic fuzzy numbers - The ESN neural network model 1 of WOA

[0059] The AANN neural network model of interval dynamic fuzzy numbers - The ESN neural network model 1 of WOA is the series connection of the AANN neural network model and the ESN neural network model of WOA.

[0060] The ESN neural network model is a new type of dynamic neural network, which has all the advantages of dynamic neural networks. At the same time, due to the introduction of the "reservoir" concept in the echo state network, this method can better adapt to the nonlinear system identification of water quality parameters than general dynamic neural networks. The "reservoir" is to transform the intermediate connection part of the traditional dynamic neural network into a randomly connected "reservoir". The whole learning process is actually the process of learning how to connect the "reservoir". The "reservoir" is actually a large-scale recursive structure of randomly generated water quality parameters, in which the neurons are sparsely connected. Usually, SD represents the percentage of interconnected neurons in the total number of neurons N. The state equation of the ESN neural network model is:

[0061]

[0062] In the formula, W is the water quality parameter state variable of the neural network, and W in is the input water quality parameter variable of the neural network; W back is the connection weight matrix of the output water quality parameter state variable of the neural network; x(n) represents the internal state of the water quality parameter of the neural network; W out is the connection weight matrix between the core reservoir of the ESN neural network model, the input of the neural network, and the output of the neural network; is the deviation of the output water quality parameter of the neural network or can represent the water quality parameter noise; f = f[f 1 , f 2 , …, f n are the n activation functions of the internal neurons of the "reservoir"; f i is the hyperbolic tangent function; f out is the ε output functions of the ESN neural network model.

[0063] The process of using the Whale Optimization Algorithm (WOA) to optimize the Echo State Network (ESN) neural network model for the WOA-ESN neural network model is as follows: (1) Initialize the weights and thresholds of the ESN neural network model, which can be randomly generated; (2) Set the parameters of the WOA algorithm, including the maximum number of iteration steps and the population size of whales, etc., and set the mean square error as the objective function for iterative optimization; (3) When the WOA algorithm reaches the maximum number of evolutionary generations or meets the accuracy requirements of water quality parameters, import the optimized weights and thresholds into the ESN neural network model; (4) Use the ESN neural network model optimized by the WOA to train the prediction target of water quality parameters and conduct test analysis on the model. The five parameters a, b, u, v, and q output by the Adaptive Neuro-Fuzzy Inference System (ANFIS) neural network model - WOA-ESN neural network model 1 form the interval dynamic fuzzy number of the detected parameters as [(a, b); u; (v, q)].

[0064] II. Design of the Intelligent Multi-Point Water Quality Monitoring Subsystem

[0065] The intelligent multi-point water quality monitoring subsystem includes a TDL press delay timer, SOM neural network models 1-4, several water quality information detection modules, Transformer model - WOA-ESN neural network models 1-4, and ANFIS neural network model - WOA-ESN neural network model 2 for interval dynamic fuzzy numbers.

[0066] 1. Design of the SOM Neural Network Model

[0067] The SOM neural network model is called a self-organizing feature mapping network. This SOM neural network model is a teacherless self-organizing and self-learning network composed of a fully connected neuron array. When a SOM neural network model receives water quality parameters, it will be divided into different response regions, and each region has different response characteristics to the input water quality parameters. In the present invention, the SOM neural network model is used to classify the samples of the output values of the input water quality parameter sensors, and the sample parameters of each type are input into the corresponding water quality information detection module. The learning algorithm of the SOM neural network model is as follows:

[0068] (1) Initialization of connection weights. Assign smaller weights to the connection weights from the output values of N water quality parameter sensors as the input neurons to the output neurons of the SOM neural network model, which are the output values of the temperature sensor, the pH value sensor, and the TDS sensor respectively.

[0069] (2) Calculate the Euclidean distance d j , that is, the distance between the output sample X of the input water quality parameter sensor and each output neuron j, and calculate a neuron j with the smallest distance * , that is, determine a certain unit k.

[0070] (3) Modify the output neuron j * and the weights of its "adjacent neurons".

[0071] (4) Calculate the output of the SOM neural network model to classify the types of output values of multiple water quality parameter sensors.

[0072] 2. Design of the AANN neural network model - WOA's ESN neural network model 2 for interval dynamic fuzzy numbers

[0073] The AANN neural network model - WOA's ESN neural network model 2 for interval dynamic fuzzy numbers has the same structure as the AANN neural network model - WOA's ESN neural network model 1, which is the series connection of the AANN neural network model and WOA's ESN neural network model. The five parameters x, y, j, k, m output by the AANN neural network model - WOA's ESN neural network model 2 form an interval dynamic fuzzy number of the detected water quality level as [(x, y); j; (k, m)]. This interval dynamic fuzzy number classifies the detected water quality safety into 5 different levels from Class I to Class V according to water quality standards. The correspondence table between the 5 interval dynamic fuzzy numbers and the 5 safety level grades of water quality safety is as follows in Table 1.

[0074] Table 1. Correspondence table between water quality safety grades and interval dynamic fuzzy numbers

[0075] Serial number Safety level Interval dynamic fuzzy number 1 Three categories [(0.45，0.55)；0.7；(0.8，0.95)] 2 Two categories [(0.55，0.75)；0.9；(0.95，1)] 3 One category [(0.85，0.95)；1；(1，1)] 4 Four categories [(0.25，0.35)；0.5；(0.65，0.75)] 5 Five categories [(0，0.15)；0.3；(0.45，0.55)]

[0076] III. Design of the multi - point water quality intelligent monitoring big data Internet of Things system

[0077] The multi - point water quality intelligent monitoring big data Internet of Things system includes a water quality measurement terminal, a water quality monitoring cloud platform, a water quality remote monitoring terminal, and a water quality monitoring mobile APP terminal. The water quality measurement terminal realizes the real - time detection of the detected water quality parameters. The water quality remote monitoring terminal has an intelligent multi - point water quality monitoring subsystem to realize the intelligent prediction and classification of the water quality of the detected object. Information interaction is realized among the water quality measurement terminal, the water quality remote monitoring terminal, and the water quality monitoring mobile APP terminal through the water quality monitoring cloud platform. The water quality monitoring mobile APP terminal and the water quality remote monitoring terminal access the water quality monitoring cloud platform through the 5G network to realize water quality supervision. The structure of the multi - point water quality intelligent monitoring big data Internet of Things system is shown in Figure 3 as follows.

[0078] 1. Design of the water quality measurement terminal

[0079] A large number of water quality measurement terminals based on NB-IoT modules are used as water quality parameter perception terminals. The water quality measurement terminals realize information interaction with the water quality monitoring cloud platform through NB-IoT modules. The water quality measurement terminals include sensors for collecting temperature, dissolved oxygen, pH value, turbidity, TDS water quality, and conductivity, corresponding signal conditioning circuits, STM32 single-chip microcomputers, cameras, and NB-IoT modules. The software of the water quality measurement terminals mainly realizes NB-IoT module communication and the collection and preprocessing of water quality parameters. The software is designed using the C language program, with high compatibility, greatly improving the work efficiency of software design and development, and enhancing the reliability, readability, and portability of the program code. The structure of the water quality measurement terminal is shown in Figure 4 .

[0080] 4. Software of the water quality remote monitoring terminal

[0081] The water quality remote monitoring terminal is an industrial control computer. The water quality remote monitoring terminal mainly realizes the monitoring and management of water quality parameters and the information interaction with the water quality monitoring cloud platform. The main functions of the water quality remote monitoring terminal are the setting of water quality communication parameters, water quality data analysis and data management, and an intelligent multi-point water quality monitoring subsystem. The software functions of the water quality remote monitoring terminal are shown in Figure 5 .

Claims

1. A multi-point water quality intelligent monitoring method, characterized in that: The steps include: Step 1: Install multiple temperature sensors, multiple TDS sensors, multiple turbidity sensors, and multiple pH sensors at multiple points in the water quality area to be tested, and obtain multiple sensor values ​​respectively; Step 2: Input the detection values ​​of multiple temperature sensors, multiple TDS sensors, multiple turbidity sensors, and multiple pH sensors into SOM neural network models 1 to 4 respectively to obtain corresponding 1-C types of temperature sensor values, 1-C types of TDS sensor values, 1-C types of turbidity sensor values, and 1-C types of pH sensor values; Step 3: The 1-C types of temperature sensor values, 1-C types of TDS sensor values, 1-C types of turbidity sensor values, and 1-C types of pH sensor values ​​obtained in step 2 are respectively calibrated and predicted by the water quality information detection module to obtain the calibrated predicted values ​​of the temperature sensor, TDS sensor, turbidity sensor, and pH sensor; Step 4: multiple temperature sensor correction prediction values, multiple TDS sensor correction prediction values, multiple turbidity sensor correction prediction values, and multiple pH value sensor correction prediction values ​​are respectively input into the ESN neural network model 1-4 of the Transformer model-WOA for data processing; Step 5: The output values ​​of Transformer model-WOA ESN neural network models 1-4 are used as the input of interval dynamic fuzzy number AANN neural network model-WOA ESN neural network model 2, and the output values ​​of Transformer model-WOA ESN neural network models 1-4 are used as the input of other three Transformer model-WOA ESN neural network models in turn for decoupling calibration; Step 6: The output of the AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 2 of WOA is fed back to the Transformer model - ESN neural network model 1-4 of WOA after passing through the TDL beat delay device. The AANN neural network model of interval dynamic fuzzy numbers - ESN neural network model 2 of WOA outputs 5 interval dynamic fuzzy numbers to divide the water quality of the detected area into five levels.

2. A multi-point water quality intelligent monitoring method according to claim 1, characterized in that: There are a total of 4C water quality information detection modules, of which C water quality information detection modules are used to process 1-C types of temperature sensor values ​​output by SOM neural network model 1, C water quality information detection modules are used to process 1-C types of TDS sensor values ​​output by SOM neural network model 2, C water quality information detection modules are used to process 1-C types of turbidity sensor values ​​output by SOM neural network model 3, and C water quality information detection modules are used to process 1-C types of PH value sensor values ​​output by SOM neural network model 4.

3. A multi-point water quality intelligent monitoring method according to claim 2, characterized in that: The water quality information detection module includes AANN neural network model-WOA BiGRU neural network model, Transformer model-WOA TCN neural network model, Transformer model-WOA BiGRU neural network model, Transformer model-ARIMA model, WOA ESN neural network model-WOA BiGRU neural network model 1-4 and interval dynamic fuzzy number AANN neural network model-WOA ESN neural network model 1; The parameter sensor outputs are respectively used as the inputs of the AANN neural network model-WOA's BiGRU neural network model, the Transformer model-WOA's TCN neural network model, the Transformer model-WOA's BiGRU neural network model and the Transformer model-ARIMA model; the differences between the outputs of the Transformer model-WOA's TCN neural network model and the outputs of the ESN neural network model-WOA's BiGRU neural network model 4 are respectively used as the inputs of the ESN neural network model-WOA's BiGRU neural network model 1 and the AANN neural network model-WOA's ESN neural network model 1 of interval dynamic fuzzy numbers; the sums of the outputs of the ESN neural network model-WOA's BiGRU neural network model 1 and the ESN neural network model-WOA's BiGRU neural network model 4 of WOA are respectively used as the inputs of the Transformer model-WOA's TCN neural network model and the AANN neural network model-WOA's ESN neural network model 1 of interval dynamic fuzzy numbers; The difference between the output of the BiGRU neural network model of Transformer model-WOA and the output of the BiGRU neural network model of ESN neural network model of WOA-WOA 4 is used as the input of the BiGRU neural network model of ESN neural network model of WOA-WOA 2 and the AANN neural network model of interval dynamic fuzzy number-WOA ESN neural network model 1, and the sum of the output of the BiGRU neural network model of ESN neural network model of WOA-WOA 2 and the output of the BiGRU neural network model of ESN neural network model of WOA-WOA 4 is used as the input of the BiGRU neural network model of Transformer model-WOA and the AANN neural network model of interval dynamic fuzzy number-WOA ESN neural network model 1; The difference between the output of Transformer model-ARIMA model and the output of ESN neural network model of WOA-BiGRU neural network model 4 of WOA is used as the input of ESN neural network model of WOA-BiGRU neural network model 3 of WOA and AANN neural network model of interval dynamic fuzzy number-ESN neural network model 1 of WOA, and the sum of the output of ESN neural network model of WOA-BiGRU neural network model 3 and the output of ESN neural network model of WOA-BiGRU neural network model 4 of WOA is used as the input of Transformer model-ARIMA model and AANN neural network model of interval dynamic fuzzy number-ESN neural network model 1 of WOA; The output of AANN neural network model of interval dynamic fuzzy number - ESN neural network model 1 of WOA and the output of AANN neural network model - BiGRU neural network model of WOA are used as the input of ESN neural network model of WOA - BiGRU neural network model 4 of WOA. The five parameters a, b, u, v, q output by AANN neural network model of interval dynamic fuzzy number - ESN neural network model 1 of WOA constitute the interval dynamic fuzzy number of the detected parameter [(a, b); u; (v, q)], where a, b, u, v, q respectively represent the minimum value, minimum value, mean value, maximum value and maximum value of the predicted value of the parameter sensor output.

4. A multi-point water quality intelligent monitoring method according to claim 1, characterized in that: In step 5, the connection relationship between the Transformer model-WOA ESN neural network model 1-4 and the interval dynamic fuzzy number AANN neural network model-WOA ESN neural network model 2 is: The output of Transformer model-WOA ESN neural network model 1 is used as the input of AANN neural network model-WOA ESN neural network model 2, Transformer model-WOA ESN neural network model 2, Transformer model-WOA ESN neural network model 3 and Transformer model-WOA ESN neural network model 4 of interval dynamic fuzzy number; The output of Transformer model-WOA ESN neural network model 2 is used as the input of interval dynamic fuzzy number AANN neural network model-WOA ESN neural network model 2, Transformer model-WOA ESN neural network model 1, Transformer model-WOA ESN neural network model 3 and Transformer model-WOA ESN neural network model 4; The output of Transformer model-WOA ESN neural network model 3 is used as the input of AANN neural network model-WOA ESN neural network model 2, Transformer model-WOA ESN neural network model 1, Transformer model-WOA ESN neural network model 2 and Transformer model-WOA ESN neural network model 4 of interval dynamic fuzzy number; The output of Transformer model-WOA's ESN neural network model 4 is used as the input of AANN neural network model-WOA's ESN neural network model 2, Transformer model-WOA's ESN neural network model 1, Transformer model-WOA's ESN neural network model 2 and Transformer model-WOA's ESN neural network model 3 of interval dynamic fuzzy numbers.

5. A multi-point water quality intelligent monitoring method according to claim 1, characterized in that: In the step 6, the output of the AANN neural network model of interval dynamic fuzzy numbers-the ESN neural network model 2 of WOA is used as the input of the TDL beat delay device, and the output of the TDL beat delay device is used as the corresponding input of the Transformer model-the ESN neural network model 1 of WOA, the Transformer model-the ESN neural network model 2 of WOA, the Transformer model-the ESN neural network model 3 of WOA, and the Transformer model-the ESN neural network model 4 of WOA. The AANN neural network model of interval dynamic fuzzy numbers-the ESN neural network model 2 of WOA outputs 5 interval dynamic fuzzy numbers to divide the water quality of the detected area into five levels.

6. A multi-point water quality intelligent monitoring method according to claim 5, characterized in that: The relationship between the interval dynamic fuzzy number and the safety level in the water quality monitoring results output by the AANN neural network model of the interval dynamic fuzzy number-WOA ESN neural network model 2 is: 。 7. A multi-point water quality intelligent monitoring method according to claim 3, characterized in that: The AANN neural network model of interval dynamic fuzzy numbers-ESN neural network model 1 of WOA and the AANN neural network model of interval dynamic fuzzy numbers-ESN neural network model 2 of WOA have the same structure, both of which are the AANN neural network model and the ESN neural network model of WOA connected in series; the AANN neural network model-WOA BiGRU neural network model is the AANN neural network model and the BiGRU neural network model of WOA connected in series; the Transformer model-ARIMA model is the Transformer model and the ARIMA model connected in series; the Transformer model-WOA BiGRU neural network model is the Transformer model and the BiGRU neural network model of WOA connected in series; the Transformer model-WOA TCN neural network model is the Transformer model and the TCN neural network model of WOA connected in series; the ESN neural network model of WOA-WOA BiGRU neural network models 1-4 have the same structure, both of which are the ESN neural network model of WOA and the BiGRU neural network model of WOA connected in series.

8. A multi-point water quality intelligent monitoring method according to claim 1, characterized in that: The Transformer model-WOA ESN neural network models 1-4 have the same structure, which are all a series connection of the Transformer model and the WOA ESN neural network model.

9. A multi-point water quality intelligent monitoring big data Internet of Things system, characterized in that: The method comprises a water quality measuring terminal, a water quality monitoring cloud platform, a water quality remote monitoring terminal and a water quality monitoring mobile phone APP terminal. The water quality measuring terminal realizes real-time detection of the detected water quality parameters. The water quality remote monitoring terminal is provided with an intelligent multi-point water quality monitoring subsystem. The intelligent multi-point water quality monitoring subsystem realizes the multi-point water quality intelligent monitoring method described in any one of claims 1 to 8, realizes intelligent prediction of the water quality of the detected object, realizes information interaction among the water quality measuring terminal, the water quality remote monitoring terminal and the water quality monitoring mobile phone APP terminal through the water quality monitoring cloud platform, and the water quality monitoring mobile phone APP terminal and the water quality remote monitoring terminal access the water quality monitoring cloud platform through the 5G network to realize water quality supervision.

Images