Water quality monitoring method and device based on Internet of Things

Abstract

The invention discloses a water quality monitoring method and device based on Internet of Things, and relates to the technical field of data analysis. Acquiring monitoring data of the target water area in a preset time period through a plurality of monitoring nodes arranged in the target water area; extracting key feature data from the monitoring data according to the key variable set corresponding to the node; the key variable set is determined by historical monitoring data; preprocessing the key feature data to obtain a standard time sequence matrix; the standard time sequence matrix comprises a plurality of parameter time sequence vectors; and taking the standard time sequence matrix as the input of the deep learning model corresponding to the target node to obtain a dissolved oxygen prediction sequence. Under different water environments, contributions of influence factors of dissolved oxygen changes are different, and characteristic parameters closest to dissolved oxygen can be found through historical monitoring data of nodes, so that redundant data are eliminated, interference of noise data is reduced, and accuracy of machine learning model prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a water quality monitoring method and device based on the Internet of Things. Background Art

[0002] Water quality monitoring is a key link in environmental protection and water resource management. Remote real-time monitoring based on the Internet of Things can timely understand the health of water bodies. Among the many water quality parameters, dissolved oxygen (DO) is one of the most important indicators. Dissolved oxygen refers to the amount of oxygen dissolved in water, which is a necessary condition for maintaining the normal life activities of aquatic organisms. Too low dissolved oxygen levels can lead to the death of aquatic organisms, the decline of the self-purification ability of water bodies, and the deterioration of water quality; while too high dissolved oxygen levels may cause bubble disease, destroy the ecological balance and promote the massive reproduction of algae. Therefore, in water quality management, maintaining an appropriate dissolved oxygen level is crucial to the health of water bodies and the survival of aquatic organisms.

[0003] In order to carry out timely and effective water quality management, it is necessary to predict the future trend of dissolved oxygen so that preventive measures can be taken in advance. Patent CN115166176A discloses a pond dissolved oxygen concentration prediction method based on the TCN model. By decomposing the original sequence, it captures the original characteristics and internal correlation characteristics of the dissolved oxygen time series at a deeper level, thereby improving the prediction accuracy of dissolved oxygen concentration.

[0004] However, the change of dissolved oxygen is affected by many factors, including the combined effects of temperature, pH value, total nitrogen, total phosphorus, turbidity and other parameters, and there are complex nonlinear relationships between these parameters. The prediction method that only relies on the time series data of dissolved oxygen itself has poor adaptability and low accuracy when the influencing factors change. Summary of the invention

[0005] The purpose of the present invention is to solve the problem of low accuracy mentioned in the above background technology, and to propose a water quality monitoring method and device based on the Internet of Things.

[0006] The first aspect of the present invention provides a water quality monitoring method based on the Internet of Things, the method comprising:

[0007] By deploying multiple monitoring nodes in the target waters, the monitoring data of the target waters within a preset time period are obtained; the monitoring data include dissolved oxygen, water temperature, pH, conductivity, turbidity, total phosphorus, total nitrogen, permanganate, ammonia nitrogen concentration, chlorophyll a, water level and flow;

[0008] Extracting key feature data from the monitoring data according to a key variable set corresponding to a target node; the target node is any one of a plurality of monitoring nodes; the key variable set is determined by historical monitoring data;

[0009] Preprocessing the key feature data to obtain a standard timing matrix; the standard timing matrix includes a plurality of parameter timing vectors;

[0010] The standard time series matrix is ​​used as the input of the deep learning model corresponding to the target node to obtain a dissolved oxygen prediction sequence.

[0011] Optionally, the specific process of determining the key variable set from historical monitoring data includes:

[0012] Acquire historical monitoring data of the target node, and preprocess the historical monitoring data to obtain a standard data set;

[0013] Calculating the maximum information coefficient of the target variable and the characteristic variable to obtain a correlation index; the target variable is dissolved oxygen, and the characteristic variable is any one of the other monitoring parameters except dissolved oxygen;

[0014] The characteristic variables whose correlation index is greater than the correlation threshold are taken as key characteristic parameters and added into the key variable set.

[0015] Optionally, the preprocessing of the historical monitoring data to obtain a standard data set includes:

[0016] Performing time alignment on the historical monitoring data to obtain a plurality of first time series data;

[0017] Perform outlier detection on each first time series data, and delete the outlier to obtain multiple second time series data;

[0018] Performing interpolation processing on each second time series data to obtain a plurality of third time series data;

[0019] Using wavelet transform, denoising is performed on each third time series data to obtain a plurality of fourth time series data;

[0020] Normalizing each fourth time series data to obtain a plurality of fifth time series data;

[0021] Combine multiple fifth time series data to obtain a standard data set.

[0022] Optionally, the maximum information coefficient is calculated for the target variable and the characteristic variable to obtain a correlation index including:

[0023] Step 1, initialize the acquisition step size and information coefficient set;

[0024] Step 2: Update the acquisition step length;

[0025] Step 3: extracting a target variable subsequence and a feature variable subsequence from the standard data set according to the acquisition step size;

[0026] Step 4, calculating the maximum information coefficient of the target variable subsequence and the feature variable subsequence, and adding the maximum information coefficient to the information coefficient set;

[0027] Step 5: If the acquisition step reaches the preset maximum limit, proceed to step 6; otherwise, return to step 1;

[0028] Step six, calculating the average value of the elements in the information coefficient set as the correlation index between the target variable and the feature variable.

[0029] Optionally, the deep learning model includes a feature extraction network and a prediction network; the feature extraction network includes a plurality of gated recurrent units and a fully connected layer; the prediction network is an N-Beats network; wherein:

[0030] Multiple gated recurrent units are used to process different parameter time series vectors respectively to obtain multiple feature vectors;

[0031] The fully connected layer is used to fuse multiple feature vectors to obtain a comprehensive feature representation;

[0032] The prediction network is used to receive the comprehensive feature representation as input and output a prediction sequence of dissolved oxygen.

[0033] A second aspect of the present invention provides a water quality monitoring device based on the Internet of Things, the device comprising:

[0034] A data acquisition module is used to acquire monitoring data of the target water area within a preset time period through multiple monitoring nodes arranged in the target water area; the monitoring data includes dissolved oxygen, water temperature, pH, conductivity, turbidity, total phosphorus, total nitrogen, permanganate, ammonia nitrogen concentration, chlorophyll a, water level and flow;

[0035] A key parameter extraction module, used to extract key feature data from the monitoring data according to a key variable set corresponding to a target node; the target node is any one of a plurality of monitoring nodes; the key variable set is determined by historical monitoring data;

[0036] A preprocessing module, used for preprocessing the key feature data to obtain a standard timing matrix; the standard timing matrix includes a plurality of parameter timing vectors;

[0037] The prediction module is used to use the standard time series matrix as the input of the deep learning model corresponding to the target node to obtain a dissolved oxygen prediction sequence.

[0038] Optionally, the device further includes a key variable set determination module; the key variable set determination module includes:

[0039] A historical data integration module is used to obtain historical monitoring data of the target node and pre-process the historical monitoring data to obtain a standard data set;

[0040] A correlation analysis module, used to calculate the maximum information coefficient of the target variable and the characteristic variable to obtain a correlation index; the target variable is dissolved oxygen, and the characteristic variable is any other monitoring parameter except dissolved oxygen;

[0041] The variable set determination module is used to take the characteristic variables whose correlation index is greater than the correlation threshold as key characteristic parameters and add them into the key variable set.

[0042] Optionally, the historical data integration module includes:

[0043] A data alignment module, used for performing time alignment on the historical monitoring data to obtain a plurality of first time series data;

[0044] An outlier processing module, used to perform outlier detection on each first time series data and delete the outlier to obtain a plurality of second time series data;

[0045] A missing value processing module, used for performing interpolation processing on each second time series data to obtain a plurality of third time series data;

[0046] A denoising module, used for performing denoising processing on each third time series data by using wavelet transform to obtain a plurality of fourth time series data;

[0047] A normalization module, used for normalizing each fourth time series data to obtain a plurality of fifth time series data;

[0048] The data set determination module is used to combine multiple fifth time series data to obtain a standard data set.

[0049] Optionally, the correlation analysis module includes:

[0050] Initialization module, used to initialize the acquisition step size and information coefficient set;

[0051] Step size update module, used to update the acquisition step size;

[0052] A subsequence extraction module, used to extract a target variable subsequence and a feature variable subsequence from the standard data set according to the acquisition step size;

[0053] A mutual information calculation module, used to calculate the maximum information coefficient of the target variable subsequence and the feature variable subsequence, and add the maximum information coefficient to the information coefficient set;

[0054] A loop judgment module, used for entering the index determination module if the acquisition step reaches a preset maximum limit; otherwise, returning to the step update module;

[0055] The index determination module is used to calculate the average value of the elements in the information coefficient set as the correlation index between the target variable and the feature variable.

[0056] Optionally, the deep learning model includes a feature extraction network and a prediction network; the feature extraction network includes a plurality of gated recurrent units and a fully connected layer; the prediction network is an N-Beats network; wherein:

[0057] Multiple gated recurrent units are used to process different parameter time series vectors respectively to obtain multiple feature vectors;

[0058] The fully connected layer is used to fuse multiple feature vectors to obtain a comprehensive feature representation;

[0059] The prediction network is used to receive the comprehensive feature representation as input and output a prediction sequence of dissolved oxygen.

[0060] Beneficial effects of the present invention:

[0061] The present invention proposes a water quality monitoring method based on the Internet of Things, which comprises: obtaining monitoring data of a target water area within a preset time period by deploying a plurality of monitoring nodes in a target water area; the monitoring data comprises dissolved oxygen, water temperature, pH, conductivity, turbidity, total phosphorus, total nitrogen, permanganate, ammonia nitrogen concentration, chlorophyll a, water level and flow; extracting key feature data from the monitoring data according to a key variable set corresponding to the target node; the target node is any one of the plurality of monitoring nodes; the key variable set is determined by historical monitoring data; preprocessing the key feature data to obtain a standard time series matrix; the standard time series matrix comprises a plurality of parameter time series vectors; and using the standard time series matrix as an input of a deep learning model corresponding to the target node to obtain a dissolved oxygen prediction sequence.

[0062] Under different water environments, the contribution of factors affecting dissolved oxygen changes is not the same. Through the historical monitoring data of the nodes, we can find the characteristic parameters most closely related to dissolved oxygen, thereby eliminating redundant data, reducing the interference of noise data, and improving the accuracy of machine learning model predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The present invention will be further described below in conjunction with the accompanying drawings.

[0064] Figure 1 A flow chart of a water quality monitoring method based on the Internet of Things is provided for an embodiment of the present invention;

[0065] Figure 2A structural diagram of a water quality monitoring device based on the Internet of Things is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] The embodiment of the present invention provides a water quality monitoring method based on the Internet of Things. Figure 1 , Figure 1 A flow chart of a water quality monitoring method based on the Internet of Things provided in an embodiment of the present invention. The method comprises the following steps:

[0068] S101, acquiring monitoring data of a target water area within a preset time period by deploying a plurality of monitoring nodes in the target water area.

[0069] S102, extracting key feature data from the monitoring data according to the key variable set corresponding to the target node.

[0070] S103, preprocessing the key feature data to obtain a standard time series matrix.

[0071] S104, using the standard time series matrix as the input of the deep learning model corresponding to the target node to obtain a dissolved oxygen prediction sequence.

[0072] Among them, the monitoring data include dissolved oxygen, water temperature, pH, conductivity, turbidity, total phosphorus, total nitrogen, permanganate, ammonia nitrogen concentration, chlorophyll a, water level and flow; the target node is any one of multiple monitoring nodes; the key variable set is determined by historical monitoring data; the standard time series matrix includes multiple parameter time series vectors.

[0073] According to a water quality monitoring method based on the Internet of Things provided by an embodiment of the present invention, the characteristic parameters most closely related to dissolved oxygen can be found through the historical monitoring data of the nodes, thereby eliminating redundant data, reducing the interference of noise data, and improving the accuracy of machine learning model prediction.

[0074] In one implementation, the elements in the key variable set are those that are highly correlated with changes in dissolved oxygen. At different monitoring nodes, due to differences in meteorology and water environment, the contribution of various factors is different, resulting in different sets of key variables at different monitoring nodes. For example, the key variable set of a monitoring node A is {dissolved oxygen, water temperature, pH, conductivity, ammonia nitrogen concentration, total nitrogen, water depth, flow}, while the key variable set of monitoring node B is {dissolved oxygen, water temperature, pH, conductivity, turbidity, total nitrogen, water depth, flow}. By evaluating the key variables for each monitoring node, more accurate model matching and training can be achieved.

[0075] In one embodiment, the specific process of determining the key variable set from historical monitoring data includes:

[0076] Step 1: Obtain historical monitoring data of the target node and preprocess the historical monitoring data to obtain a standard data set.

[0077] Step 2: Calculate the maximum information coefficient of the target variable and the characteristic variable to obtain the correlation index.

[0078] Step three: The feature variables whose correlation index is greater than the correlation threshold are taken as key feature parameters and added into the key variable set.

[0079] Among them, the target variable is dissolved oxygen, and the characteristic variable is any one of the other monitoring parameters except dissolved oxygen.

[0080] In one implementation, the correlation threshold may be an average value of all correlation indices.

[0081] In one implementation, the maximum information coefficient calculation can effectively capture the nonlinear relationship between variables, making the selection of key features more accurate, thereby avoiding the interference of redundant data.

[0082] In one embodiment, the historical monitoring data is preprocessed to obtain a standard data set including:

[0083] Step 1: Time-align the historical monitoring data to obtain multiple first time series data.

[0084] Step 2: Perform outlier detection on each first time series data and delete the outliers to obtain multiple second time series data.

[0085] Step three, interpolate each second time series data to obtain multiple third time series data.

[0086] Step 4: Use wavelet transform to perform denoising on each third time series data to obtain multiple fourth time series data.

[0087] Step five, normalize each fourth time series data to obtain multiple fifth time series data.

[0088] Step six, combining multiple fifth time series data to obtain a standard data set.

[0089] In one implementation, missing values ​​can be supplemented through interpolation methods such as polynomial interpolation or mean interpolation, which improves the continuity of time series data and ensures data integrity during model training.

[0090] In one implementation, the denoising process of the wavelet transform effectively reduces the background noise, making the key features more prominent and enhancing the quality of the data.

[0091] In one embodiment, the maximum information coefficient is calculated for the target variable and the feature variable to obtain the correlation index including:

[0092] Step 1: Initialize the acquisition step size and information coefficient set.

[0093] Step 2: Update the acquisition step size.

[0094] Step three: Extract the target variable subsequence and feature variable subsequence from the standard data set according to the acquisition step size.

[0095] Step 4: Calculate the maximum information coefficient of the target variable subsequence and the feature variable subsequence, and add the maximum information coefficient to the information coefficient set.

[0096] Step 5: If the acquisition step reaches the preset maximum limit, go to step 6; otherwise, return to step 1.

[0097] Step six, calculate the average value of the elements in the information coefficient set as the correlation index between the target variable and the feature variable.

[0098] In one implementation, the acquisition step size may be updated successively in the order of 0.5 hours, 1 hour, 2 hours, and 3 hours, so as to measure the correlation between the target variable and the characteristic variable at different time scales, and ensure that the extracted key characteristic parameters are highly correlated with dissolved oxygen.

[0099] In one implementation, the process of calculating the maximum information coefficient of the target variable subsequence and the feature variable subsequence is:

[0100] Through different division methods, the target variable subsequence and the feature variable subsequence are divided into several intervals respectively;

[0101] Under different partitioning methods, the mutual information values ​​of the target variable subsequence and the feature variable subsequence are calculated;

[0102] The largest mutual information value is selected as the maximum information coefficient.

[0103] Among them, under one division method (dividing the target variable subsequence into m intervals and the feature variable subsequence into n intervals), the specific process of calculating the mutual information value of the target variable subsequence and the feature variable subsequence is:

[0104] Step 1: According to the division method, the target variable subsequence and the feature variable subsequence are divided into several intervals to obtain discrete interval sequences X and Y; where X = {x1, x2, ..., x m}; Y = {y1, y2, ..., y n}.

[0105] Step 2: Create a two-dimensional frequency table based on the target variable subsequence and the feature variable subsequence, as shown in Table 1.

[0106] Table 1

[0107] <![CDATA[x1]]> <![CDATA[x2]]> ... <![CDATA[y1]]> <![CDATA[C 11 ]]> <![CDATA[C 12 ]]> ... <![CDATA[y2]]> <![CDATA[C 21 ]]> <![CDATA[C 21 ]]> ... ... ... ... ...

[0108] Among them, any x i and j A grid area is formed in the two-dimensional coordinate system, and the (feature variable value F, target variable value D) data pair at each acquisition moment is used as the coordinate value to match the grid area to obtain the number C of data pairs in the grid area. ij .

[0109] Step 3: According to the two-dimensional frequency table, the joint probability distribution is obtained:

[0110]

[0111] Step 4: Calculate the mutual information value of the target variable subsequence and the feature variable subsequence according to the joint probability distribution:

[0112]

[0113] Where I is the mutual information value required;

[0114] The maximum information coefficient method can effectively capture the nonlinear relationship between variables and enhance the scientific nature of feature selection.

[0115] In one embodiment, the deep learning model includes a feature extraction network and a prediction network; the feature extraction network includes a plurality of gated recurrent units and a fully connected layer; the prediction network is an N-Beats network; wherein:

[0116] Multiple gated recurrent units are used to process different parameter time series vectors respectively to obtain multiple feature vectors.

[0117] The fully connected layer is used to fuse multiple feature vectors to obtain a comprehensive feature representation.

[0118] The prediction network is used to receive the comprehensive feature representation as input and output a prediction sequence of dissolved oxygen.

[0119] In one implementation, the gated recurrent unit (GRU) is a recursive neural network that can effectively process different parameter time series vectors and fully capture the time series characteristics and dynamic changes in the time series. N-Beats is a feedforward neural network with a relatively simple architecture, fast training speed, and is not prone to gradient vanishing or exploding problems. The N-Beats network itself can directly perform time series prediction. When combined with GRU, the high-level features extracted from the GRU can be used to further predict and enhance the accuracy and stability of the prediction.

[0120] The embodiment of the present invention provides a water quality monitoring device based on the Internet of Things. Figure 2 , Figure 2 This is a structural diagram of a water quality monitoring device based on the Internet of Things provided by an embodiment of the present invention. The device includes:

[0121] The data acquisition module is used to acquire the monitoring data of the target water area within a preset time period through multiple monitoring nodes deployed in the target water area.

[0122] The key parameter extraction module is used to extract key feature data from the monitoring data according to the key variable set corresponding to the target node.

[0123] The preprocessing module is used to preprocess the key feature data to obtain a standard time series matrix.

[0124] The prediction module is used to use the standard time series matrix as the input of the deep learning model corresponding to the target node to obtain the dissolved oxygen prediction sequence.

[0125] Among them, the monitoring data include dissolved oxygen, water temperature, pH, conductivity, turbidity, total phosphorus, total nitrogen, permanganate, ammonia nitrogen concentration, chlorophyll a, water level and flow; the target node is any one of multiple monitoring nodes; the key variable set is determined by historical monitoring data; the standard time series matrix includes multiple parameter time series vectors.

[0126] A water quality monitoring device based on the Internet of Things provided by an embodiment of the present invention can find the characteristic parameters most closely related to dissolved oxygen through the historical monitoring data of the nodes, thereby eliminating redundant data, reducing the interference of noise data, and improving the accuracy of machine learning model prediction.

[0127] In one embodiment, the device further includes a key variable set determination module; the key variable set determination module includes:

[0128] The historical data integration module is used to obtain the historical monitoring data of the target node and pre-process the historical monitoring data to obtain a standard data set.

[0129] The correlation analysis module is used to calculate the maximum information coefficient of the target variable and the characteristic variable to obtain the correlation index.

[0130] The variable set determination module is used to take the characteristic variables whose correlation index is greater than the correlation threshold as key characteristic parameters and add them into the key variable set.

[0131] Among them, the target variable is dissolved oxygen, and the characteristic variable is any one of the other monitoring parameters except dissolved oxygen.

[0132] In one embodiment, the historical data integration module includes:

[0133] The data alignment module is used to perform time alignment on the historical monitoring data to obtain a plurality of first time series data.

[0134] The outlier processing module is used to perform outlier detection on each first time series data and delete the outlier to obtain multiple second time series data.

[0135] The missing value processing module is used to perform interpolation processing on each second time series data to obtain multiple third time series data.

[0136] The denoising module is used to perform denoising processing on each third time series data by using wavelet transform to obtain multiple fourth time series data.

[0137] The normalization module is used to normalize each fourth time series data to obtain multiple fifth time series data.

[0138] The data set determination module is used to combine multiple fifth time series data to obtain a standard data set.

[0139] In one embodiment, the correlation analysis module includes:

[0140] The initialization module is used to initialize the acquisition step size and information coefficient set.

[0141] The step size update module is used to update the acquisition step size.

[0142] The subsequence extraction module is used to extract the target variable subsequence and feature variable subsequence from the standard data set according to the acquisition step size.

[0143] The mutual information calculation module is used to calculate the maximum information coefficient of the target variable subsequence and the feature variable subsequence, and add the maximum information coefficient to the information coefficient set.

[0144] The loop judgment module is used to enter the index determination module if the acquisition step reaches the preset maximum limit; otherwise, it returns to the step update module.

[0145] The index determination module is used to calculate the average value of the elements in the information coefficient set as the correlation index between the target variable and the feature variable.

[0146] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A water quality monitoring method based on the Internet of Things, characterized in that: The method comprises: By deploying multiple monitoring nodes in the target waters, the monitoring data of the target waters within a preset time period are obtained; the monitoring data include dissolved oxygen, water temperature, pH, conductivity, turbidity, total phosphorus, total nitrogen, permanganate, ammonia nitrogen concentration, chlorophyll a, water level and flow; Extracting key feature data from the monitoring data according to a key variable set corresponding to a target node; the target node is any one of a plurality of monitoring nodes; the key variable set is determined by historical monitoring data; Preprocessing the key feature data to obtain a standard timing matrix; the standard timing matrix includes a plurality of parameter timing vectors; The standard time series matrix is ​​used as the input of the deep learning model corresponding to the target node to obtain a dissolved oxygen prediction sequence.

2. A water quality monitoring method based on the Internet of Things according to claim 1, characterized in that: The specific process of determining the key variable set from historical monitoring data includes: Acquire historical monitoring data of the target node, and preprocess the historical monitoring data to obtain a standard data set; Calculating the maximum information coefficient of the target variable and the characteristic variable to obtain a correlation index; the target variable is dissolved oxygen, and the characteristic variable is any one of the other monitoring parameters except dissolved oxygen; The characteristic variables whose correlation index is greater than the correlation threshold are taken as key characteristic parameters and added into the key variable set.

3. A water quality monitoring method based on the Internet of Things according to claim 2, characterized in that: The preprocessing of the historical monitoring data to obtain a standard data set includes: Performing time alignment on the historical monitoring data to obtain a plurality of first time series data; Perform outlier detection on each first time series data, and delete the outlier to obtain multiple second time series data; Performing interpolation processing on each second time series data to obtain a plurality of third time series data; Using wavelet transform, denoising is performed on each third time series data to obtain a plurality of fourth time series data; Normalizing each fourth time series data to obtain a plurality of fifth time series data; Combine multiple fifth time series data to obtain a standard data set.

4. A water quality monitoring method based on the Internet of Things according to claim 2, characterized in that: The maximum information coefficient is calculated for the target variable and the characteristic variable to obtain the correlation index, which includes: Step 1, initialize the acquisition step size and information coefficient set; Step 2: Update the acquisition step length; Step 3: extracting a target variable subsequence and a feature variable subsequence from the standard data set according to the acquisition step size; Step 4, calculating the maximum information coefficient of the target variable subsequence and the feature variable subsequence, and adding the maximum information coefficient to the information coefficient set; Step 5: If the acquisition step reaches the preset maximum limit, proceed to step 6; otherwise, return to step 1; Step six, calculating the average value of the elements in the information coefficient set as the correlation index between the target variable and the feature variable.

5. The water quality monitoring method based on the Internet of Things according to claim 1 is characterized in that: The deep learning model includes a feature extraction network and a prediction network; the feature extraction network includes a plurality of gated recurrent units and a fully connected layer; the prediction network is an N-Beats network; in: Multiple gated recurrent units are used to process different parameter time series vectors respectively to obtain multiple feature vectors; The fully connected layer is used to fuse multiple feature vectors to obtain a comprehensive feature representation; The prediction network is used to receive the comprehensive feature representation as input and output a prediction sequence of dissolved oxygen.

6. A water quality monitoring device based on the Internet of Things, characterized in that: The device comprises: A data acquisition module is used to acquire monitoring data of the target water area within a preset time period through multiple monitoring nodes arranged in the target water area; the monitoring data includes dissolved oxygen, water temperature, pH, conductivity, turbidity, total phosphorus, total nitrogen, permanganate, ammonia nitrogen concentration, chlorophyll a, water level and flow; A key parameter extraction module, used to extract key feature data from the monitoring data according to a key variable set corresponding to a target node; the target node is any one of a plurality of monitoring nodes; the key variable set is determined by historical monitoring data; A preprocessing module, used for preprocessing the key feature data to obtain a standard timing matrix; the standard timing matrix includes a plurality of parameter timing vectors; The prediction module is used to use the standard time series matrix as the input of the deep learning model corresponding to the target node to obtain a dissolved oxygen prediction sequence.

7. The water quality monitoring device based on the Internet of Things according to claim 6, characterized in that: The device also includes a key variable set determination module; The key variable set determination module comprises: A historical data integration module is used to obtain historical monitoring data of the target node and pre-process the historical monitoring data to obtain a standard data set; A correlation analysis module, used to calculate the maximum information coefficient of the target variable and the characteristic variable to obtain a correlation index; the target variable is dissolved oxygen, and the characteristic variable is any other monitoring parameter except dissolved oxygen; The variable set determination module is used to take the characteristic variables whose correlation index is greater than the correlation threshold as key characteristic parameters and add them into the key variable set.

8. The water quality monitoring device based on the Internet of Things according to claim 7, characterized in that: The historical data integration module includes: A data alignment module, used for performing time alignment on the historical monitoring data to obtain a plurality of first time series data; An outlier processing module, used to perform outlier detection on each first time series data and delete the outlier to obtain a plurality of second time series data; A missing value processing module, used for performing interpolation processing on each second time series data to obtain a plurality of third time series data; A denoising module, used for performing denoising processing on each third time series data by using wavelet transform to obtain a plurality of fourth time series data; A normalization module, used for normalizing each fourth time series data to obtain a plurality of fifth time series data; The data set determination module is used to combine multiple fifth time series data to obtain a standard data set.

9. The water quality monitoring device based on the Internet of Things according to claim 7, characterized in that: The correlation analysis module comprises: Initialization module, used to initialize the acquisition step size and information coefficient set; Step size update module, used to update the acquisition step size; A subsequence extraction module, used to extract a target variable subsequence and a feature variable subsequence from the standard data set according to the acquisition step size; A mutual information calculation module, used to calculate the maximum information coefficient of the target variable subsequence and the feature variable subsequence, and add the maximum information coefficient to the information coefficient set; A loop judgment module, used for entering the index determination module if the acquisition step reaches a preset maximum limit; otherwise, returning to the step update module; The index determination module is used to calculate the average value of the elements in the information coefficient set as the correlation index between the target variable and the feature variable.

10. The water quality monitoring device based on the Internet of Things according to claim 6, characterized in that: The deep learning model includes a feature extraction network and a prediction network; the feature extraction network includes a plurality of gated recurrent units and a fully connected layer; the prediction network is an N-Beats network; in: Multiple gated recurrent units are used to process different parameter time series vectors respectively to obtain multiple feature vectors; The fully connected layer is used to fuse multiple feature vectors to obtain a comprehensive feature representation; The prediction network is used to receive the comprehensive feature representation as input and output a prediction sequence of dissolved oxygen.

Images