Water quality parameter monitoring and predicting method based on signal decomposition and Informer network

Abstract

The invention belongs to the field of water quality monitoring, and particularly relates to a water quality parameter monitoring and predicting method based on signal decomposition and an Informer network. The method comprises the following steps: acquiring regional water quality parameter data; decomposing the water quality parameter data by using an empirical orthogonal function to obtain a space function and a time function; decomposing the decomposed time function by using an ensemble empirical mode; the decomposed high-frequency components are predicted by adopting an Informer model; integrating prediction results into a new time function; integrating the new time function and the space function to obtain a water quality parameter predicted value; and judging whether a water quality abnormal event occurs or not according to the threshold. The monitoring and predicting method provided by the invention combines the empirical orthogonal function, integrates the advantages of the empirical mode decomposition and the Informer model, can better process the nonlinear and unsteady state characteristics of the water quality parameter change, and improves the prediction accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of water quality monitoring, and in particular relates to a water quality parameter monitoring and prediction method based on signal decomposition and Informer network. Background Art

[0002] With the intensification of human activities and the impact of global climate change, changes in water quality have had an increasing impact on ecosystems and human life. For complex deep-sea waters, water quality is a key factor affecting fish growth, directly affecting the production, reproduction and growth of marine aquatic products in the aquaculture area, and is the most important monitoring project in deep-sea fishery aquaculture. Therefore, solving the problem of real-time monitoring and prediction of changes in water quality parameters (such as seawater temperature, dissolved oxygen content, pH) in the field of water quality monitoring is of great significance to ensure healthy aquaculture.

[0003] Existing water quality monitoring technologies have obvious deficiencies in data analysis capabilities and prediction accuracy. Many methods are limited by data processing capabilities when quantitatively analyzing the response of water quality parameters to environmental changes, making it difficult to accurately reflect the dynamic changes in water quality. In addition, existing prediction models are often too complex, such as random forests, long short-term memory networks (LSTMs), and convolutional neural networks (CNNs) in machine learning. Although they can handle nonlinear relationships, they still face great accuracy challenges when predicting the duration and intensity of water quality parameters.

[0004] Existing methods often face data quality issues during model training and testing. The acquired data may be noisy, missing, or incomplete, resulting in poor model training results, thus affecting the reliability of the prediction. Summary of the invention

[0005] In response to the above technical problems, the present invention provides a water quality parameter monitoring and prediction method based on signal decomposition and Informer network. The method provided by the present invention uses signal decomposition technology to effectively extract and reduce noise of complex water quality monitoring data, making the data clearer and more reliable; combined with the powerful sequence modeling capability of the Informer network, it can capture the long-term dependencies in the time series when processing multi-dimensional water quality parameters, thereby significantly improving the prediction accuracy.

[0006] The present invention is achieved through the following technical solutions:

[0007] A water quality parameter monitoring and prediction method based on signal decomposition and informer network, the method comprising:

[0008] (1) Obtain regional water quality parameter data;

[0009] (2) decomposing the water quality parameter data using an empirical orthogonal function to obtain a spatial function and a temporal function;

[0010] (3) Decomposing the time function obtained by the decomposition in step (2) using a set empirical mode;

[0011] (4) Using the Informer model to predict the high-frequency components decomposed in step (3); integrating the prediction results into a new time function;

[0012] (5) integrating the new time function with the space function in step (2) to obtain a predicted value of a water quality parameter;

[0013] (6) Determine whether an abnormal water quality event has occurred based on the threshold.

[0014] Furthermore, in step (1), the water quality parameter data obtained includes any one of seawater temperature, dissolved oxygen content, and pH.

[0015] Furthermore, in step (2), the water quality parameter data is preprocessed by dimensionality reduction using an empirical orthogonal function, and the spatiotemporal three-dimensional data set is split into mutually unrelated spatial functions and time functions.

[0016] Furthermore, step (2) specifically includes:

[0017] (2.1) Data preprocessing: performing denoising, filling missing values, and normalization preprocessing on the original water quality parameter data collected in step (1);

[0018] (2.2) Constructing the covariance matrix: Perform covariance calculation on the preprocessed data. For an N×M data matrix X, where N is the number of time steps and M is the number of spatial points, calculate the covariance matrix C of the data matrix X. The formula is as follows:

[0019]

[0020] Among them, X T is the transpose of X;

[0021] (2.3) Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and corresponding eigenvectors; the eigenvalue decomposition formula is:

[0022] C·V=V·Λ

[0023] Wherein, V contains eigenvectors; Λ is a diagonal matrix, Λ contains eigenvalues; the eigenvectors constitute the spatial function EOFs; the eigenvalues ​​represent the importance of the corresponding eigenvectors;

[0024] (2.4) Select the spatial function: sort the eigenvalues ​​in Λ according to their numerical values, select the spatial function EOFs consisting of the eigenvector corresponding to the largest eigenvalue and proceed to step (2.5);

[0025] (2.5) Calculating the time coefficient: Project the original water quality parameter data collected in step (1) onto the spatial function EOFs selected in step (2.4); for each time step, obtain the time coefficient PCs by performing a dot multiplication between the original water quality parameter data and the selected spatial function EOFs, and the formula is as follows:

[0026] PCs = X·V;

[0027] The spatial functions EOFs represent the spatial functions in the water quality parameter data, and the temporal coefficients PCs represent the temporal function of each spatial function.

[0028] Furthermore, in step (3), the time function obtained by the empirical orthogonal decomposition in step (2) is decomposed using the set empirical mode, including the following steps:

[0029] (3.1) Finding extreme points: Taking the time coefficient PCs as the original signal i(t), find all the maximum and minimum points of the original signal i(t);

[0030] (3.2) Fitting envelope: Connect all maximum points into the upper envelope and all minimum points into the lower envelope by using cubic spline interpolation method;

[0031] (3.3) Calculating the mean envelope: Calculating the average value of the upper envelope and the lower envelope to obtain the mean envelope h(t);

[0032] (3.4) generating an intermediate signal: subtracting the mean envelope h(t) from the original signal i(t) to obtain an intermediate signal m(t);

[0033] (3.5) Determine whether the intermediate signal m(t) satisfies the two constraints of IMF: If so, the intermediate signal m(t) is the first high-frequency component decomposed and is recorded as IMF1, and proceed to step (3.6); if not, return to step (3.1) and iterate repeatedly;

[0034] Among them, the two constraints are: Constraint 1: In the entire data segment, the number of extreme value points and the number of zero-crossing points must be equal or the difference cannot exceed one at most; Constraint 2: At any time, the average value of the upper envelope formed by the local maximum value points and the lower envelope formed by the local minimum value points is zero, that is, the upper envelope and the lower envelope are locally symmetrical with respect to the time axis;

[0035] (3.6) Subtract the high-frequency component IMF1 in step (3.5) from the original signal i(t) to obtain a new signal without the high-frequency component; repeat steps (3.1)-(3.5) to obtain the next high-frequency component IMF2;

[0036] (3.7) Repeat steps (3.6)-(3.6) until the residual component res is a monotonic function or a constant, and the decomposition process stops;

[0037] The original signal is decomposed and expressed as:

[0038] i(t)=∑m i (t)+res;

[0039] mi(t) is the frequency component decomposed each time, and i is the number of all decomposed modes.

[0040] Furthermore, step (4) is specifically as follows: using the Informer model, taking the high-frequency component decomposed in (3) as input, outputting the prediction result m'(t) of the high-frequency component at the future time point

[0041] The prediction results of the high-frequency components at different time points in the future are interpolated and smoothed to obtain a continuous time series, that is, the new signal i'(t) is obtained by integration:

[0042] i'(t) = ∑m'(t) + res;

[0043] The signal i'(t) is the new time function obtained.

[0044] Furthermore, step (5) is specifically as follows: superimposing the new time function and the space function to obtain predicted values ​​of water quality parameters at different time points at different spatial points.

[0045] The present invention proposes a water quality parameter monitoring and prediction method based on signal decomposition and informer network, which includes the following key points:

[0046] The present invention adopts the EOF-EEMD-Informer model: water quality parameters (seawater temperature, dissolved oxygen content, pH) present complex multidimensional distribution in time and space, and directly processing these variables will result in too many input variables; the empirical orthogonal function (EOF) analysis method is used to pre-process the water quality parameter data by dimensionality reduction, establish the spatial relationship between discrete grid points, extract the dominant spatial relationship, and split the complex three-dimensional data set into a series of unrelated spatial functions (EOFs) and time functions (PCs). EEMD is used to decompose the non-stationary time function into several intrinsic mode functions again, and finally the Informer deep network is used for prediction.

[0047] Processing nonlinearity and nonstationarity of water quality parameters: Water quality parameters (such as temperature, salinity, dissolved oxygen) have nonlinear and nonstationary characteristics. The present invention uses empirical orthogonal decomposition to decompose water quality data into time function and space function, and then uses ensemble empirical mode decomposition to decompose the time function. This not only simplifies the amount of predicted data, but also eliminates the nonlinear and nonstationary characteristics in the data that affect the prediction, thereby improving the accuracy and timeliness of water quality parameter monitoring and prediction.

[0048] Beneficial technical effects of the present invention:

[0049] The present invention provides a water quality parameter monitoring and prediction method that combines signal decomposition with an informer network, aiming to improve the real-time monitoring and prediction accuracy of water quality parameters, overcome the limitations of traditional methods in data analysis, model complexity, prediction factor selection, and data quality, and overcome the main technical difficulties in monitoring and predicting water quality parameters.

[0050] The water quality parameter monitoring and prediction method provided by the present invention combines the advantages of empirical orthogonal function, ensemble empirical mode decomposition and Informer model, can better handle the nonlinear and non-steady-state characteristics of water quality parameter changes, and improve the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of a method for monitoring and predicting water quality parameters based on signal decomposition and Informer network in an embodiment of the present invention;

[0052] Figure 2 is the RMSE index for seawater temperature prediction in the embodiment of the present invention;

[0053] Figure 3 It is the MAE index of seawater temperature prediction in the embodiment of the present invention;

[0054] Figure 4 It is the MAPE index for seawater temperature prediction in the embodiment of the present invention;

[0055] Figure 5 is a TNR result index for predicting seawater temperature changes in an embodiment of the present invention;

[0056] Figure 6 is a TPR result indicator for predicting seawater temperature changes in an embodiment of the present invention;

[0057] Figure 7 It is the F1score result indicator for predicting seawater temperature changes in the embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the detailed description of the present invention below. Those skilled in the art can fully understand the present invention without the description of these details.

[0060] Embodiment 1: A method for monitoring and predicting water quality parameters based on signal decomposition and informer network, the method comprising:

[0061] (1) Obtain regional water quality parameter data;

[0062] (2) decomposing the water quality parameter data using an empirical orthogonal function to obtain a spatial function and a temporal function;

[0063] (3) Decomposing the time function obtained by the decomposition in step (2) using a set empirical mode;

[0064] (4) Using the Informer model to predict the high-frequency components decomposed in step (3); integrating the prediction results into a new time function;

[0065] (5) integrating the new time function with the space function in step (2) to obtain a predicted value of a water quality parameter;

[0066] (6) Determining whether an abnormal water quality event has occurred based on a threshold value. Specifically, the threshold value is determined based on historical data or based on environmental standards; based on environmental standards specifically means setting the upper and lower limits based on the baseline value of the water quality parameter and the environmental standard, and then determining the normal range of the water quality parameter; based on historical data analysis specifically means analyzing past data to determine the typical change pattern and threshold value of abnormal events.

[0067] In step (1) of this embodiment, the water quality parameter data obtained includes any one of seawater temperature, dissolved oxygen content, and pH. And the water quality parameter data obtained is a spatiotemporal three-dimensional data set. Specifically, the present invention collects water quality parameter data of the area to be predicted through a multi-sensor network;

[0068] In step (2) of this embodiment, the water quality parameter data is preprocessed by dimensionality reduction using an empirical orthogonal function, and the spatiotemporal three-dimensional data set is split into mutually unrelated spatial functions and time functions.

[0069] In this embodiment, step (2) specifically includes:

[0070] (2.1) Data preprocessing: performing denoising, filling missing values, and normalization preprocessing on the original water quality parameter data collected in step (1);

[0071] (2.2) Constructing the covariance matrix: Perform covariance calculation on the preprocessed data. For an N×M data matrix X, where N is the number of time steps and M is the number of spatial points, calculate the covariance matrix C of the data matrix X. The formula is as follows:

[0072]

[0073] Among them, X T is the transpose of X;

[0074] (2.3) Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and corresponding eigenvectors; the eigenvalue decomposition formula is:

[0075] C·V=V·Λ

[0076] Wherein, V contains eigenvectors; Λ is a diagonal matrix, Λ contains eigenvalues; the eigenvectors constitute the spatial function EOFs; the eigenvalues ​​represent the importance of the corresponding eigenvectors;

[0077] (2.4) Select the spatial function: sort the eigenvalues ​​in Λ according to their numerical values, select the spatial function EOFs consisting of the eigenvector corresponding to the largest eigenvalue and proceed to step (2.5);

[0078] (2.5) Calculating the time coefficient: Project the original water quality parameter data collected in step (1) onto the spatial function EOFs selected in step (2.4); for each time step, obtain the time coefficient PCs by performing a dot multiplication between the original water quality parameter data and the selected spatial function EOFs, and the formula is as follows:

[0079] PCs = X·V;

[0080] The spatial function EOFs represents the spatial function in the water quality parameter data, and the temporal coefficient PCs represents the temporal function of each spatial function. Step (2) can reduce the dimension of the data, thereby better describing the characteristics and changing trends of the data.

[0081] In step (3) of this embodiment, the time function obtained by the empirical orthogonal decomposition in step (2) is decomposed using the set empirical mode, including the following steps:

[0082] (3.1) Finding extreme points: Taking the time coefficient PCs as the original signal i(t), find all the maximum and minimum points of the original signal i(t);

[0083] (3.2) Fitting envelope: Connect all the maximum points into the upper envelope and all the minimum points into the lower envelope by cubic spline interpolation; all the extreme points must be included in the upper and lower envelopes;

[0084] (3.3) Calculating the mean envelope: Calculating the average value of the upper envelope and the lower envelope to obtain the mean envelope h(t);

[0085] (3.4) generating an intermediate signal: subtracting the mean envelope h(t) from the original signal i(t) to obtain an intermediate signal m(t);

[0086] (3.5) Determine whether the intermediate signal m(t) satisfies the two constraints of IMF: If so, the intermediate signal m(t) is the first high-frequency component decomposed and is recorded as IMF1, and proceed to step (3.6); if not, return to step (3.1) and iterate repeatedly;

[0087] Among them, the two constraints are: Constraint 1: In the entire data segment, the number of extreme points and the number of zero-crossing points must be equal or the difference cannot exceed one; this means that the sample signal needs to repeatedly cross the time axis, that is, there must be at least one crossing between two adjacent extreme points. Constraint 2: At any time, the average value of the upper envelope formed by the local maximum points and the lower envelope formed by the local minimum points is zero, that is, the upper envelope and the lower envelope are locally symmetrical with respect to the time axis; this means that the upper envelope and the lower envelope of the sample signal need to remain symmetrical on the time axis.

[0088] (3.6) Subtract the high-frequency component IMF1 in step (3.5) from the original signal i(t) to obtain a new signal without the high-frequency component; repeat steps (3.1)-(3.5) to obtain the next high-frequency component IMF2;

[0089] (3.7) Repeat steps (3.6)-(3.6) until the residual component res is a monotonic function or a constant, and the decomposition process stops;

[0090] The original signal is decomposed and expressed as:

[0091] i(t)=∑m i (t)+res.

[0092] mi(t) is the frequency component decomposed each time, and i is the number of all decomposed modes.

[0093] In this embodiment, step (4) is specifically as follows: using the Informer model, taking the high-frequency component decomposed in (3) as input, and outputting the prediction result m′(t) of the high-frequency component at the future time point

[0094] The prediction results of the high-frequency components at different time points in the future are interpolated and smoothed to obtain a continuous time series, that is, the new signal i'(t) is obtained by integration:

[0095] i′(t)=∑m′(t)+res;

[0096] The signal i'(t) is the new time function obtained.

[0097] In this embodiment, step (5) specifically includes: superimposing the new time function and the space function to obtain predicted values ​​of water quality parameters at different time points at different spatial points.

[0098] In the present invention, Informer is used to predict high-frequency components, which has the following technical effects: Informer adopts a sparse self-attention mechanism, which has higher importance and plays an important role in calculating attention distribution and focusing on key information, avoiding serious information loss.

[0099] In the present invention, Informer adopts distilled self-attention to reduce the overall memory usage.

[0100] In the present invention, the Informer adopts a generative decoder to prevent data leakage.

[0101] The seawater temperature is predicted using the method provided in Example 1; Figure 2-4 It can be seen that the overall index is small, and the prediction effect is good. The reason why MAPE is high in some areas is that the true value of SST is close to or equal to 0. The calculation process of MAPE needs to be divided by the true value, and the program will return a larger number. The average value of RMSE is 0.5418, the average value of MAE is 0.3391, and the average value of MAPE is 0.1856. Among them, RMSE represents the average value of the square difference between the predicted value and the true value, MAE represents the average value of the absolute difference between the predicted value and the true value, MAPE represents the average value of the absolute percentage difference between the predicted value and the true value, and RMSE is the square root of MSE, which is used to measure the standard deviation between the predicted value and the true value. The smaller the value of the above indicators, the better the prediction ability of the model.

[0102] Figure 5-7These are the three indicators after the seawater temperature is predicted using the EOF-EEMD-Informer model. The seawater temperature in the Bohai Sea area is also predicted with a step length of 30 days, and the SST for 365 days in 2022 is predicted. Then the number of days with abnormal seawater temperature is counted, and compared with the actual situation, TP, FP, TN, and FN are obtained, and then the three indicators of TNR, TPR, and F1Score are calculated. The average value of F1Score is 0.7324, the average value of TPR is 0.6944, and the average value of TNR is 0.9339.

[0103] Among them, F1Score is a deep learning evaluation indicator that combines precision and recall scores; its calculation formula is as follows:

[0104]

[0105] The F1Score ranges from 0 to 1, where the maximum value (reached when FN=FP=0) indicates perfect prediction performance, and the minimum value (reached when TP=0) indicates that the prediction performance of the proposed prediction model is very poor. The average F1Score corresponding to the prediction using the method of the present invention is 0.7324, which shows that the prediction performance provided by the present invention is very superior.

[0106] In the present invention, TP is a true positive value, indicating that MHW events are identified in both the observed and predicted data sets. FN is a false negative value, indicating that MHW events are detected in the observed data set but not in the predicted data set. FP is a false positive value, indicating that there are no MHW events in the observed data set but MHW events are found in the predicted data set. TN is a true negative value, indicating that there are no MHW events in both the observed data set and the predicted data set. The details are shown in Table 1:

[0107] Table 1 Confusion matrix of MHW events

[0108]

[0109] In ROC space, the true positive rate (TPR) is defined as the correct detection of MHW occurrence, that is, the observed MHW can be detected in the prediction model. Similarly, the false positive rate (FPR) is defined as the incorrect detection of MHW occurrence, that is, the predicted MHW does not exist in the real observed data. It is worth noting that TPR indicates the confidence of the predicted MHW occurrence. They are calculated as:

[0110]

[0111] In addition, the ratios of false negative (FNR) and true negative (TNR) are calculated as FNR=1-TPR and TNR=1-FPR, respectively. The average TPR corresponding to the prediction using the method of the present invention is 0.6944, and the average TNR is 0.9339. It can be seen that the prediction performance using the method provided by the present invention is very superior.

[0112] The present invention provides a water quality parameter monitoring and prediction method using an EOF-EEMD-Informer model. Since water quality parameters present a complex multidimensional distribution in time and space, directly processing these variables will result in too many input variables. The present invention uses an empirical orthogonal function (EOF) analysis method to perform dimensionality reduction preprocessing on water quality parameter data, establish a spatial relationship between discrete grid points, extract the dominant spatial relationship, and split the complex three-dimensional data set into a series of unrelated spatial functions (EOFs) and time functions (PCs); use EEMD to decompose the non-stationary time function into several intrinsic mode functions again, and finally use the Informer deep network for prediction; it has the following advantages:

[0113] (1) Processing the nonlinear and non-stationary characteristics of water quality parameters: Water quality parameters have nonlinear and non-stationary characteristics. The method provided by the present invention uses empirical orthogonal decomposition to decompose water quality data into time function and space function, and then uses ensemble empirical mode decomposition to decompose the time function, which not only simplifies the amount of predicted data, but also eliminates the nonlinear and non-stationary characteristics in the data that affect the prediction, making the prediction easier, thereby improving the accuracy and timeliness of water quality parameter monitoring and prediction;

[0114] (2) Efficient processing of large-scale water quality data: Given that water quality monitoring requires processing a large amount of data, the present invention uses dimensionality reduction technology to effectively reduce the burden of model processing. Through signal decomposition and data dimensionality reduction, the model is more efficient in processing large-scale water quality data, can quickly respond to changes, and provide monitoring results and prediction information in a timely manner.

[0115] (3) Improving prediction accuracy: The present invention combines the advantages of signal decomposition and informer network to more comprehensively capture the complex patterns and interrelationships of water quality parameters, thereby improving the ability to identify water quality changes and enhancing prediction accuracy.

[0116] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A water quality parameter monitoring and prediction method based on signal decomposition and informer network, characterized in that: The method comprises: (1) Obtain regional water quality parameter data; (2) decomposing the water quality parameter data using an empirical orthogonal function to obtain a spatial function and a temporal function; (3) Decomposing the time function obtained by the decomposition in step (2) using a set empirical mode; (4) Using the Informer model to predict the high-frequency components decomposed in step (3); integrating the prediction results into a new time function; (5) integrating the new time function with the space function in step (2) to obtain a predicted value of a water quality parameter; (6) Determine whether an abnormal water quality event has occurred based on the threshold.

2. According to claim 1, a water quality parameter monitoring and prediction method based on signal decomposition and informer network is characterized in that: In step (1), the water quality parameter data obtained includes any one of seawater temperature, dissolved oxygen content, and pH.

3. According to claim 2, a water quality parameter monitoring and prediction method based on signal decomposition and informer network is characterized in that: In step (2), the water quality parameter data is preprocessed by dimensionality reduction using an empirical orthogonal function, and the spatiotemporal three-dimensional data set is split into mutually unrelated spatial functions and time functions.

4. According to claim 3, a water quality parameter monitoring and prediction method based on signal decomposition and informer network is characterized in that: Step (2) specifically includes: (2.1) Data preprocessing: performing denoising, filling missing values, and normalization preprocessing on the original water quality parameter data collected in step (1); (2.2) Constructing the covariance matrix: Perform covariance calculation on the preprocessed data. For an N×M data matrix X, where N is the number of time steps and M is the number of spatial points, calculate the covariance matrix C of the data matrix X. The formula is as follows: Among them, X T is the transpose of X; (2.3) Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and corresponding eigenvectors; the eigenvalue decomposition formula is: C·V=V·Λ Among them, V contains eigenvectors; Λ is a diagonal matrix, and Λ contains eigenvalues; (2.4) Select the spatial function: sort the eigenvalues ​​in Λ according to their numerical values, select the spatial function EOFs consisting of the eigenvector corresponding to the largest eigenvalue and proceed to step (2.5); (2.5) Calculating the time coefficient: Project the original water quality parameter data collected in step (1) onto the spatial function EOFs selected in step (2.4); for each time step, obtain the time coefficient PCs by performing a dot multiplication between the original water quality parameter data and the selected spatial function EOFs, and the formula is as follows: PCs = X·V; The spatial functions EOFs represent the spatial functions in the water quality parameter data, and the temporal coefficients PCs represent the temporal function of each spatial function.

5. According to claim 4, a water quality parameter monitoring and prediction method based on signal decomposition and informer network is characterized in that: In step (3), the time function obtained by the empirical orthogonal decomposition in step (2) is decomposed using the set empirical mode, including the following steps: (3.1) Finding extreme points: Taking the time coefficient PCs as the original signal i(t), find all the maximum and minimum points of the original signal i(t); (3.2) Fitting envelope: Connect all maximum points into the upper envelope and all minimum points into the lower envelope by using cubic spline interpolation method; (3.3) Calculating the mean envelope: Calculating the average value of the upper envelope and the lower envelope to obtain the mean envelope h(t); (3.4) generating an intermediate signal: subtracting the mean envelope h(t) from the original signal i(t) to obtain an intermediate signal m(t); (3.5) Determine whether the intermediate signal m(t) satisfies the two constraints of IMF: If so, the intermediate signal m(t) is the first high-frequency component decomposed and is recorded as IMF1, and proceed to step (3.6); if not, return to step (3.1) and iterate repeatedly; Among them, the two constraints are: Constraint 1: In the entire data segment, the number of extreme value points and the number of zero-crossing points must be equal or the difference cannot exceed one at most; Constraint 2: At any time, the average value of the upper envelope formed by the local maximum value points and the lower envelope formed by the local minimum value points is zero, that is, the upper envelope and the lower envelope are locally symmetrical with respect to the time axis; (3.6) Subtract the high-frequency component IMF1 in step (3.5) from the original signal i(t) to obtain a new signal without the high-frequency component; repeat steps (3.1)-(3.5) to obtain the next high-frequency component IMF2; (3.7) Repeat steps (3.6)-(3.6) until the residual component res is a monotonic function or a constant, and the decomposition process stops; The original signal is decomposed and expressed as: i(t)=Σm i (t)+res; mi(t) is the frequency component decomposed each time, and i is the number of all decomposed modes.

6. The method for monitoring and predicting water quality parameters based on signal decomposition and informer network according to claim 1, characterized in that: Step (4) is as follows: using the Informer model, taking the high-frequency components decomposed in (3) as input, and outputting the prediction result m'(t) of the high-frequency components at the future time point The prediction results of the high-frequency components at different time points in the future are interpolated and smoothed to obtain a continuous time series, that is, the new signal i'(t) is obtained by integration: i'(t) = ∑m'(t) + res; The signal i'(t) is the new time function obtained.

7. The method for monitoring and predicting water quality parameters based on signal decomposition and informer network according to claim 1, characterized in that: Step (5) is specifically as follows: superimposing the new time function and the space function to obtain the predicted values ​​of water quality parameters at different time points at different spatial points.

Images