Water quality prediction method fusing spatial-temporal characteristics
By constructing a deep prediction model that integrates convolutional neural network, bidirectional gating recurrent units and multi-head attention mechanism, combining spatial weight matrix and Osprey optimization algorithm, the problem of insufficient fusion of spatiotemporal features in water quality prediction is solved, and the prediction accuracy and model adaptability are improved.
Patent Information
- Application Number
- CN202510639097.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-19
- Publication Date
- 2025-08-01
AI Technical Summary
The existing water quality prediction methods fail to effectively integrate spatial and temporal characteristics, ignore the influence of spatial location, and lack prediction accuracy and generalization capabilities.
A deep prediction model combining a convolutional neural network, a bidirectional gating cyclic unit and a multi-head attention mechanism is adopted, combining a spatial weight matrix and an Osprey optimization algorithm to optimize hyperparameters and extract the spatiotemporal characteristics of water quality data.
It significantly improves the accuracy of water quality prediction and the generalization ability of the model. It is suitable for water quality prediction of water quality in water stations and multi-indexes, and improves the stability and adaptability of the model.
Smart Images
Figure CN120408528A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of water environment monitoring and water quality prediction, and in particular to a water quality prediction method integrating temporal and spatial characteristics. Background Art
[0002] With the acceleration of industrialization and urbanization, surface water bodies face increasingly severe pollution risks. Water quality prediction, as a key technical approach to scientific water environment management and risk early warning, has garnered widespread attention in recent years. Existing water quality prediction methods are mostly based on time series, utilizing statistical models, machine learning, or deep learning techniques to predict future trends in water quality indicators at a single monitoring station. However, water quality changes are driven not only by temporal factors but also by multiple spatial factors, including spatial location, hydrogeological characteristics, and upstream discharge behavior.
[0003] In existing technologies, research on spatiotemporal feature fusion is still in its early stages and generally suffers from the following deficiencies:
[0004] First, traditional models generally ignore the potential impact of spatial location on water quality evolution and lack modeling of the spatial dependency between different monitoring sections; second, although some methods introduce spatial features, the processing methods are rough and fail to effectively quantify the spatial correlation between sites; finally, in terms of model architecture design, existing methods mostly use a single neural network structure, which makes it difficult to simultaneously take into account the accuracy of time series feature extraction and the adequacy of spatial feature fusion, resulting in limited prediction accuracy and generalization ability.
[0005] Therefore, there is an urgent need for a new method that can systematically integrate the temporal and spatial characteristics of water quality data, explore the potential correlation patterns between sections, and improve the model's response ability and prediction accuracy to water quality changes.
[0006] To this end, a water quality prediction method integrating temporal and spatial characteristics is designed to provide a technical solution to the above technical problems. Summary of the Invention
[0007] Based on this, it is necessary to provide a water quality prediction method that integrates spatiotemporal characteristics to address the above technical problems, which can be used to solve the technical problems of insufficient spatiotemporal information fusion, inaccurate spatial dependency characterization, and weak generalization ability of prediction models in existing water quality prediction technologies.
[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0009] A water quality prediction method integrating spatiotemporal features, 1. A water quality prediction method integrating spatiotemporal features, characterized by the following steps:
[0010] S1: Preprocess water quality data to generate a time series dataset that can be used for model training;
[0011] S2: Construct a spatial weight matrix. By setting the spatial adjacency relationship and weight coefficients, generate the time-series input data containing spatial information in a weighted fusion manner;
[0012] S3: Construct a deep prediction model integrating a convolutional neural network, a bidirectional gated recurrent unit, and a multi-head attention mechanism;
[0013] S4: Optimize the hyperparameters of the deep prediction model using the osprey optimization algorithm to obtain the optimal parameter combination for final training;
[0014] S5: Input the historical water quality monitoring data and output the predicted values of the key indicators at the target section in the future time period.
[0015] As a preferred implementation manner of the water quality prediction method integrating spatio-temporal features provided by the present invention, in step S1, the steps are as follows:
[0016] Preprocess the original water quality time-series data collected from multiple monitoring sections through missing value filling, outlier replacement, wavelet denoising, sliding window partitioning, normalization processing, and test set partitioning.
[0017] As a preferred implementation manner of the water quality prediction method integrating spatio-temporal features provided by the present invention, for the missing value filling, the linear interpolation method is used to complete the missing values, and the expression is as follows:
[0018]
[0019] In the formula: X k is the missing value; X m is the previous known value of X k ; X n is the next known value of X k .
[0020] The normalization operation adopts the maximum-minimum normalization method, and the expression is as follows:
[0021]
[0022] In the formula: y is the normalized data, x is the data to be normalized, max(x) is the maximum value of the data, and min(x) is the minimum value of the data.
[0023] As a preferred implementation manner of the water quality prediction method integrating spatio-temporal features provided by the present invention, for the outlier replacement, the outliers are judged based on 3σ and replaced with adjacent values;
[0024] The wavelet denoising uses the wavelet transform method, selects coif10 as the wavelet basis function, and sets the soft threshold to remove high-frequency noise;
[0025] The sliding window partitioning method is used to construct time series training samples, and the window length is set according to the periodic characteristics of water quality data;
[0026] The data partitioning divides the processed sample set into a training set and a test set.
[0027] As a preferred implementation of the water quality prediction method that fuses spatio-temporal features provided by the present invention, in step S2, the steps are as follows:
[0028] Based on the correlation factors of geographical distance, watershed water system structure and water quality indicators, construct the spatial adjacency relationship between multiple monitoring sections;
[0029] Adopt the weighted method based on the Pearson correlation coefficient to construct the spatial weight matrix;
[0030] Perform weighted averaging on the historical data of the target monitoring section and the data of its spatially adjacent sections to generate an input sequence with spatial features.
[0031] As a preferred implementation of the water quality prediction method that fuses spatio-temporal features provided by the present invention, in step S3, the steps are as follows:
[0032] S31: Extract spatial features through a convolutional neural network;
[0033] S32: Extract the bidirectional dependence features in the water quality time series through a bidirectional gated recurrent unit;
[0034] S33: Allocate attention weights between different time steps through a multi-head attention mechanism
[0035] S34: Construct a deep learning model and use the training set for training, and the validation set is used for parameter tuning.
[0036] As a preferred implementation of the water quality prediction method that fuses spatio-temporal features provided by the present invention, in step S4, the osprey optimization algorithm is used to optimize the hyperparameters of the deep prediction model, and the steps are as follows:
[0037] S41: Initialize the osprey population;
[0038] S42: Construct the hyperparameter search space through parameter setting;
[0039] S43: Set the fitness function as the prediction error index;
[0040] S44: Execute the iterative optimization process;
[0041] S45: Output the optimal hyperparameter combination.
[0042] As a preferred embodiment of the water quality prediction method integrating spatio-temporal features provided by the present invention, in step S44, the iterative optimization process simulates the foraging behavior of ospreys. Through global search and local adjustment, the individual positions are continuously updated to improve the overall fitness value.
[0043] In step S45, the optimal hyperparameter combination is output after meeting the convergence condition or reaching the maximum number of iterations, and is used to construct the final water quality prediction model.
[0044] As a preferred embodiment of the water quality prediction method integrating spatio-temporal features provided by the present invention, in step S5, the steps are as follows:
[0045] Apply the trained and hyperparameter-optimized hybrid water quality prediction model to the water quality data sequence to be predicted, and generate the predicted values of water quality indicators at future moments for each monitoring section.
[0046] As a preferred embodiment of the water quality prediction method integrating spatio-temporal features provided by the present invention, in step S5, the obtained prediction results include one or more indicators among COD, NH3-N, TP, and DO.
[0047] It can be undoubtedly seen that through the above technical solution of the present application, the technical problems to be solved by the present application can surely be solved.
[0048] Meanwhile, through the above technical solutions, the present invention has at least the following beneficial effects:
[0049] 1. A water quality prediction method integrating spatio-temporal features provided by the present invention effectively extracts the spatio-temporal features of water quality data by introducing a spatial weight mechanism to integrate the spatial dependence relationship between multiple monitoring sections and combining with the CNN-BiGRU-MHA deep learning model, significantly improving the prediction accuracy and the generalization ability of the model. At the same time, the osprey optimization algorithm is used to realize the automatic adjustment of hyperparameters, avoiding manual intervention, improving the stability and adaptability of the model, and being applicable to the water quality prediction tasks of multi-site and multi-index river basins, having good engineering application prospects and promotion value.
[0050] 2. The present invention constructs a spatial weight matrix to mine the spatial dependence relationship between monitoring sections, and combines with the CNN-BiGRU-MHA deep learning model to fully extract the spatial and temporal features in water quality data, realizing the accurate prediction of the water quality conditions of multiple sites.
[0051] 3. The present invention is applicable to the analysis of river basin water quality monitoring data and the prediction of water quality trends, especially applicable to surface water environments with multiple monitoring sections and significant spatial dependence relationships. Brief Description of the Drawings
[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0053] Figure 1 It is the overall schematic diagram of the present invention;
[0054] Figure 2 It is the overall flowchart of the present invention;
[0055] Figure 3 It is the schematic flowchart of the data preprocessing of the present invention;
[0056] Figure 4 It is the schematic structural diagram of the hybrid water quality prediction model of the present invention;
[0057] Figure 5 It is the actual prediction effect diagram of the model of the present invention at a certain monitoring section, showing the comparison between the true value and the predicted value. Detailed Embodiment
[0058] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the 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] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings.
[0060] It should be noted that, without conflict, the embodiments in the present invention and the features and technical solutions in the embodiments can be combined with each other.
[0061] It should be noted that similar reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.
[0062] Referring to Figures 1 - 5 , a water quality prediction method integrating spatio-temporal features, applicable to the water quality prediction task of a basin with multiple monitoring sections, the steps are as follows:
[0063] Step 1: Water quality data preprocessing.
[0064] Specifically, historical water quality time series data of multiple monitoring sections in the target watershed are collected. The data includes common indicators such as chemical oxygen demand (COD), ammonia nitrogen (NH3-N), total phosphorus (TP), dissolved oxygen (DO), etc., and the time granularity is at the daily or monthly scale.
[0065] Specifically, preprocessing operations are performed on the original data, including missing value filling, outlier replacement, wavelet denoising, sliding window partitioning, normalization operation, and test set partitioning.
[0066] For missing value filling, the linear interpolation method is used to complete the missing values, and the formula used is:
[0067]
[0068] In the formula: X k is the missing value; X m is the previous known value of X k ; X n is the next known value of X k .
[0069] For outlier replacement, outliers are judged based on 3σ and replaced with neighboring values;
[0070] For wavelet denoising, the wavelet transform (WT) method is adopted, coif10 is selected as the wavelet basis function, and high-frequency noise is removed through soft threshold filtering to retain the main trend information;
[0071] The sliding window partitioning method is used to construct time series training samples, and the window length is set according to the periodic characteristics of the water quality data, with a length of 50 time steps;
[0072] For the normalization operation, the maximum-minimum normalization method is adopted to unify the data scale and improve the model training efficiency. The formula used is:
[0073]
[0074] In the formula: y is the normalized data, x is the data to be normalized, max(x) is the maximum value of the data, and min(x) is the minimum value of the data;
[0075] The data partitioning divides the processed sample set into a training set and a test set, and the common ratio is 7:3.
[0076] Step 2: Spatial weight construction.
[0077] Specifically, to introduce spatial correlation information, a spatial weight matrix between monitoring sections is constructed to perform spatial enhancement processing on the water quality data.
[0078] Specifically, the spatial enhancement processing steps for water quality data include constructing a spatial weight matrix, setting weight factors based on water quality correlation, and establishing a weighted fusion mechanism.
[0079] The establishment of spatial adjacency relationships is based on factors such as geographical distance, the structure of the river basin water system, and the correlation of water quality indicators to determine the spatial association strength between each monitoring section.
[0080] The construction method of the spatial weight matrix includes using a weighted method based on the Pearson correlation coefficient to quantify the spatial influence degree between monitoring sections.
[0081] Among them, the formula for the Pearson correlation coefficient is as follows:
[0082]
[0083] In the formula: n is the number of data points; x i and y i are the values of the two variables of the i-th data point respectively; and are the means of the two variables respectively; r is the Pearson correlation coefficient, and its value ranges from [-1, 1]. The correlation between data is confirmed according to the absolute value of the Pearson correlation coefficient value. The closer the r value is to 1, the higher the correlation.
[0084] The spatial weighted fusion method includes weighted averaging of the historical data of the target monitoring section and the data of its spatially adjacent sections to generate an enhanced input sequence containing spatial dependence information for subsequent model training.
[0085] Among them, the weight assignment of the weighted method based on the Pearson correlation coefficient is based on the following formula:
[0086]
[0087] In the formula: ω ij represents the spatial weight of site i to site j; ρ ij is the influence factor, and ∑ k∈N ρ ik is the sum of the Pearson correlation coefficients of site i to all neighboring sites.
[0088] Step 3: Model construction and training.
[0089] Specifically, construct a deep learning prediction model that fuses spatio-temporal features, and use the training set for training. The validation set is used for parameter tuning to obtain a water quality prediction model with high-precision prediction ability.
[0090] The convolutional neural network (CNN) is used to extract spatial features, and the bidirectional gated recurrent unit (BiGRU) is used to extract temporal features. Combining with the multi-head attention mechanism (MHA) can enhance the model's ability to identify key temporal information;
[0091] The convolutional neural network (CNN) is used to extract local spatial features from the input data, and a two-dimensional convolution method is adopted to identify the spatial patterns of water quality data between different monitoring sections;
[0092] The max-pooling layer is used to perform dimensionality reduction on the convolutional feature map, reducing the computational complexity and retaining the main features;
[0093] The transcription layer is used to convert the convolutional output features into a time series structure to adapt to the subsequent temporal modeling network. The formula for the convolution operation is as follows:
[0094]
[0095] In the formula: f and x are functions of t; * is the convolution operation; f(i) is the value corresponding to i; h(t - i) is the weight corresponding to i.
[0096] The bidirectional gated recurrent unit (BiGRU) is used to extract the bidirectional dependence features in the water quality time series, enhancing the model's learning ability for long-term change trends. The mathematical model of BiGRU is as follows:
[0097] r t =σ(W r ·[h t-1 ,x t );
[0098] u t =σ(W u ·[h t-1 ,x t );
[0099]
[0100] In the formula: is the candidate current state, that is, the new candidate state calculated through the reset gate; h t is the current state, that is, the final hidden state controlled by the update gate; W r , W u , are weight matrices; Xt is the bias vector; σ(·) is the Sigmoid activation function.
[0101] Through the forward GRU, h0, h1,... h t are calculated, and through the GRU, l0, l1,... l t are calculated. Then, the features in the two directions are concatenated to obtain the final output Rt:
[0102] R t = [h t , l t
[0103] In the formula: h t is the forward output of the memory unit at time t; l t is the backward output of the memory unit at time t.
[0104] The multi-head attention mechanism (MHA) is used to allocate attention weights between different time steps, improving the model's ability to identify key time segments. The mathematical formula of the multi-head attention mechanism is as follows:
[0105] Attention(Q, K, V = Concat(head1,…, head h )W O ;
[0106] In the formula: Q, K, and V respectively represent the matrix forms of query, key, and value. head i represents the output of the i-th attention head, which is obtained through the following calculation:
[0107]
[0108] In the formula: and are respectively the weight matrices of the query, key, and value of the i-th attention head. W O is the weight matrix that concatenates the outputs of all attention heads and maps them to the final result.
[0109] The fully connected layer is used to map the extracted comprehensive features to the prediction output space, generating the predicted values of the target water quality indicators.
[0110] In the model training process, the supervised learning method is adopted. The input is the water quality sequence after spatio-temporal fusion, the output is the predicted value of the water quality at the target time, the loss function is the mean square error (MSE), and the optimization algorithm can adopt Nadam.
[0111] Step Four: Hyperparameter optimization.
[0112] Specifically, to improve the prediction accuracy and training efficiency of the model, the osprey optimization algorithm (OOA) is used to automatically tune the hyperparameters of the deep learning model.
[0113] Specifically, the hyperparameter optimization steps include initializing the osprey population, constructing the hyperparameter search space, setting the fitness function, performing the iterative optimization process, and outputting the optimal hyperparameter combination.
[0114] Initialize the osprey population to set the population size, initial position, and maximum number of iterations of the algorithm, and determine the basic conditions for the optimization process;
[0115] Set the initial hyperparameter search space, including the size, number, and stride of the convolutional kernel, the number of BiGRU units, the learning rate, the batch size, etc.;
[0116] The fitness function is used to evaluate the quality of each set of hyperparameter combinations, and the mean squared error (MSE) on the validation set is selected as the measurement standard;
[0117] The iterative optimization process simulates the foraging behavior of ospreys. Through global search and local adjustment, the individual positions are continuously updated to improve the overall fitness value;
[0118] The optimal hyperparameter combination is output after meeting the convergence condition or reaching the maximum number of iterations, and is used to construct the final water quality prediction model.
[0119] Through the above hyperparameter optimization process, the prediction performance and generalization ability of the model can be effectively improved, and the subjectivity and inefficiency in the manual hyperparameter tuning process can be avoided.
[0120] Step Five: Water quality index prediction.
[0121] Specifically, apply the trained and hyperparameter-optimized hybrid water quality prediction model to the water quality data sequence to be predicted, and generate the predicted values of the water quality indexes at each monitoring section in the future.
[0122] Specifically, the water quality index prediction steps include inputting the data to be predicted, performing model inference, outputting the prediction results, and conducting error evaluation and result display.
[0123] The input data to be predicted is time series data processed by the same preprocessing method as in the training stage, including the fused input of the target monitoring section and its adjacent upstream section;
[0124] The model inference process is based on the optimized CNN-BiGRU-MHA model structure, extracts features from the input data and performs time series modeling, and outputs the predicted values for one or more future time steps;
[0125] The output prediction results include at least one water quality index, and the indexes include one or more of COD, NH3-N, TP, and DO. Single-index or multi-index prediction can be performed according to actual needs;
[0126] Error evaluation includes calculating MAE, RMSE, MAPE, R 2 indexes between the predicted values and the true values, which are used to evaluate the prediction performance of the model;
[0127] The result display method may include a time series trend chart to assist water environment management personnel in analysis, judgment, and decision-making support.
[0128] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and utilize the present invention well. The present invention is only limited by the claims and their full scope and equivalents.
Claims
1. A water quality prediction method integrating spatio-temporal features, characterized in that, The steps are as follows: S1: Preprocess the water quality data to generate a time series dataset for model training; S2: Construct a spatial weight matrix, and generate time series input data containing spatial information by setting spatial adjacency relationships and weight coefficients and using a weighted fusion method; S3: Construct a deep prediction model integrating a convolutional neural network, a bidirectional gated recurrent unit, and a multi-head attention mechanism; S4: Optimize the hyperparameters of the deep prediction model using the osprey optimization algorithm to obtain the optimal parameter combination for final training; S5: Input the historical water quality monitoring data and output the predicted values of key indicators at the target section in the future period.
2. The water quality prediction method integrating spatio-temporal features according to claim 1, wherein, In step S1, the steps are as follows: Preprocess the original water quality time series data collected from multiple monitoring sections through missing value filling, outlier replacement, wavelet denoising, sliding window partitioning, normalization processing, and test set partitioning.
3. A water quality prediction method integrating spatio-temporal features according to claim 2, characterized in that, The missing value filling uses the linear interpolation method to complete the missing values, and the expression is as follows: Where: X k is a missing value; X m is the previous known value of X k ; X n is the next known value of X k . The normalization operation uses the maximum-minimum normalization method, and the expression is as follows: In the formula: y is the normalized data, x is the data to be normalized, max(x) is the maximum value of the data, and min(x) is the minimum value of the data.
4. The water quality prediction method integrating spatio-temporal features according to claim 2, characterized in that The outlier replacement judges outliers based on 3σ and replaces them with adjacent values; The wavelet denoising uses the wavelet transform method, selects coif10 as the wavelet basis function, and sets a soft threshold to remove high-frequency noise; The sliding window partitioning method is used to construct time series training samples, and the window length is set according to the periodic characteristics of the water quality data; The data partitioning divides the processed sample set into a training set and a test set.
5. A water quality prediction method integrating spatio-temporal features according to claim 1, characterized in that, In step S2, the steps are as follows: Based on the correlation factors of geographical distance, basin water system structure, and water quality indicators, construct the spatial adjacency relationship between multiple monitoring sections; Use the weighted method based on the Pearson correlation coefficient to construct the spatial weight matrix; Perform weighted averaging on the historical data of the target monitoring section and the data of its spatially adjacent sections to generate an input sequence with spatial characteristics.
6. A water quality prediction method integrating spatio-temporal features according to claim 1, characterized in that, In step S3, the steps are as follows: S31: Extract spatial features through a convolutional neural network; S32: Extract the bidirectional dependence features in the water quality time series through a bidirectional gated recurrent unit; S33: Allocate attention weights between different time steps through a multi-head attention mechanism; S34: Construct a deep learning model and use the training set for training, and the validation set is used for parameter tuning.
7. A water quality prediction method integrating spatio-temporal features according to claim 1, characterized in that In step S4, the osprey optimization algorithm is used to optimize the hyperparameters of the deep prediction model, and the steps are as follows: S41: Initialize the osprey population; S42: Construct a hyperparameter search space through parameter settings; S43: Set the fitness function as the prediction error index; S44: Execute the iterative optimization process; S45: Output the optimal hyperparameter combination.
8. A water quality prediction method integrating spatio-temporal features according to claim 7, characterized in that In step S44, the iterative optimization process simulates the foraging behavior of ospreys, and continuously updates the individual positions through global search and local adjustment; In step S45, the optimal hyperparameter combination is output after meeting the convergence condition or reaching the maximum number of iterations, and is used to construct the final water quality prediction model.
9. The water quality prediction method integrating spatio-temporal features according to claim 1, wherein In step S5, the steps are as follows: Apply the hybrid water quality prediction model that has been trained and optimized for hyperparameters to the water quality data sequence to be predicted, and generate the predicted values of water quality indicators at each monitoring section for future time points.
10. A water quality prediction method integrating spatio-temporal features according to claim 9, characterized in that, In step S5, the obtained prediction results include one or more indicators of COD, NH3-N, TP, and DO.