A surface water prediction system and method based on hydrological optimization informer model

Abstract

The present invention discloses a surface water prediction system and method based on a hydrologically optimized Informer model, comprising the following steps: collecting multi-source data of upstream monitoring points and downstream monitoring points of surface water; preprocessing the multi-source data to obtain preprocessed data; extracting features from the preprocessed data to obtain water quality characteristics and hydrological spatiotemporal characteristics of the upstream monitoring points and the downstream monitoring points, and constructing cumulative dissolved oxygen saturation characteristics; inputting the upstream and downstream water quality characteristics, hydrological spatiotemporal characteristics, and dissolved oxygen saturation characteristics into a pre-trained multimodal prediction model of an improved Informer model to obtain prediction results of upstream surface water and downstream surface water; and outputting the fused surface water prediction results from the multimodal prediction model of the improved Informer model; the present invention optimizes the Informer model and combines it with a GRU model, using an attention mechanism to introduce hydrological spatiotemporal characteristics and cumulative dissolved oxygen saturation, thereby improving prediction accuracy, reducing errors, and enhancing the early warning capability for sudden pollution incidents.

Description

Technical Field

[0001] The present invention relates to the technical field of surface water prediction based on an improved Informer model, and in particular to a surface water prediction system and method based on a hydrologically optimized Informer model. Background Art

[0002] Surface water is a vital component of water resources. Changes in its quantity and quality have profound impacts on ecosystems, agricultural production, industrial water use, and human health. With global climate change and intensified human activities, the dynamics of surface water are becoming increasingly complex. Accurately predicting surface water changes is crucial for the rational use of water resources, the protection of the aquatic environment, and flood and drought prevention.

[0003] Traditional surface water forecasting methods mainly rely on the assumption of physical processes and predict surface water changes by simulating hydrological processes such as precipitation, evaporation, and runoff. These methods include:

[0004] Hydrological models: such as the linear ARDL model and the nonlinear Informer model. Although these models can capture certain hydrological characteristics, they have limitations in processing complex nonlinear and non-stationary data.

[0005] Time series models: such as the ARMA model. Although they can use historical data for prediction, they have high computational complexity when processing long series of data and have difficulty capturing global features.

[0006] Existing models have limited consideration of hydrological conditions and often ignore the continuous impact of environmental factors such as hydrological spatiotemporal characteristics and dissolved oxygen saturation in water on surface water quality, resulting in the inability of prediction models to accurately predict the changing trends of surface water quantity, water quality and water level. Summary of the Invention

[0007] The purpose of the present invention is to provide a surface water prediction system and method based on a hydrological optimization Informer model to solve the above-mentioned problems existing in the prior art.

[0008] The specific application is as follows:

[0009] A surface water prediction method based on a hydrological optimization informer model includes the following steps:

[0010] S1. Collect multi-source data related to surface water, including upstream and downstream monitoring points, including hydrological data, meteorological data, topographic data, and water quality data;

[0011] S2. Preprocessing the multi-source data to obtain preprocessed data; the preprocessing includes data cleaning, normalization, and missing value filling;

[0012] S3. Extracting features from the preprocessed data to obtain water quality characteristics and hydrological spatiotemporal characteristics of the upstream and downstream monitoring points, and constructing cumulative dissolved oxygen saturation characteristics, wherein the hydrological spatiotemporal characteristics include hydrological time characteristics and hydrological spatial characteristics;

[0013] S4. Input the upstream and downstream water quality characteristics, hydrological spatiotemporal characteristics, and dissolved oxygen saturation characteristics into the pre-trained multimodal prediction model of the improved Informer model to obtain the prediction results of upstream surface water and downstream surface water. Use the mean square error method to fuse the prediction results of upstream surface water and downstream surface water, and use the genetic algorithm to optimize the model weights so that the prediction results of the multimodal prediction model of the improved Informer model can be adjusted in different scenarios.

[0014] S5. Improve the surface water prediction results obtained by fusion of the multimodal prediction model output of the Informer model, preset warning thresholds, and if the prediction results meet the preset warning thresholds, issue early warnings of surface water emergencies at the monitoring points; the prediction results include changes in water quantity, water quality, and water level.

[0015] Furthermore, step S3 includes the following steps:

[0016] Perform adversarial perturbations on all water quality characteristics and hydrological spatiotemporal characteristics in the preprocessed data to create random forest trees; calculate the normalized prediction accuracy of out-of-sample data For out-of-sample data feature X e Apply the adversarial perturbation factor and calculate the normalized prediction accuracy after the e-th feature perturbation Calculate X e Importance of P e , the calculation expression is:

[0017]

[0018] Among them, P e is the importance of the e-th feature, G is the number of training samples;

[0019] P e Sort in ascending order to get the feature importance ranking Q e ; Then build the SelectKBest scoring system; the sorted features Q e , input the SelectKBest scoring system to obtain the final SelectKBest scoring system score of each feature, and select features as subsequent model input based on the SelectKBest scoring system score.

[0020] Furthermore, the calculation method of the cumulative dissolved oxygen saturation characteristic in step S3 includes:

[0021] The dissolved oxygen saturation data of the cumulative period are recorded in reverse order to obtain the dissolved oxygen saturation data set. The influence on the dissolved oxygen saturation at time t is calculated according to Henry's law. The calculation expression is:

[0022]

[0023] Among them, BF t is the air pressure influence intensity at time t, ABC is the saturated dissolved oxygen concentration per unit time, T t is the air pressure intensity at time t;

[0024] Calculate the time weight at time t. The calculation expression is:

[0025] BS t =10 a*(t-lt) ,

[0026] Among them, BS t is the time weight, lt is the delay (h), and a is the time weight coefficient;

[0027] Calculate the cumulative dissolved oxygen saturation after removing the time delay. The calculation expression is:

[0028]

[0029] Wherein, AR is the cumulative dissolved oxygen saturation, f represents the starting point at time t=f, and a1 represents the end point at time t=a1.

[0030] Furthermore, the multimodal prediction model of the improved Informer model in step S4 includes a GRU model and an Informer model, wherein the GRU model obtains the prediction result of the upstream monitoring point based on the water quality characteristics, hydrological spatiotemporal characteristics and cumulative dissolved oxygen saturation characteristics of the upstream monitoring point, and the Informer model obtains the prediction result of the downstream surface water based on the water quality characteristics and hydrological spatiotemporal characteristics of the downstream monitoring point.

[0031] Furthermore, the specific steps of the training process of the multimodal prediction model of the improved Informer model pre-trained in step S4 include:

[0032] S41, inputting the water quality characteristics of the upstream monitoring point, the hydrological time characteristics in the hydrological spatiotemporal characteristics, and the cumulative dissolved oxygen saturation characteristics into the GRU model to obtain the prediction result at the upstream monitoring point;

[0033] S42, inputting the water quality characteristics of the downstream monitoring point, the hydrological time characteristics, and the hydrological space characteristics in the hydrological spatiotemporal characteristics into the Informer model to obtain a prediction result of the downstream surface water;

[0034] S43. The Informer model and the GRU model are used as basic training models, and the models are fused according to the mean square error method to obtain a multimodal prediction model of the improved Informer model.

[0035] Furthermore, in step S41, the GRU model is an improved GRU model, and the improvement method is specifically as follows:

[0036] S411, input feature T i Converted into a serialized time dataset L Ti ;

[0037] S412, using attention mechanism to improve information acquisition of GRU algorithm;

[0038] Get GRU processing L Ti The hidden state is obtained when G=(g i-k ,…,g i ) indicates that the time format matrix is ​​extracted using two-dimensional convolution, and the calculation expression is:

[0039]

[0040] Among them, C is an iterator, is the feature value extracted by the qth iterator at time m, k is the preset iterator length, n is the padding length of the two-dimensional convolution calculation, and the time formats include short window 6 hours, medium window 12 hours, and long window 24 hours;

[0041] The attention formula is obtained by calculating the weight:

[0042]

[0043] Among them, MT i is the attention feature vector, σ is the Sigmoid activation function, MA is the weight matrix, n1 means at time n1, g i Represents the serialized time dataset L Ti The corresponding hidden state;

[0044] MT i With g i The final predicted value is obtained by adding the matrix after mapping;

[0045] S413. Use genetic algorithm combined with whale algorithm to optimize the time step, learning rate and filter length of the GRU model with the attention mechanism;

[0046] S414. Input the optimized attention vector, time step, learning rate, and iterator length into the GRU model to obtain the prediction result.

[0047] Furthermore, in step S42, the Informer model is an improved Informer model, and the improvement method is specifically as follows:

[0048] S421. Construct a time feature code. Decompose the hydrological time feature in the hydrological spatiotemporal feature through EMD to obtain discrete IMF components and continuous IMF components. Use the sine function feature coding method to generate the feature code of the discrete IMF component at time d:

[0049]

[0050] Among them, Z is the time encoding method, is the sine function encoding under Z time encoding mode, per Z is the discrete IMF component, d represents the preset period length under the Z time encoding method;

[0051] The CountEncoder encoding method is used to generate the feature encoding of the continuous IMF component at time d;

[0052] S422. Construct the longitude and latitude coding of spatial features, use Gaussian kernel function mapping to encode the hydrological spatial features in the hydrological spatiotemporal features, and encode the basin topology relationship:

[0053]

[0054] Among them, c lon , c lat Represents the longitude and latitude coordinates of the basin center, lon i ,lat i They represent the longitude and latitude coordinates of the monitoring point respectively, β represents the spatial attenuation coefficient, Represents the spatial feature code of the x-th monitoring point, wherein the hydrological spatial feature includes the latitude and longitude coordinate vectors of the monitoring point and the key image frame features;

[0055]

[0056] Among them, Y (H1+L1) It is represented as a representation vector after H1 graph convolution and L1 graph pooling. and are the associated representations of the key image frame and its adjacent key image frames in the H1th graph convolution, Yconv(·) represents the graph convolution layer, ReLU represents the activation function, and Ypool(·) represents the graph pooling layer;

[0057] S424. Use a multi-scale three-layer perceptron (MLP) to process the input temporal feature encoding and spatial feature encoding, respectively processing them into a long-window temporal feature set L, a medium-window temporal feature set M, a short-window temporal feature set S, and a spatial feature set D, to obtain the final multi-scale perceptual fusion feature:

[0058]

[0059] Among them, E represents the hydrological spatial feature encoding in the hydrological spatiotemporal feature, H represents the key image frame feature encoding in the hydrological spatial feature, NOR represents the normalization algorithm, MLP represents the multi-layer perceptron operation, b1 represents the feature set of different scales, represents splicing, G′ LMSD represents the normalized fusion feature, G LMSD Represents the final multi-scale perception fusion feature, and the spatial feature set D represents the final spatial embedding feature vector obtained by splicing the hydrological spatial feature code E in the hydrological spatiotemporal feature and the key image frame feature code H in the hydrological spatial feature through the multi-scale perception machine MLP;

[0060] S425. Input the obtained multi-scale perception fusion features into the Informer model as input feature encoding to obtain an improved Informer model. The improved Informer model also introduces a window sparse self-attention mechanism, a distillation mechanism and a fully connected output mechanism. The output of the convolutional neural network is used to access the first input channel of the improved Informer model.

[0061] Furthermore, the window sparse self-attention mechanism in step 425 includes:

[0062] Use sliding window combined with Fourier transform to characterize the self-attention mechanism;

[0063] Use a sliding window to limit the scope of the attention mechanism so that it only calculates the attention weight within a local window;

[0064] Using Fourier transform to convert time series data from the time domain to the frequency domain can not only capture local time fluctuations but also process global trends.

[0065] Furthermore, the calculation expression for model fusion according to the mean square error method is:

[0066] f F =W I *f d +W T *f D ,

[0067]

[0068] Among them, f d is the predicted value of upstream surface water obtained by the Informer model, f D is the predicted value of downstream surface water obtained by the GRU model, f F is the final prediction value after fusion, W I 、W T are the weight coefficients of the Informer model and the GRU model, θ I ,θ L Represent the mean square error of the prediction results of the Informer model and the GRU model respectively.

[0069] A surface water prediction system based on a hydrological optimization informer model, the system being used to execute any one of the surface water prediction methods based on a hydrological optimization informer model, the system comprising a data collection module, a data preprocessing module, a feature extraction module, a model training module, and a model prediction module;

[0070] The data collection module is used to collect multi-source data related to surface water, including upstream monitoring points and downstream monitoring points, and the data includes hydrological data, meteorological data, topographic data and water quality data;

[0071] The data preprocessing module is used to preprocess the multi-source data to obtain preprocessed data; the preprocessing includes data cleaning, normalization and missing value filling;

[0072] The feature extraction module is used to extract features from the preprocessed data to obtain water quality features and hydrological spatiotemporal features of the upstream monitoring point and the downstream monitoring point, and to construct cumulative dissolved oxygen saturation features, wherein the hydrological spatiotemporal features include hydrological time features and hydrological space features;

[0073] The model training module is used to input upstream and downstream water quality characteristics, hydrological spatiotemporal characteristics, and dissolved oxygen saturation characteristics into a pre-trained multimodal prediction model of an improved Informer model to obtain upstream surface water prediction results and downstream surface water prediction results, and to fuse the upstream surface water prediction results and the downstream surface water prediction results using a mean square error method;

[0074] The model prediction module is used to improve the surface water prediction results obtained by fusion of the multimodal prediction model output of the Informer model, and the prediction results include changes in water quantity, water quality and water level.

[0075] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0076] An embodiment of the present invention provides a method for collecting multi-source data related to surface water, including upstream monitoring points and downstream monitoring points; preprocessing the multi-source data to obtain preprocessed data; performing feature extraction on the preprocessed data to obtain water quality characteristics and hydrological spatiotemporal characteristics of the upstream monitoring points and the downstream monitoring points, and constructing cumulative dissolved oxygen saturation characteristics; inputting the upstream and downstream water quality characteristics, hydrological spatiotemporal characteristics and dissolved oxygen saturation characteristics into a multimodal prediction model of a pre-trained improved Informer model to obtain prediction results of upstream surface water and downstream surface water; the multimodal prediction model of the improved Informer model outputs the fused surface water prediction results; the present invention improves calculation accuracy, reduces surface water prediction errors and provides early warning for sudden pollution by optimizing the Informer model through hydrology and introducing an attention connection between cumulative dissolved oxygen saturation and hydrological spatiotemporal characteristics in combination with a GRU model. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a flow chart of a surface water prediction method based on a hydrological optimization informer model provided by an embodiment of the present invention;

[0078] Figure 2 This is an architecture diagram of a surface water prediction system based on a hydrological optimization informer model provided by an embodiment of the present invention;

[0079] Figure 3 This is an Informer model architecture diagram of a surface water prediction system based on a hydrological optimization Informer model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0080] The present invention will be described in detail below with reference to the accompanying drawings.

[0081] Example 1

[0082] First, the technical terms involved in the embodiments of the present application are explained.

[0083] (1) Informer model

[0084] The Informer model is an efficient time series prediction model, particularly suitable for processing long sequence data. It significantly improves the model's computational efficiency and prediction performance by introducing a sparse self-attention mechanism and a series of optimization techniques. The Informer model has performed well in fields such as meteorological forecasting, financial time series analysis, and hydrological forecasting.

[0085] (2) GRU model

[0086] GRU (Gated Recurrent Unit) is an improved recurrent neural network (RNN) architecture used to process long-term and short-term dependencies in sequence data. GRU controls the flow of information by introducing update gates and reset gates, effectively alleviating the vanishing gradient problem of traditional RNNs when processing long sequences. The GRU model has been widely used in natural language processing, time series prediction, speech recognition and other fields.

[0087] The embodiment of the present invention provides a surface water prediction method based on a hydrological optimization informer model, such as Figure 1 , including the following steps:

[0088] S1. Collect multi-source data related to surface water, including upstream and downstream monitoring points, including hydrological data, meteorological data, topographic data, and water quality data;

[0089] S2. Preprocessing the multi-source data to obtain preprocessed data; the preprocessing includes data cleaning, normalization, and missing value filling;

[0090] S3. Extracting features from the preprocessed data to obtain water quality characteristics and hydrological spatiotemporal characteristics of the upstream and downstream monitoring points, and constructing cumulative dissolved oxygen saturation characteristics, wherein the hydrological spatiotemporal characteristics include hydrological time characteristics and hydrological spatial characteristics;

[0091] S4. Input the upstream and downstream water quality characteristics, hydrological spatiotemporal characteristics, and dissolved oxygen saturation characteristics into the pre-trained multimodal prediction model of the improved Informer model to obtain the prediction results of upstream surface water and downstream surface water. Use the mean square error method to fuse the prediction results of upstream surface water and downstream surface water, and use the genetic algorithm to optimize the model weights so that the prediction results of the multimodal prediction model of the improved Informer model can be adjusted in different scenarios.

[0092] S5. Improve the surface water prediction results obtained by fusion of the multimodal prediction model output of the Informer model, preset warning thresholds, and if the prediction results meet the preset warning thresholds, issue an early warning of surface water emergencies at the monitoring point; the prediction results include changes in water quantity, water quality, and water level.

[0093] Specifically, the method collects multi-source data of surface water including upstream monitoring points and downstream monitoring points; pre-processes the multi-source data to obtain pre-processed data; performs feature extraction on the pre-processed data to obtain water quality characteristics and hydrological spatiotemporal characteristics of the upstream monitoring points and the downstream monitoring points, and constructs cumulative dissolved oxygen saturation characteristics; inputs the upstream and downstream water quality characteristics, hydrological spatiotemporal characteristics and dissolved oxygen saturation characteristics into a multimodal prediction model of a pre-trained improved Informer model to obtain prediction results of upstream surface water and downstream surface water; the multimodal prediction model of the improved Informer model outputs the fused surface water prediction results; this embodiment improves calculation accuracy, reduces surface water prediction errors and provides early warning for sudden pollution by optimizing the Informer model and combining it with the GRU model. The sudden surface water conditions at least include sudden surface water pollution, sudden surface water floods, etc.

[0094] It should be noted that the Informer model is suitable for hydrological prediction. For example, its long sequence capture capability is better than other time series capture models such as LSTM. Therefore, the present invention uses the Informer model for hydrological prediction.

[0095] By introducing cumulative dissolved oxygen saturation and based on historical time series relationships, the Informer model excels in capturing long sequences, enhancing the model's long-term prediction capabilities and improving its prediction accuracy, which is one of the technical highlights of this invention.

[0096] It should be noted that the collection and pre-treatment of surface water in step S1 and step S2 include:

[0097] Collection of original water quality time series data at each monitoring point (pH, DO, NH3-N, TP and other water quality parameters, sampling frequency of each monitoring point once every half hour): time series concentration data of each monitoring point within 24 hours;

[0098] The data collection and processing process is as follows:

[0099] Real-time fluctuation calculation of each monitoring point, and hourly change rate calculation of each collected water quality parameter Dynamically evaluate the stability of data collection, where r t : real-time volatility at time t (dimensionless), x t : Water quality parameter value at time t (such as DO concentration, unit: mg / L), x t-1 : water quality parameter value at time t-1;

[0100] Window adaptive adjustment collection mode:

[0101] Stable period collection (|r t∣<5%): Use a long window (24 hours) to enhance periodic characteristics, suitable for stable periods (such as slow changes in DO during the dry season), and smooth out diurnal cycle noise;

[0102] Fluctuation period collection (5%≤|r t ∣≤15%): Switch to the medium window (12 hours), balance trends and details, cope with the impact of early rainfall or tides, and retain the semi-diurnal cycle characteristics;

[0103] Mutation period collection (|r t >15%): Enable a short window (6 hours) to quickly respond to sudden changes and minimize lag effects for sudden pollution (such as factory leaks);

[0104] The window adaptive adjustment acquisition method adopts multi-scale feature generation and parallel computing:

[0105] For each collected data, the moving average values ​​of the three windows of 6h / 12h / 24h are calculated independently to generate three types of time series features:

[0106]

[0107] Where win represents the window hours, Win∈{6,12,24};

[0108] Data preprocessing, filling missing values, and using reflection padding to avoid sequence truncation;

[0109] An example of mirror filling in this embodiment:

[0110] Original DO sequence: [8.2,8.5,7.9,...,9.1],

[0111] 6h window MA (capture hourly fluctuations): [8.3, 8.2, ..., 8.9],

[0112] 24h window MA (highlighting daily cycle trends): [8.4,8.4,...,8.6].

[0113] In the above embodiment, specifically, step S3 includes the following steps:

[0114] Perform adversarial perturbations on all water quality characteristics and hydrological spatiotemporal characteristics in the preprocessed data to create random forest trees; calculate the normalized prediction accuracy of out-of-sample data For out-of-sample data feature X e Apply the adversarial perturbation factor and calculate the normalized prediction accuracy after the e-th feature perturbation Calculate X e Importance of P e, the calculation expression is:

[0115]

[0116] Among them, P e is the importance of the e-th feature, G is the number of training samples;

[0117] P e Sort in ascending order to get the feature importance ranking Q e ; Then build the SelectKBest scoring system; the sorted features Q e , input the SelectKBest scoring system to obtain the final SelectKBest scoring system score of each feature, and select features as subsequent model input based on the SelectKBest scoring system score.

[0118] It should be noted that by performing counter-disturbance on water quality characteristics and hydrological spatiotemporal characteristics, a SelectKBest scoring system is constructed, and features are screened according to the scores. Compared with the existing technology, the amount of training data for subsequent models is reduced, the training time of subsequent models is increased, and the accuracy of feature screening is improved. In addition to applying counter-disturbance factors, other algorithm factors can also be added, including but not limited to Gaussian noise factors, Poisson noise factors, etc., which will not be repeated in this embodiment.

[0119] In the above embodiment, specifically, the calculation method of the cumulative dissolved oxygen saturation characteristic in step S3 includes:

[0120] The dissolved oxygen saturation data of the cumulative period are recorded in reverse order to obtain the dissolved oxygen saturation data set. The influence on the dissolved oxygen saturation at time t is calculated according to Henry's law. The calculation expression is:

[0121]

[0122] Among them, BF t is the air pressure influence intensity at time t, ABC is the saturated dissolved oxygen concentration per unit time, T r is the air pressure intensity at time t;

[0123] Calculate the time weight at time t. The calculation expression is:

[0124] BS t =10 a*(t-lt) ,

[0125] Among them, BS t is the time weight, lt is the delay, and a is the time weight coefficient;

[0126] Calculate the cumulative dissolved oxygen saturation after removing the time delay. The calculation expression is:

[0127]

[0128] Wherein, AR is the cumulative dissolved oxygen saturation, f represents the starting point at time t=f, and a1 represents the end point at time t=a1.

[0129] It should be noted that existing models have limited consideration of hydrology and often ignore the continuous impact of environmental factors such as dissolved oxygen saturation on water quality disasters and sudden pollution. As a result, the prediction model cannot accurately describe the changing trends of water quality. This embodiment introduces the principle that the decomposition of substances in water bodies is highly correlated with the dissolved oxygen saturation characteristics. It continuously analyzes the dissolved oxygen saturation characteristics through the Informer neural network, and thus predicts sudden water pollution accidents of surface water with high precision.

[0130] In the above embodiment, specifically, the multimodal prediction model of the improved Informer model in step S4 includes a GRU model and an Informer model, the GRU model obtains the prediction result of the upstream monitoring point based on the water quality characteristics, hydrological spatiotemporal characteristics and cumulative dissolved oxygen saturation characteristics of the upstream monitoring point, and the Informer model obtains the prediction result of the downstream surface water based on the water quality characteristics and hydrological spatiotemporal characteristics of the downstream monitoring point.

[0131] It should be noted that this embodiment predicts surface water by selecting the fusion of the GRU model and the Informer model. This is because the advantages of the GRU model in processing time series are combined with the advantages of the Informer model in processing spatiotemporal series. After the fusion of the two models, the accuracy of the prediction analysis of the fusion model is improved. Compared with the existing technology, the fusion of the two models is the technical highlight of the present invention.

[0132] In the above embodiment, specifically, the specific steps of the training process of the multimodal prediction model of the improved Informer model pre-trained in step S4 include:

[0133] S41, inputting the water quality characteristics of the upstream monitoring point, the hydrological time characteristics in the hydrological spatiotemporal characteristics, and the cumulative dissolved oxygen saturation characteristics into the GRU model to obtain the prediction result at the upstream monitoring point;

[0134] S42, inputting the water quality characteristics of the downstream monitoring point, the hydrological time characteristics, and the hydrological space characteristics in the hydrological spatiotemporal characteristics into the Informer model to obtain a prediction result of the downstream surface water;

[0135] S43. The Informer model and the GRU model are used as basic training models, and the models are fused according to the mean square error method to obtain a multimodal prediction model of the improved Informer model.

[0136] In the above embodiment, specifically, in step S41, the GRU model is an improved GRU model, and the improvement method is specifically as follows:

[0137] S411, input feature T i Converted into a serialized time dataset L Ti ;

[0138] S412, using attention mechanism to improve information acquisition of GRU algorithm;

[0139] Get GRU processing L Ti The hidden state is obtained when G=(g i-k ,…,g i ) indicates that the time format matrix is ​​extracted using two-dimensional convolution, and the calculation expression is:

[0140]

[0141] Among them, C is an iterator, is the feature value extracted by the qth iterator at time m, k is the preset iterator length, n is the padding length of the two-dimensional convolution calculation, and the time formats include short window 6 hours, medium window 12 hours, and long window 24 hours;

[0142] The attention formula is obtained by calculating the weight:

[0143]

[0144] Among them, MT i is the attention feature vector, σ is the Sigmoid activation function, MA is the weight matrix, n1 means at time n1, g i Represents a serialized time dataset The corresponding hidden state;

[0145] MT i With g i The final predicted value is obtained by adding the matrix after mapping;

[0146] S413. Use genetic algorithm combined with whale algorithm to optimize the time step, learning rate and filter length of the GRU model with the attention mechanism;

[0147] S414. Input the optimized attention vector, time step, learning rate, and iterator length into the GRU model to obtain the prediction result.

[0148] Specifically, the genetic algorithm optimizes the time step, learning rate, and filter length of the GRU model with the attention mechanism as follows:

[0149] M1: parameter encoding, using genetic algorithm to optimize the time step, learning rate and filter length in the attention mechanism, converting the time step, learning rate and filter length in the attention mechanism into binary digital form, and encoding the time step, learning rate and filter length in the attention mechanism into genetic algorithm antibodies;

[0150] M2: Calculate the fitness function, the time step, the learning rate, and the filter length optimization in the attention mechanism. The key issue is the selection of the fitness function. The absolute error integral is used as the evaluation indicator. The fitness function is:

[0151]

[0152] M3: Update memory cells. The high fitness of antibody cells is retained through the memory function and allocated to the time step, learning rate, and filter length in the attention mechanism. The individual fitness function value is calculated to determine the optimal fitness value of the population. If the searched optimal fitness value is less than the optimal fitness value in the immune network, the immune memory starts to search for the optimal value again and uses it as the optimal immune antibody. Otherwise, the calculated optimal fitness value is added to the antibody memory list.

[0153] M4: Maintain diversity and use selection probability to adjust the diversity of the population. The calculation formula is:

[0154]

[0155] Among them, a and b are random numbers in the interval [0, 1], F i (i) is the individual fitness value, C is the concentration of the antibody;

[0156] M5: Crossover operation, the population individuals are operated using a two-point crossover method, and the crossover equation is:

[0157]

[0158] Among them, X′ and Y′ are new individuals of the crossover, and r is a random number in the interval [0, 1];

[0159] M6: Mutation operation, using Gaussian mutation to operate on the time step, learning rate and filter length in the attention mechanism. Its mutation equation is:

[0160]

[0161] Where λ is a random number in the interval [0, 1], and μ is a Gaussian operator;

[0162] M7: The algorithm terminates and compares the fitness function values ​​of two adjacent times. If the error meets the conditions or reaches the maximum number of iterations, the algorithm terminates. The results are used to adjust the parameters of the time step, learning rate, and filter length in the attention mechanism in real time. Otherwise, go to step M2.

[0163] The whale algorithm will not be described in detail in this embodiment, and can be understood by looking up the existing technology.

[0164] It should be noted that by using genetic algorithms combined with the whale algorithm to optimize the time step, learning rate and filter length in the attention mechanism of the GRU model, the prediction accuracy of the GRU model is improved and the prediction speed of the GRU model is accelerated.

[0165] In the above embodiment, specifically, in step S42, the Informer model is an improved Informer model, and the improvement method is specifically as follows:

[0166] S421. Construct a time feature code. Decompose the hydrological time feature in the hydrological spatiotemporal feature through EMD to obtain discrete IMF components and continuous IMF components. Use the sine function feature coding method to generate the feature code of the discrete IMF component at time d:

[0167]

[0168] Among them, Z is the time encoding method, is the sine function encoding under Z time encoding mode, per Z is the discrete IMF component, d represents the preset period length under the Z time encoding method;

[0169] The CountEncoder encoding method is used to generate the feature encoding of the continuous IMF component at time d;

[0170] It should be noted that sine function feature encoding and CountEncoder encoding are used for IMF components respectively, because CountEncoder encoding has advantages in processing continuous IMF components, while sine function feature encoding has advantages in encoding discrete IMF components. This can improve encoding efficiency and accuracy, and provide better support for subsequent model training and recognition.

[0171] S422. Construct the longitude and latitude coding of spatial features, use Gaussian kernel function mapping to encode the hydrological spatial features in the hydrological spatiotemporal features, and encode the basin topology relationship:

[0172]

[0173] Among them, c lon , c latRepresents the longitude and latitude coordinates of the basin center, lon i ,lat i They represent the longitude and latitude coordinates of the monitoring point respectively, β represents the spatial attenuation coefficient, Represents the spatial feature code of the x-th monitoring point, wherein the hydrological spatial feature includes the latitude and longitude coordinate vectors of the monitoring point and the key image frame features;

[0174] It should be noted that the longitude and latitude codes of the spatial features are constructed, and the longitude and latitude codes correspond to the longitude and latitude position information of each monitoring point;

[0175]

[0176] Among them, Y (H1+L1) It is represented as a representation vector after H1 graph convolution and L1 graph pooling. and are the associated representations of the key image frame in the H1th graph convolution and its adjacent key image frames, Yconv(·) represents the graph convolution layer, ReLU represents the activation function, and Ypool(·) represents the graph pooling layer. The key image frame features represent the turbidity image features of the surface water surface of the monitoring point, the water level features of the surface water surface of the monitoring point, and the water flow features of the surface water surface of the monitoring point, as shown in Figure 2 As shown;

[0177] It should be noted that the key image frame feature encoding for constructing spatial features is performed on the key image frames collected at each monitoring point. The spatial feature encoding includes the latitude and longitude encoding of the spatial features corresponding to each monitoring point and the key image frame feature encoding. The convolutional neural network described in this embodiment adopts a feature pyramid network (FPN). The feature pyramid network constructs a multi-scale feature pyramid and uses feature maps at different levels to process targets of different scales, thereby improving the model's ability to process multi-scale data and improving the recognition of surface water pollution image features associated with the key image frame features of the monitoring points.

[0178] Specifically, the feature pyramid network in this embodiment is divided into four layers, and the specific processing process of the four layers of the feature pyramid network is as follows:

[0179] The first layer of the pyramid is used to process the first key frame image resolution, and the brightness of the image belongs to the first level of brightness;

[0180] The second layer of the pyramid is used to process the second key frame image resolution, and the brightness of the image belongs to the second level brightness;

[0181] The third layer of the pyramid is used to process the third key frame image resolution, and the brightness of the image belongs to the third level of brightness;

[0182] The fourth layer of the pyramid is used to process the image resolution of the fourth key frame, and the brightness of the image belongs to the fourth-level brightness;

[0183] The resolution levels of the said images are specifically divided as follows:

[0184] Resolution of the first key frame image: w < 150, h < 150;

[0185] Resolution of the second key frame image: 150 < w ≤ 320, 150 < h ≤ 350;

[0186] Resolution of the third key frame image: 350 < w ≤ 1080, 350 < h ≤ 1080;

[0187] Resolution of the fourth key frame image: 1080 < w, 1080 < h, where w represents the width of the image and h represents the height of the image;

[0188] The brightness of the said images adopts the method of calculating the mean brightness of the image by using a histogram, and is specifically divided as follows:

[0189] First mean brightness: 0 ≤ L ≤ 70;

[0190] Second mean brightness: 70 < L ≤ 110;

[0191] Third mean brightness: 110 < L ≤ 170;

[0192] Fourth mean brightness: 170 < L ≤ 255, where L represents the mean brightness value of the image;

[0193] Two branch networks are added behind the feature maps of each layer of the pyramid. One branch is used for classification and the other branch is used for regression. Each branch first performs 6 convolutions on the feature map to enhance the key image frames respectively. The convolution kernel size is 3×3 and the number is 64. By setting the feature recognition of four different brightness levels, the recognition degree of the turbidity of the surface water of the monitoring point by the model is improved.

[0194] S424. Use the multi-scale three-layer perceptron MLP to process the input temporal feature encoding and spatial feature encoding, and process them into the long-window temporal feature set L, the medium-window temporal feature set M, the short-window temporal feature set S and the spatial feature set D respectively, to obtain the final multi-scale perceptual fusion feature:

[0195]

[0196] Among them, E represents the hydrological spatial feature encoding in the hydrological spatio-temporal feature, H represents the key image frame feature encoding in the hydrological spatial feature, NOR represents the normalization algorithm, MLP represents the multi-layer perceptron operation, b1 represents different scale feature sets, represents splicing, G′LMSD represents the normalized fusion feature, G LMSD Represents the final multi-scale perception fusion feature, and the spatial feature set D represents the final spatial embedding feature vector obtained by splicing the hydrological spatial feature code E in the hydrological spatiotemporal feature and the key image frame feature code H in the hydrological spatial feature through the multi-scale perception machine MLP;

[0197] S425. Input the obtained multi-scale perception fusion features into the Informer model as input feature encoding to obtain an improved Informer model. The improved Informer model also introduces a window sparse self-attention mechanism, a distillation mechanism and a fully connected output mechanism. The output of the convolutional neural network is used to access the first input channel of the improved Informer model.

[0198] In the above embodiment, specifically, the window sparse self-attention mechanism in step 425 includes:

[0199] Use sliding window combined with Fourier transform to characterize the self-attention mechanism;

[0200] Use a sliding window to limit the scope of the attention mechanism so that it only calculates the attention weight within a local window;

[0201] Using Fourier transform to convert time series data from the time domain to the frequency domain can not only capture local time fluctuations but also process global trends.

[0202] It should be noted that the window sparse self-attention mechanism includes:

[0203] The self-attention mechanism is characterized using the window form. The original self-attention mechanism can be expressed as:

[0204]

[0205] After using the window form, the self-attention of the i-th q can be expressed as:

[0206]

[0207] Where Q, K, and V represent the query, key, and value matrices respectively, and q, k, and v represent the corresponding vectors respectively, softmax is the activation function, and d represents the dimension of the key.

[0208] In the above embodiment, specifically, the calculation expression for model fusion according to the mean square error method is:

[0209] f F =W I *f d +WT *f D ,

[0210]

[0211] Among them, f d is the predicted value of upstream surface water obtained by the Informer model, f D is the predicted value of downstream surface water obtained by the GRU model, f F is the final prediction value after fusion, W I 、W T are the weight coefficients of the Informer model and the GRU model, θ I ,θ L Represent the mean square error of the prediction results of the Informer model and the GRU model respectively.

[0212] It should be noted that the specific implementation process of optimizing the model weights using the genetic algorithm in step S4 is as follows:

[0213] A genetic algorithm (GA) is an optimization algorithm based on the principles of natural selection and genetics. It simulates the "selection," "crossover," and "mutation" operations of biological evolution to find the optimal solution. The following is a detailed process for optimizing the weights of the Informer and GRU models using a genetic algorithm:

[0214] 1. Initialize the population

[0215] Population size: A set of populations is randomly generated, each individual represents a set of model hyperparameter configurations.

[0216] Hyperparameter range: Set a reasonable value range for each hyperparameter, including learning rate, number of network layers, number of neurons, etc.

[0217] 2. Evaluate fitness

[0218] Model training: Train the Informer and GRU models using each set of hyperparameter configurations;

[0219] Performance evaluation: Evaluate model performance on the validation set. Common evaluation metrics include mean square error (MSE) and accuracy.

[0220] 3. Select an action

[0221] Selection strategy: select individuals with higher fitness to enter the next generation based on their performance evaluation. Common selection methods include roulette wheel selection and tournament selection.

[0222] 4. Crossover Operation

[0223] Crossover strategy: Perform a crossover operation on the selected individuals, exchanging some of their genes (i.e., hyperparameters) to generate new individuals;

[0224] 5. Mutation Operation

[0225] Mutation strategy: Make small modifications to certain genes of individuals according to a certain probability to increase population diversity;

[0226] 6. Iterative Optimization

[0227] Iterative process: Repeat the above selection, crossover and mutation operations until the predetermined number of iterations is reached or the stopping criterion is met;

[0228] Stop condition: Set the maximum number of iterations or the fitness value is less than the set threshold as the stopping condition;

[0229] 7. Output the optimal solution

[0230] Optimal parameters: After multiple rounds of iteration, find a set of optimal hyperparameter combinations;

[0231] Model training: Retrain the model using optimal hyperparameters to achieve the best performance;

[0232] Example 2

[0233] A surface water prediction system based on a hydrological optimization informer model, the system comprising a data collection module A1, a data preprocessing module A2, a feature extraction module A3, a model training module A4 and a model prediction module A5;

[0234] The data collection module A1 is used to collect multi-source data related to surface water, including upstream monitoring points and downstream monitoring points, and the data includes hydrological data, meteorological data, topographic data and water quality data;

[0235] The data preprocessing module A2 is used to preprocess the multi-source data to obtain preprocessed data; the preprocessing includes data cleaning, normalization and missing value filling;

[0236] The feature extraction module A3 is used to extract features from the preprocessed data to obtain water quality features and hydrological spatiotemporal features of upstream and downstream monitoring points, and to construct cumulative dissolved oxygen saturation features, wherein the hydrological spatiotemporal features include hydrological time features and hydrological spatial features;

[0237] The model training module A4 is used to input the water quality characteristics, hydrological spatiotemporal characteristics, and dissolved oxygen saturation characteristics of the upstream and downstream into the multimodal prediction model of the pre-trained improved Informer model to obtain the prediction results of the upstream surface water and the downstream surface water, fuse the prediction results of the upstream surface water and the downstream surface water using the mean square error method, and optimize the model weights using a genetic algorithm to make the prediction results of the multimodal prediction model of the improved Informer model adjustable in different scenarios;

[0238] The model prediction module A5 is used to improve the surface water prediction results obtained by fusion of the multimodal prediction model output of the Informer model, and the prediction results include changes in water quantity, water quality and water level.

[0239] It should be understood that the above embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations thereof based on the present invention; the variations and modifications made by ordinary technicians in this industry through the present invention without making groundbreaking innovations all fall within the scope of protection of the present invention.

Claims

1. A surface water prediction method based on a hydrological improved Informer model, characterized in that: The following steps are involved: S1. Collect multi-source data related to surface water, including upstream and downstream monitoring points, including hydrological data, meteorological data, topographic data, and water quality data; S2. Preprocessing the multi-source data to obtain preprocessed data; the preprocessing includes data cleaning, normalization, and missing value filling; S3. Extracting features from the preprocessed data to obtain water quality characteristics and hydrological spatiotemporal characteristics of the upstream and downstream monitoring points, and constructing cumulative dissolved oxygen saturation characteristics, wherein the hydrological spatiotemporal characteristics include hydrological time characteristics and hydrological spatial characteristics; S4. Inputting the water quality characteristics, hydrological spatiotemporal characteristics, and dissolved oxygen saturation characteristics of the upstream and downstream regions into a pre-trained multimodal prediction model to obtain the prediction results of the upstream surface water and the downstream surface water, and fusing the prediction results of the upstream surface water and the downstream surface water using the mean square error method; S5. Surface water prediction results obtained by fusion of multimodal prediction model outputs are combined with preset warning thresholds. If the prediction results meet the preset warning thresholds, an early warning of surface water emergencies at the monitoring point is issued; the prediction results include changes in water quantity, water quality, and water level; In step S4, the multimodal prediction model includes a GRU model and an Informer model. The GRU model obtains the prediction result of the upstream monitoring point based on the water quality characteristics, hydrological spatiotemporal characteristics, and cumulative dissolved oxygen saturation characteristics of the upstream monitoring point. The Informer model obtains the prediction result of the downstream surface water based on the water quality characteristics and hydrological spatiotemporal characteristics of the downstream monitoring point. The Informer model is an improved Informer model, and the improvement method is as follows: S421. Construct a time feature code. Decompose the hydrological time feature in the hydrological spatiotemporal feature through EMD to obtain discrete IMF components and continuous IMF components. Use the sine function feature coding method to generate the feature code of the discrete IMF component at time d: ， Among them, Z is the time encoding method, It is the sine function encoding under the Z time encoding method, is the discrete IMF component, d represents the preset period length under the Z time encoding method; The CountEncoder encoding method is used to generate the feature encoding of the continuous IMF component at time d; S422. Construct the longitude and latitude coding of spatial features, use Gaussian kernel function mapping to encode the hydrological spatial features in the hydrological spatiotemporal features, and encode the basin topology relationship: ， in, , Represent the longitude and latitude coordinates of the basin center, , Respectively represent the longitude and latitude coordinates of the monitoring point, represents the spatial attenuation coefficient, Represents the spatial feature code of the x-th monitoring point, wherein the hydrological spatial feature includes the latitude and longitude coordinate vectors of the monitoring point and the key image frame features; S423: Construct key image frame feature encoding of spatial features, and use a convolutional neural network to encode the key image frame features of the time series. This process can be expressed as: ， in, It is represented as a representation vector after H1 graph convolution and L1 graph pooling. and are the associated representations of the key image frame and its adjacent key image frames in the H1th graph convolution, Yconv(·) represents the graph convolution layer, ReLU represents the activation function, and Ypool(·) represents the graph pooling layer; S424. Use a multi-scale three-layer perceptron (MLP) to process the input temporal feature encoding and spatial feature encoding, respectively processing them into a long-window temporal feature set L, a medium-window temporal feature set M, a short-window temporal feature set S, and a spatial feature set D, to obtain the final multi-scale perceptual fusion feature: D = MLP(E⊕H), = NOR(MLP(b ={L,M,S,D}⊕ })), ， Among them, E represents the hydrological spatial feature encoding in the hydrological spatiotemporal feature, H represents the key image frame feature encoding in the hydrological spatial feature, NOR represents the normalization algorithm, MLP represents the multi-layer perceptron operation, b1 represents the feature set of different scales, ⊕ represents splicing, represents the normalized fusion features, Represents the final multi-scale perception fusion feature, and the spatial feature set D represents the final spatial embedding feature vector obtained by splicing the hydrological spatial feature code E in the hydrological spatiotemporal feature and the key image frame feature code H in the hydrological spatial feature through the multi-scale perception machine MLP; S425. Input the obtained multi-scale perception fusion features into the Informer model as input feature encoding to obtain an improved Informer model. The improved Informer model also introduces a window sparse self-attention mechanism, a distillation mechanism and a fully connected output mechanism. The output of the convolutional neural network is used to access the first input channel of the improved Informer model.

2. A surface water prediction method based on a hydrological improved Informer model according to claim 1, characterized in that: Step S3 includes the following steps: Perform adversarial perturbations on all water quality characteristics and hydrological spatiotemporal characteristics in the preprocessed data to create random forest trees; calculate the normalized prediction accuracy of out-of-sample data ; For out-of-sample data characteristics Apply the adversarial perturbation factor and calculate the normalized prediction accuracy after the e-th feature perturbation ;calculate Importance , the calculation expression is: ， in, is the importance of the e-th feature, G is the number of training samples; right Sort in ascending order to get the feature importance ranking ; Then build the SelectKBest scoring system; the sorted features , input the SelectKBest scoring system to obtain the final SelectKBest scoring system score of each feature, and select features as subsequent model input based on the SelectKBest scoring system score.

3. The surface water prediction method based on the hydrological improved Informer model according to claim 1 is characterized in that: The calculation method of the cumulative dissolved oxygen saturation characteristic in step S3 includes: The dissolved oxygen saturation data of the cumulative period are recorded in reverse order to obtain the dissolved oxygen saturation data set. The influence on the dissolved oxygen saturation at time t is calculated according to Henry's law. The calculation expression is: ， in, is the air pressure influence intensity at time t, ABC is the saturated dissolved oxygen concentration per unit time, is the air pressure intensity at time t; Calculate the time weight at time t. The calculation expression is: ， in, is the time weight, lt is the delay, and a is the time weight coefficient; Calculate the cumulative dissolved oxygen saturation after removing the time delay. The calculation expression is: ， Where AR is the cumulative dissolved oxygen saturation, f represents the starting point at t=f, and a1 represents the end point at t=a1.

4. The surface water prediction method based on the hydrological improved Informer model according to claim 1 is characterized in that: The specific steps of the training process of the pre-trained multimodal prediction model in step S4 include: S41, inputting the water quality characteristics of the upstream monitoring point, the hydrological time characteristics in the hydrological spatiotemporal characteristics, and the cumulative dissolved oxygen saturation characteristics into the GRU model to obtain the prediction result at the upstream monitoring point; S42, inputting the water quality characteristics of the downstream monitoring point, the hydrological time characteristics, and the hydrological space characteristics in the hydrological spatiotemporal characteristics into the Informer model to obtain a prediction result of the downstream surface water; S43. Use the Informer model and the GRU model as the basic training models, perform model fusion according to the mean square error method, and obtain a multimodal prediction model.

5. The surface water prediction method based on the hydrological improved Informer model according to claim 4 is characterized in that: In step S41, the GRU model is an improved GRU model, and the improvement method is specifically as follows: S411, input features Convert to serialized time dataset ; S412, using attention mechanism to improve information acquisition of GRU algorithm; Get GRU processing The hidden state obtained when ,…, ) indicates that the time format matrix is ​​extracted using two-dimensional convolution, and the calculation expression is: ， Among them, C is an iterator, is the feature value extracted by the qth iterator at time m, k is the preset iterator length, n is the padding length of the two-dimensional convolution calculation, and the time formats include short window 6 hours, medium window 12 hours, and long window 24 hours; The attention formula is obtained by calculating the weight: ， in, is the attention feature vector, σ is the Sigmoid activation function, MA is the weight matrix, n1 represents the time n1, Represents a serialized time dataset The corresponding hidden state; Will and The final predicted value is obtained by adding the matrix after mapping; S413. Use genetic algorithm combined with whale algorithm to optimize the time step, learning rate and filter length of the GRU model with the attention mechanism; S414. Input the optimized attention vector, time step, learning rate, and iterator length into the GRU model to obtain the prediction result.

6. The surface water prediction method based on the hydrological improved Informer model according to claim 1 is characterized in that: The window sparse self-attention mechanism in step 425 includes: Use sliding window combined with Fourier transform to characterize the self-attention mechanism; Use a sliding window to limit the scope of the attention mechanism so that it only calculates the attention weight within a local window; Using Fourier transform to convert time series data from the time domain to the frequency domain can not only capture local time fluctuations but also process global trends.

7. The surface water prediction method based on the hydrological improved Informer model according to claim 4 is characterized in that: The calculation expression for model fusion based on the mean square error method is: ， ， ， in, is the predicted value of upstream surface water obtained by the Informer model, is the predicted value of downstream surface water obtained by the GRU model, is the final prediction value after fusion, 、 are the weight coefficients of the Informer model and the GRU model respectively, 、 Represent the mean square error of the prediction results of the Informer model and the GRU model respectively.

8. A surface water prediction system based on a hydrological improved Informer model, characterized in that: The system is used to perform the method according to any one of claims 1 to 7, and the system includes a data collection module, a data preprocessing module, a feature extraction module, a model training module and a model prediction module; The data collection module is used to collect multi-source data related to surface water, including upstream monitoring points and downstream monitoring points, and the data includes hydrological data, meteorological data, topographic data and water quality data; The data preprocessing module is used to preprocess the multi-source data to obtain preprocessed data; The preprocessing includes data cleaning, normalization and missing value filling; The feature extraction module is used to extract features from the preprocessed data to obtain water quality features and hydrological spatiotemporal features of the upstream monitoring point and the downstream monitoring point, and to construct cumulative dissolved oxygen saturation features, wherein the hydrological spatiotemporal features include hydrological time features and hydrological space features; The model training module is used to input the water quality characteristics, hydrological spatiotemporal characteristics and dissolved oxygen saturation characteristics of the upstream and downstream into a pre-trained multimodal prediction model to obtain the prediction results of the upstream surface water and the downstream surface water, and to fuse the prediction results of the upstream surface water and the downstream surface water using the mean square error method; The model prediction module is used to predict the surface water obtained by fusion of the multimodal prediction model outputs, and the prediction results include changes in water quantity, water quality and water level.

Images