River basin water quality prediction and early warning method, equipment and medium based on multi-model processing technology

Abstract

The present invention belongs to the field of water quality prediction technology, and particularly relates to a river basin water quality prediction and early warning method, device, and medium based on multi-model processing technology. The method first acquires and preprocesses historical and real-time water environment monitoring data for the river basin; then, a water quality prediction deep learning model is constructed, and after training and verification, the river basin water quality prediction results are output; then, a water quality indicator prediction mechanism model is constructed, and after training and verification, the prediction results of various water quality indicators in the river basin are output based on the water quality prediction results; finally, threshold judgments are performed based on the river basin water quality prediction results and the dynamic changes of various water quality indicators in the river basin, and expert emergency warning plans are retrieved based on the judgment results. The present invention can solve the problem of poor prediction results in the existing technology.

Description

Technical Field

[0001] The present invention belongs to the technical field of water quality prediction, and in particular relates to a river basin water quality prediction and early warning method, equipment and medium based on a multi-model processing technology. Background Art

[0002] With the development of society, surface river prediction and early warning in urban development areas play an important role in the water quality management of river basins. The important factors affecting the water quality prediction of surface river basins include natural factors and human factors. For example, natural factors include meteorological data and geographical data, while human factors include industrial pollution emissions and livestock and poultry breeding pollution emissions. Therefore, the water quality of river basins involves multiple aspects, and reliable water quality prediction methods are particularly important for predicting water quality change trends.

[0003] The water quality prediction method in the existing technology is relatively simple. It directly predicts by obtaining river data and importing it into the prediction model. The prediction results of this type of model are difficult to effectively predict the river water quality under the influence of multiple factors. Summary of the Invention

[0004] The technical problem solved by the present invention is to provide a river basin water quality prediction and early warning method, equipment and medium based on multi-model processing technology to solve the problem of poor prediction effect in the existing technology.

[0005] The basic solution provided by the present invention is a river basin water quality prediction and early warning method based on multi-model processing technology, including:

[0006] S1: Acquire water environment monitoring data from surface water quality monitoring stations in a river basin and perform preprocessing; the water environment monitoring data includes historical water environment monitoring data and real-time water environment monitoring data;

[0007] S2: Build a deep learning model for water quality prediction based on a deep learning network, use pre-processed historical water environment monitoring data for water quality prediction analysis training, and use pre-processed real-time water environment monitoring data to output river basin water quality prediction results;

[0008] S3: Construct a water quality index prediction mechanism model, train the water quality index prediction mechanism model based on pre-processed historical water environment monitoring data, verify the water quality index prediction mechanism model based on pre-processed real-time water environment monitoring data, input the river basin water quality prediction results into the water quality index prediction mechanism model, and output the prediction results of various water quality indicators in the river basin;

[0009] S4: Make threshold judgments based on the water quality prediction results of the river basin and the dynamic changes of various water quality indicators in the river basin, and call up expert emergency warning plans based on the judgment results.

[0010] Furthermore, the S1 includes:

[0011] S1-1: Divide the target river basin into a grid, and obtain water environment monitoring data from monitoring stations in the target river basin; the water environment monitoring data includes historical water environment monitoring data and real-time water environment monitoring data; both the historical water environment monitoring data and the real-time water environment monitoring data include monitoring station monitoring data and meteorological data;

[0012] S1-2: Preprocessing the water environment monitoring data, including missing value processing, data cleaning, normalization, filling of null value processing, spatialization processing and time series matching processing.

[0013] Furthermore, the S2 includes:

[0014] S2-1: Build a deep learning model for water quality prediction based on deep learning network;

[0015] S2-2: Preset a water quality prediction task, build a sliding window based on the monitoring station values ​​of the preset water quality prediction task, and fill the monitoring data and meteorological data of the monitoring station of the acquired historical water environment monitoring data into the sliding window;

[0016] S2-3: Generate data samples for the sliding window with a step size of 1, and divide them into training set, evaluation set and test set in chronological order with a preset ratio;

[0017] S2-4: Use the training set to train the model parameters of the water quality prediction deep learning model, use the evaluation set to optimize the hyperparameters of the water quality prediction deep learning model, use the test machine to verify the performance of the water quality prediction deep learning model, and output the trained water quality prediction deep learning model;

[0018] S2-5: Use the pre-processed real-time water environment monitoring data to input the trained water quality prediction deep learning model and output the river basin water quality prediction results;

[0019] S2-6: Encapsulate the deep learning model for water quality prediction.

[0020] Furthermore, the water quality prediction deep learning model is a fusion of a water quality prediction deep learning model based on a spatiotemporal graph convolutional network, a water quality prediction deep learning model based on a multi-layer attention mechanism network, and a water quality prediction deep learning model based on a long short-term memory network;

[0021] The water quality prediction deep learning model based on spatiotemporal graph convolutional network is used to construct an undirected graph of water quality monitoring network, and predict the water quality of different monitoring sections in the future region based on the temporal and spatial information in the water environment monitoring data;

[0022] The water quality prediction deep learning model based on the multi-layer attention mechanism network is based on the Encoder-Decoder structure and introduces a multi-layer attention mechanism to predict water quality parameters in water environment monitoring data;

[0023] The water quality prediction deep learning model based on the long short-term memory network is based on the LSTM water quality prediction AI model architecture to realize water quality parameter prediction of water environment monitoring data.

[0024] Furthermore, the S3 includes:

[0025] S3-1: Construct spatial database and attribute database based on the acquired water environment monitoring data, and perform ArcGIS data processing;

[0026] S3-2: Construct a water quality index prediction mechanism model that integrates the water quality model and the hydrodynamic model, and conduct training and verification based on data processed by ArcGIS;

[0027] S3-3: Analyze the spatiotemporal distribution characteristics of water quality in the river basin water quality prediction results based on the integrated water quality index prediction mechanism model after training, and output the prediction results of various water quality indicators in the river basin;

[0028] S3-4: Encapsulate and invert the water quality prediction mechanism model to generate a dynamic water quality rendering map.

[0029] Further, the S4 includes:

[0030] S4-1: Regularly read the online water environment monitoring data from the river basin monitoring stations to build a water quality evaluation optimization model; the water quality evaluation optimization model has preset warning level and warning limit data;

[0031] S4-2: Input the most recent water environment monitoring data into the water quality evaluation optimization model and output the current water quality evaluation data;

[0032] S4-3: Input the read water environment monitoring data into the encapsulated water quality prediction deep learning model and water quality prediction mechanism model, output the water quality deep learning prediction results and water quality mechanism prediction results respectively, and input them into the water quality evaluation optimization model, outputting the water quality deep learning prediction evaluation data and water quality mechanism prediction evaluation data;

[0033] S4-4: Compare the water quality deep learning prediction and evaluation data and the water quality mechanism prediction and evaluation data with the current water quality evaluation data, and use the model whose evaluation data error meets the preset standard in the comparison results as the water quality prediction model for the current preset period;

[0034] S4-5: Compare the water quality prediction results of the selected water quality prediction model with the preset warning level and warning limit data to generate a water quality warning result, perform warning classification operations based on the water quality warning result, and output the warning classification result;

[0035] S4-6: Extract the corresponding expert emergency plan based on the warning classification results and transmit it to the corresponding river basin management office.

[0036] The river basin water quality prediction and early warning device based on the multi-model processing technology is applied to the above-mentioned river basin water quality prediction and early warning method based on the multi-model processing technology, including:

[0037] Data acquisition module: used to obtain water environment monitoring data from surface water quality monitoring stations in river basins, the water environment monitoring data including historical water environment monitoring data and real-time water environment monitoring data;

[0038] Data preprocessing module: used to preprocess the acquired water environment monitoring data;

[0039] Water quality prediction module: Builds a deep learning model for water quality prediction based on a deep learning network, uses pre-processed historical water environment monitoring data for water quality prediction analysis training, and uses pre-processed real-time water environment monitoring data to output river basin water quality prediction results;

[0040] Water quality index prediction module: Build a water quality index prediction mechanism model, train the water quality index prediction mechanism model based on pre-processed historical water environment monitoring data, verify the water quality index prediction mechanism model based on pre-processed real-time water environment monitoring data, input the river basin water quality prediction results into the water quality index prediction mechanism model, and output the prediction results of various water quality indicators in the river basin;

[0041] Early warning module: Make threshold judgments based on the water quality prediction results of the river basin and the dynamic changes of various water quality indicators in the river basin, and call up expert emergency early warning plans based on the judgment results.

[0042] An electronic device includes a processor and a memory, wherein the memory stores programs or instructions, and the processor executes the river basin water quality prediction and early warning method based on multi-model processing technology as described above by calling the programs or instructions stored in the memory.

[0043] A computer-readable storage medium stores a program or instruction, which enables a computer to execute the river basin water quality prediction and early warning method based on the multi-model processing technology as described above.

[0044] The principles and advantages of the present invention are: in the technical solution of the present application, the use of a single prediction model in the prior art to predict the water quality of a river basin has the problem of poor prediction effect. The present application adopts a multi-model fusion technology. First, after processing the acquired water environment monitoring data, a deep learning network is used to construct a water quality prediction deep learning model. The model is based on the automatic feature learning of the deep learning network and can well predict the water quality of the river basin. On this basis, a water quality index prediction mechanism model is constructed by integrating the basin distributed water environment model, the hydrological water quality model and the hydrodynamic model. It can not only identify the factors affecting the water quality of the river basin, but also identify the distribution of relevant characteristics in the water quality, such as the key source area of ​​pollution.

[0045] Therefore, the solution of the present application effectively solves the problem that the water quality prediction model in the prior art has poor prediction results for water quality in complex river basins. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flowchart of an embodiment of the present invention;

[0047] Figure 2 This is an architecture diagram of a deep learning model for water quality prediction based on a spatiotemporal graph convolutional network according to an embodiment of the present invention;

[0048] Figure 3 is an undirected graph in an embodiment of the present invention;

[0049] Figure 4 Schematic diagram of the spatiotemporal dependence of different monitoring sections in monitoring data according to an embodiment of the present invention;

[0050] Figure 5 This is a structural diagram of a deep learning model for water quality prediction based on a multi-layer attention mechanism network in an embodiment of the present invention;

[0051] Figure 6 This is a structural diagram of a deep learning model for water quality prediction based on a long short-term memory network in an embodiment of the present invention;

[0052] Figure 7 Schematic diagram of the prediction results of ammonia nitrogen monitoring values ​​of each model in the embodiment of the present invention;

[0053] Figure 8 Schematic diagram of the predicted results of total phosphorus monitoring values ​​of each model in the embodiment of the present invention;

[0054] Figure 9 Schematic diagram of the prediction results of dissolved oxygen monitoring values ​​of various models in the embodiment of the present invention;

[0055] Figure 10 A comparison chart of the calculated and measured COD values ​​of the Lianghekou (R0) section shown in an example of the present invention;

[0056] Figure 11 A comparison diagram of the COD calculated value and the measured value of the new bridge (R2) section in an embodiment of the present invention;

[0057] Figure 12 1. A comparison chart of the COD calculated value and the measured value of the Liaojia Oil Mill (R12) section, as exemplified in an embodiment of the present invention;

[0058] Figure 13 Schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The following is further described in detail through specific implementation methods:

[0060] The symbols in the drawings of the specification include: electronic device 400 , processor 401 , memory 402 , input device 403 , and output device 404 .

[0061] The embodiment is basically as shown in the attached Figure 1 Shown: River basin water quality prediction and early warning method based on multi-model processing technology, including:

[0062] S1: Acquire water environment monitoring data from surface water quality monitoring stations in a river basin and preprocess the data; the water environment monitoring data includes historical water environment monitoring data and real-time water environment monitoring data; wherein S1 includes:

[0063] S1-1: Divide the target river basin into a grid and obtain water environment monitoring data from monitoring stations in the target river basin;

[0064] S1-2: Preprocess water environment monitoring data.

[0065] In this embodiment, the water environment monitoring data includes monitoring data and meteorological data obtained from monitoring sites, and the monitoring data obtained from the monitoring sites include geographical data, hydrological data, pollution source data, river topography data and socio-economic data, etc. In meteorological data, meteorological factors have a considerable impact on the generation and spread of water pollution. Therefore, the water environment monitoring data in this application takes meteorological data into account, and after subsequent analysis, the complex impact relationship of meteorological factors on the water quality of the river basin can be obtained; in this embodiment, the meteorological factors in the meteorological data, that is, meteorological indicators, include temperature, air pressure, humidity and rainfall, etc., and this application is for meteorological The monitoring data obtained from monitoring indicators and monitoring stations have a valid value range set, that is, meteorological data and monitoring data that are not within the valid value range are regarded as outliers and are regarded as missing values ​​in the subsequent processing flow, and need to be preprocessed. Therefore, the preprocessing of this application includes missing value processing and normalization. Specifically, many problems that arise in actual data collection will lead to missing and abnormal data. Among them, abnormal data can be regarded as missing values ​​after being judged in the valid range. Linear interpolation is used to fill missing values ​​during implementation. The linear interpolation method assumes that a continuous sequence containing missing values ​​satisfies a linear change relationship, and a linear function is used to fit the data changes.

[0066] In other embodiments of this embodiment, the preprocessing measures also include data cleaning, filling null value processing, spatial processing and time series matching processing, so as to ensure the temporal, spatial and logical consistency of the water environment monitoring data obtained from different sources.

[0067] S2: Build a deep learning model for water quality prediction based on a deep learning network, use pre-processed historical water environment monitoring data for water quality prediction analysis training, and use pre-processed real-time water environment monitoring data to output river basin water quality prediction results; S2 includes:

[0068] S2-1: Build a deep learning model for water quality prediction based on deep learning network;

[0069] S2-2: Preset a water quality prediction task, build a sliding window based on the monitoring station values ​​of the preset water quality prediction task, and fill the monitoring data and meteorological data of the monitoring station of the acquired historical water environment monitoring data into the sliding window;

[0070] S2-3: Generate data samples for the sliding window with a step size of 1, and divide them into training set, evaluation set and test set in chronological order with a preset ratio;

[0071] S2-4: Use the training set to train the model parameters of the water quality prediction deep learning model, use the evaluation set to optimize the hyperparameters of the water quality prediction deep learning model, use the test machine to verify the performance of the water quality prediction deep learning model, and output the trained water quality prediction deep learning model;

[0072] S2-5: Use the pre-processed real-time water environment monitoring data to input the trained water quality prediction deep learning model and output the river basin water quality prediction results;

[0073] S2-6: Encapsulate the deep learning model for water quality prediction.

[0074] In this embodiment, the water quality prediction deep learning model is constructed using a deep learning network. It is based on the comparative analysis of existing model principles and is obtained through in-depth consideration. Specifically, this application considers a large number of models, such as mechanism-based models and machine learning-based methods. Among them, the mechanism-based model method is based on professional models in the field, simulating the entire process from discharge to diffusion of water pollutants, thereby predicting the future water quality of each monitoring station. This method has very high requirements for the integrity of pollution source data, and the model parameters need to be adjusted and adapted by professionals according to different water scenarios, and the generalization ability is poor.

[0075] The machine learning-based method is a data-driven method that predicts various indicators of monitoring sites in the future by mining the implicit complex dependencies of historical data. The machine learning-based method can automatically learn the complex mapping relationship between multi-source data input and actual results. This type of method can be further divided into time series analysis models, general regression models and deep learning models. The time series analysis model predicts future trends by discovering linear patterns in historical sequences, but cannot accept feature inputs other than sequence data. The general regression model can support multi-source data input, such as linear regression models, support vector machines, random forests, etc. The disadvantage of this type of method is that in order to achieve considerable prediction accuracy, complex feature engineering is required, which increases the difficulty of model implementation.

[0076] The deep learning model fits the nonlinear complex mapping from input to output by stacking multiple layers of neural networks, which can realize automated feature learning and solve the feature selection problem of traditional regression models.

[0077] The water quality prediction deep learning model based on deep learning network constructed in this application is formed by the fusion of a water quality prediction deep learning model based on spatiotemporal graph convolutional network, a water quality prediction deep learning model based on a multi-layer attention mechanism network, and a water quality prediction deep learning model based on a long short-term memory network. Among them, the fusion of the water quality prediction model based on spatiotemporal graph convolutional network and the water quality prediction deep learning model based on a multi-layer attention mechanism network can obtain the complex spatial dependencies of the monitoring data of the monitoring stations, while the fusion of the water quality prediction model based on spatiotemporal graph convolutional network and the water quality prediction deep learning model based on the long short-term memory network can characterize the complex temporal dependencies of the monitoring data.

[0078] Regarding water quality prediction models based on spatiotemporal graph convolutional networks, although traditional deep learning methods have achieved great success in extracting features from Euclidean spatial data, water environment monitoring data is generated from non-Euclidean space, and traditional deep learning methods are not satisfactory in processing non-Euclidean spatial data. In this context, in view of the extremely high complexity and nonlinear characteristics of water quality data, this application takes into account the spatial and temporal dependencies between water quality monitoring sections and proposes a water quality prediction model based on spatiotemporal graph convolutional networks to solve the water quality prediction problem. Specifically:

[0079] The temporal and spatial information contained in the data is used to predict water quality indicators for different monitoring sections in the future. The water quality monitoring network is defined as an undirected graph, and the spatiotemporal dependencies of water quality monitoring sections are modeled using spatiotemporal convolution blocks, which can effectively capture spatiotemporal correlations.

[0080] In the spatial dimension, the graph convolution formula after Chebyshev approximation and first-order approximation is used to capture spatial correlation. In the convolution process, not only the state of neighboring nodes but also the state of the node itself is considered. In the time dimension, gated convolution is used to capture time dependency. Different from the traditional convolution method, the model considers the problem of time series and uses causal convolution to perform parallel calculations on the data, which makes the model training faster. The above graph convolution and gated convolution are combined into Figure 2 The structure shown uses a bottleneck strategy to achieve scale compression and feature compression, and a normalization layer is connected after each layer to prevent overfitting. The final model is a stack of two St-Conv Blocks followed by an output layer, where the output layer first uses a time dimension convolution to merge the time dimension of the previous output data. After merging, it outputs the final predicted data through a convolution. The predicted data is a picture of the next time dimension.

[0081] Undirected graphs such as Figure 3 As shown in the figure, the undirected graph G = (V, E, A), each point on G detects water quality at the same frequency F, and the water quality prediction problem is defined as: given all historical variable values ​​of all nodes on the spatiotemporal graph G in the past period of time, predict the water quality parameters of all nodes in the future period of time.

[0082] The water quality prediction deep learning model based on the deep learning network also adopts a water quality prediction deep learning model based on a multi-layer attention mechanism network, specifically:

[0083] The deep learning model for water quality prediction based on a multi-layer attention network is based on an encoder-decoder structure. It introduces a multi-layer attention mechanism to the spatiotemporal prediction of water quality parameters. This model models the dynamic spatiotemporal correlations between sensors and improves model performance by integrating external factors such as rainfall data corresponding to the sensors in the decoder stage. The challenges of spatiotemporal prediction of water quality parameters are mainly reflected in the following two aspects: (1) dynamic spatiotemporal correlations; and (2) external factors.

[0084] The dynamic spatiotemporal correlation is reflected in the following two aspects.

[0085] 1. Complex spatial correlations between different sensor devices, such as Figure 4 As shown, the spatial correlation between sensors is highly dynamic over time, and the geographical locations are nonlinearly related;

[0086] 2. Dynamic correlation within the same sensor. Water quality time series data often follows a periodic pattern, but sometimes (such as sudden extreme weather) sensor data will fluctuate significantly, so how to choose the appropriate time interval is also a challenge.

[0087] As for external factors, the sensor data will be affected by the surrounding environment and the time (whether it is the peak emission period or not).

[0088] Therefore, the water quality prediction deep learning model based on the multi-layer attention mechanism network in this embodiment consists of the following two main parts. The model structure is as follows: Figure 5 As shown:

[0089] 1. Multi-layer Attention Mechanism: The multi-layer attention mechanism mainly consists of an encoder with two spatial attention mechanisms (local spatial attention and global spatial attention) and a decoder with temporal attention. In the encoder, two different spatial attention mechanisms are proposed, which use the implicit state of the encoder at the previous moment, the time series values ​​of sensor observations at the previous moment, and spatial context (such as sensor network information) to capture the complex spatial relationships between sensors. In the decoder, a temporal attention mechanism is used to adaptively select relevant time slices in the past for prediction.

[0090] Second, external factor integration. This module is designed in this embodiment to address the impact of external factors on geographic sensor time series, including meteorological data, time-related features, and geographic features. Specifically, this module integrates these external factors and then uses its output as part of the input to the decoder mentioned above and enters the network's calculations.

[0091] The water quality prediction deep learning model based on the long short-term memory network is specifically a water quality prediction AI model architecture based on the long short-term memory network model (LSTM). The long short-term memory network model (LSTM) is a deep learning model used to process sequence data with time dependencies, such as natural language. It is a network composed of multiple memory units, each of which has three gates (input gate, output gate, and forget gate) that can control the information stored and transmitted in the memory unit. The LSTM network uses these gates to balance short-term memory and long-term memory, thereby retaining long-term dependencies in sequence data while still effectively processing complex sequence data.

[0092] Each time step of the LSTM network contains multiple gates, including input gate, output gate and forget gate. In the input gate, the LSTM network selects important information from the input sequence and outputs it to the next time step. In the output gate, the LSTM network will select the information to be passed to the next time step and output it to the output part of the sequence. In the forget gate, the LSTM network controls which information is stored in the memory unit and which information should be forgotten. The LSTM network stores and operates sequence data through memory units. The memory unit is controlled by three gates. It can store important information in the input sequence in its local state and pass it to the next time step when needed. In each time step, the memory unit will update its state to better reflect the important information in the input sequence.

[0093] Model structure and main parameters such as Figure 6 As shown in the figure, the first layer is an LSTM training model with 300 neurons. According to the above principle, the model learns based on past fitting results and uses the memory characteristics of LSTM to optimize water quality prediction capabilities. The purpose of the second layer is to prevent overfitting and allow all neurons to learn better. It is then connected to the third layer to fully connect the neurons in the previous layer, and finally output after the activation function in the fourth layer.

[0094] The training process of the fusion model in this application is as follows: first, water environment monitoring data is obtained and preprocessed, and then the model hyperparameters are set. For example, the batch size is set to 64, the output dimension of the GCN is 16, the hidden state length of the LSTM is 32, and the number of GCN layers is 1. The model parameters are optimized using the Adam optimizer, with a learning rate of 0.002 and a training epoch of 150. To reduce randomness, the model is trained independently for 5 times, and the average of the 5 experimental results is taken as the final result in the experimental part.

[0095] A water quality prediction task is pre-defined. For example, the prediction task is defined as using 18 values ​​(Tin = 18, sampling period is 4 hours) of historical water environment monitoring data and meteorological data from all stations to predict water quality for the next 18 values ​​(Tout = 18, sampling period is 4 hours). A sample containing 18 water quality parameters and meteorological data is used as the model input, and the subsequent 18 water quality indicators are used as the actual surface water quality value. A sliding window with a window size of 36 and a step size of 1 is used to generate data samples. The data samples are divided into training set, evaluation set, and test set in chronological order with a ratio of 7:1:2;

[0096] The model parameters are optimized using samples from the training set, and the hyperparameters are optimized using the results from the evaluation set. The performance of the model is verified on the test set. The performance evaluation indicators include mean absolute error (MAE) and mean relative error (MRE), which are calculated as follows:

[0097]

[0098] Among them, Y i and They represent the true value and predicted value of the i-th sample respectively, N represents the number of samples, MAE directly calculates the absolute value between the true value and the predicted value, and MRE divides the true value by MAE to obtain the relative error rate. The result is more intuitive, but it is not suitable for data indicators with a numerical range near 0 (the relative error will be very large).

[0099] To verify the performance advantage of the fusion model in this embodiment, the fusion model of this application is compared with the existing model based on machine learning methods. The baseline model used for comparison is as follows:

[0100] 1. Support Vector Regression (SVR). This method is a regression version of the traditional support vector machine (SVM). Its goal is to find the optimal hyperplane in the feature space such that all data samples in the feature space are closest to this hyperplane. The original version of SVR only supports single-variable output. As an experimental comparison baseline, this paper establishes an SVR model for each indicator and prediction step size.

[0101] 2. Autoregressive Integrated Moving Average (ARIMA) model: This method transforms the time series into a stationary time series and regresses the dependent variable on its lagged values ​​and random error terms. It is one of the most common models in time series linear analysis.

[0102] 3. Long Short-Term Memory (LSTM) networks. LSTMs maintain memory during sequence modeling and have a more complex gating structure than RNNs and GRUs. They use a forget gate and input gate to control the retention and updating of long-term memory, and an output gate to obtain the output at each moment. LSTMs alleviate the vanishing gradient problem in RNN models and are the most commonly used neural network for modeling time series in deep learning.

[0103] In order to evaluate the performance of each model in short-term and long-term prediction tasks, the errors of the models at different prediction steps are listed; Figure 7 、 Figure 8 and Figure 9 The figure shows the prediction results of the proposed model and the comparison model on the three indicators of ammonia nitrogen, total phosphorus and dissolved oxygen mass concentration. In the figure, L is the prediction step size. It can be seen that compared with other baseline models, the fusion model of this application has obvious performance advantages.

[0104] Therefore, this application proposes a new deep learning model for water quality prediction, which uses graph neural networks and multi-layer attention mechanisms to model the complex spatial dependencies of surface water quality monitoring stations. Long short-term memory networks are used to model the complex temporal dependencies of historical indicator sequences, and to achieve multi-step prediction output of water quality indicators for all surface water quality stations. Experimental results show that compared with traditional regression methods, the use of this model can significantly improve prediction performance, and compared with time series analysis methods, general regression methods and general deep learning methods, this model can achieve performance improvement.

[0105] S3: Construct a water quality index prediction mechanism model, train the water quality index prediction mechanism model based on pre-processed historical water environment monitoring data, verify the water quality index prediction mechanism model based on pre-processed real-time water environment monitoring data, input the river basin water quality prediction results into the water quality index prediction mechanism model, and output the prediction results of various water quality indicators in the river basin; S3 includes:

[0106] S3-1: Construct spatial database and attribute database based on the acquired water environment monitoring data, and perform ArcGIS data processing;

[0107] S3-2: Construct a water quality index prediction mechanism model that integrates the water quality model and the hydrodynamic model, and conduct training and verification based on data processed by ArcGIS;

[0108] S3-3: Analyze the spatiotemporal distribution characteristics of water quality in the river basin water quality prediction results based on the integrated water quality index prediction mechanism model after training, and output the prediction results of various water quality indicators in the river basin;

[0109] S3-4: Encapsulate and invert the water quality prediction mechanism model to generate a dynamic water quality rendering map.

[0110] In this embodiment, for the acquired water environment monitoring data, in addition to the above-mentioned missing value processing and normalization processing, a spatial database and an attribute database are also constructed based on the acquired water environment monitoring data. The spatial database includes digital elevation maps, land use maps, soil type maps, watershed system maps, administrative division maps, etc., and the attribute database includes water quality databases, meteorological databases, soil databases, socio-economic databases, etc.

[0111] The data stored in the above database is then processed by ArcGIS data. ArcGIS data processing includes information layer preparation, vector-to-raster conversion, spatial interpolation processing, water system map generation, sub-watershed generation, and spatial overlay processing. The data after ArcGIS data processing is transmitted to the water quality index prediction mechanism model that is a fusion of the constructed water quality model and the hydrodynamic model. The fusion water quality index prediction mechanism model in this embodiment is constructed specifically as follows:

[0112] Step 1: Construct the control equations of the water quality index prediction mechanism model, including the two-dimensional planar phreatic water control equation and the material transport equation, specifically:

[0113] The vector form of the two-dimensional diving control equation is:

[0114]

[0115]

[0116] Where t represents time; x and y represent horizontal coordinates; U is the vector of conserved variables; F and G are the vectors of flux variables; F d and G d is the diffusion flux vector; S is the source term vector, h is the water depth, u and v are the flow velocities in the x and y directions respectively, η is the water level, τ bx and τ by are the bottom friction terms in the x and y directions, v t is the diffusion coefficient of the transported species.

[0117] Material transport equation:

[0118]

[0119] C represents the material transport variable; v t is the diffusion coefficient of the transported substance, v t =αu * h,α takes values ​​between 0.3 and 1.0. S C is the source term of the transport variable.

[0120] Step 2: Discrete processing of water flow control equations, specifically:

[0121] The finite volume method is used to solve the mathematical model of water flow based on unstructured grids. The equation (1) of the two-dimensional diving control equation is integrated on any triangular control volume to obtain:

[0122]

[0123] Let U i is the average value of the cell, stored at the center of the cell,

[0124] Applying Green's formula to the mass transport equation, the area is decomposed into line integrals along its perimeter to obtain:

[0125]

[0126] Where L i is the perimeter of the i-th control unit V, n=(n x ,n y )=(cosφ,sinφ) is the outer normal unit vector on the perimeter, φ is the angle between the outer normal vector and the positive direction of the x-axis, A i is the area of ​​calculation unit i.

[0127] The line integral of the formula obtained by integrating the two-dimensional diving control equation on any triangular control volume is discretized and sorted out to obtain:

[0128]

[0129] Where: m is the number of control volume edges. Since this paper uses a triangular control volume, m = 3; l ij is the length of each side of the unit; Represents the normal numerical flux through the jth edge of the i-th unit.

[0130] Step 3: Discretization of the transport equation, specifically:

[0131] The material transport equation is also discretized in the triangular control volume using finite volume. The integral of equation (3) on element i is:

[0132]

[0133] Use Green's formula to decompose the area into a line integral along its perimeter, and discretize the line integral part and sort it out:

[0134]

[0135] Q ij is the flow through the j-th edge of unit i; is the gradient of the transport variable in the direction of the outer normal of the j-th edge of unit i, where:

[0136]

[0137] N1 and N2 are the two vertices of the j-th edge of unit i. ij is the transport variable value of the j-th edge of unit i, calculated by upwind reconstruction interpolation:

[0138] Assume that the adjacent cells of cell i on side j are m ij , and the water flows from unit i to unit m ij , then the transport variable C at edge j ij It is obtained by interpolating the transport variables of two adjacent cell centers:

[0139]

[0140] Φ (r) is a function of the gradient r of the transport variable. In order to reduce numerical dissipation and improve calculation accuracy, this paper uses the van Leer function to restrict the gradient function:

[0141] Φ (r) =max[0,(r+|r| / (1+r))]

[0142]

[0143] s i is the edge where the flow in unit i flows out of the unit; is the transport variable value at the center of the unit grid adjacent to edge k.

[0144] After the water quality prediction mechanism model is constructed, the initial conditions, boundary conditions, and model parameters of the model are set according to the actual river conditions. For example, in this embodiment, the Binan River is used as an example, where the initial conditions are: the initial water level and initial flow velocity of the model are both set to 0;

[0145] The boundary conditions are as follows: three boundaries are divided, including two flow boundaries (the two estuaries of Banzhuershe in the main channel of Meijiang River and the two estuaries of Binan River) and one water level boundary (the two estuaries of Binan River);

[0146] The model parameters are: The riverbed roughness range for the Binan River Basin is 0.016 to 0.035. Based on this, the dry season channel roughness range was determined to be between 0.016 and 0.044. Regarding the selection of comprehensive pollutant attenuation coefficients, the attenuation coefficients for COD, ammonia nitrogen, and total phosphorus in the Binan River Basin were determined to be 0.20 / day, 0.15 / day, and 0.1 / day, respectively.

[0147] After the model is set up and trained, the simulation effect of the model is verified using real-time water environment monitoring data, taking the COD water quality indicator as an example:

[0148] The comparison results of COD calculation and measured values ​​at Lianghekou (R0), Xindaqiao (R2), and Liaojiayoufang (R12) sections are as follows: Figure 10 、 Figure 11 and Figure 12 The comparison results show that the calculated COD values ​​of each section are consistent with the measured values, with relative errors of -6.18%, 8.71%, and 1.67%, respectively.

[0149] In addition, deep learning models and mechanism models are encapsulated, including model algorithm encapsulation and data encapsulation involved in the calculation, and model calculation services are released to realize online calling of model services and real-time storage of prediction results.

[0150] Inversion and rendering, on the other hand, are to display the model's prediction results on a digital map of the river basin in a visual interface to improve the intuitive feedback of the prediction results.

[0151] In summary, the water quality index prediction mechanism model that integrates the hydrodynamic and water quality models established in this application reproduces the river flow and material transport process well. The simulation results are consistent with the measured values, and the deviations are controlled within a reasonable range. It can be used for subsequent water environment simulation analysis.

[0152] S4: Based on the water quality prediction results of the river basin and the dynamic changes of various water quality indicators in the river basin, threshold judgment is made, and expert emergency warning plans are retrieved based on the judgment results. Among them, S4 includes:

[0153] S4-1: Regularly read the online water environment monitoring data from the river basin monitoring stations to build a water quality evaluation optimization model; the water quality evaluation optimization model has preset warning level and warning limit data;

[0154] S4-2: Input the most recent water environment monitoring data into the water quality evaluation optimization model and output the current water quality evaluation data;

[0155] S4-3: Input the read water environment monitoring data into the encapsulated water quality prediction deep learning model and water quality prediction mechanism model, output the water quality deep learning prediction results and water quality mechanism prediction results respectively, and input them into the water quality evaluation optimization model, outputting the water quality deep learning prediction evaluation data and water quality mechanism prediction evaluation data;

[0156] S4-4: Compare the water quality deep learning prediction and evaluation data and the water quality mechanism prediction and evaluation data with the current water quality evaluation data, and use the model whose evaluation data error meets the preset standard in the comparison results as the water quality prediction model for the current preset period;

[0157] S4-5: Compare the water quality prediction results of the selected water quality prediction model with the preset warning level and warning limit data to generate a water quality warning result, perform warning classification operations based on the water quality warning result, and output the warning classification result;

[0158] S4-6: Extract the corresponding expert emergency plan based on the warning classification results and transmit it to the corresponding river basin management office.

[0159] In this embodiment, for the constructed water quality prediction deep learning model and water quality index prediction mechanism model, the dimension of the water quality results predicted by the water quality prediction deep learning model belongs to the monitoring section site on the target river basin, while the water quality index prediction mechanism model is aimed at the water quality index results of several grids divided on each section. For example, 2000 grids are divided to obtain the water quality index results on each grid. Therefore, it is not difficult to see that the resources and computing power required for the operation of the water quality prediction deep learning model are less than the resources and computing power required for the operation of the water quality index prediction mechanism model. In different monitoring environments, according to the results desired by the user, it is necessary to intelligently call the corresponding model for processing. Therefore, through the water quality evaluation optimization model, the evaluation data comparison results of the two fusion models over a period of time are compared, and a preset error is set according to user needs. After comparison, the corresponding fusion model that is less than the preset error and closest to the preset error is used as the model called for subsequent monitoring for a period of time. In this way, the actual needs of the user are met and the resource pressure and computing power of the system are reduced.

[0160] In another embodiment of this embodiment, a river basin water quality prediction and early warning device based on a multi-model processing technology is further included, which is applied to the above-mentioned river basin water quality prediction and early warning method based on a multi-model processing technology, including:

[0161] Data acquisition module: used to obtain water environment monitoring data from surface water quality monitoring stations in river basins, the water environment monitoring data including historical water environment monitoring data and real-time water environment monitoring data;

[0162] Data preprocessing module: used to preprocess the acquired water environment monitoring data;

[0163] Water quality prediction module: Builds a deep learning model for water quality prediction based on a deep learning network, uses pre-processed historical water environment monitoring data for water quality prediction analysis training, and uses pre-processed real-time water environment monitoring data to output river basin water quality prediction results;

[0164] Water quality index prediction module: Build a water quality index prediction mechanism model, train the water quality index prediction mechanism model based on pre-processed historical water environment monitoring data, verify the water quality index prediction mechanism model based on pre-processed real-time water environment monitoring data, input the river basin water quality prediction results into the water quality index prediction mechanism model, and output the prediction results of various water quality indicators in the river basin;

[0165] Early warning module: Make threshold judgments based on the water quality prediction results of the river basin and the dynamic changes of various water quality indicators in the river basin, and call up expert emergency early warning plans based on the judgment results.

[0166] Also included is an electronic device such as Figure 13 As shown, the electronic device 400 includes one or more processors 401 and a memory 402 .

[0167] The processor 401 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 400 to perform desired functions.

[0168] The memory 402 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 401 may run the program instructions to implement the river basin water quality prediction and early warning method based on multi-model processing technology of any embodiment of the present invention described above and / or other desired functions. Various contents such as initial external parameters, thresholds, etc. may also be stored in the computer-readable storage medium.

[0169] In one example, electronic device 400 may further include an input device 403 and an output device 404, which are interconnected via a bus system and / or other connection mechanisms (not shown). Input device 403 may include, for example, a keyboard, a mouse, etc. Output device 404 may output various information to the outside, including warning information, braking force, etc. Output device 404 may include, for example, a display, a speaker, a printer, a communication network, and remote output devices connected thereto.

[0170] Of course, to simplify, Figure 13Only some of the components related to the present invention in the electronic device 400 are shown, and components such as buses, input / output interfaces, etc. are omitted. In addition, the electronic device 400 may further include any other appropriate components according to specific application scenarios.

[0171] In addition to the above-mentioned methods and devices, an embodiment of the present invention may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the river basin water quality prediction and early warning method based on multi-model processing technology provided in any embodiment of the present invention.

[0172] The computer program product may be written in any combination of one or more programming languages ​​to implement the operations of embodiments of the present invention, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as C or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0173] In addition, an embodiment of the present invention may also be a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the processor executes the steps of the river basin water quality prediction and early warning method based on multi-model processing technology provided in any embodiment of the present invention.

[0174] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0175] The above are only embodiments of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme are not described in detail here. Ordinary technicians in the field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A river basin water quality prediction and early warning method based on multi-model processing technology is characterized by: include: S1: Acquire water environment monitoring data from surface water quality monitoring stations in a river basin and perform preprocessing; the water environment monitoring data includes historical water environment monitoring data and real-time water environment monitoring data; S2: Build a deep learning model for water quality prediction based on a deep learning network, use pre-processed historical water environment monitoring data for water quality prediction analysis training, and use pre-processed real-time water environment monitoring data to output river basin water quality prediction results; S3: Construct a water quality index prediction mechanism model, train the water quality index prediction mechanism model based on pre-processed historical water environment monitoring data, verify the water quality index prediction mechanism model based on pre-processed real-time water environment monitoring data, input the river basin water quality prediction results into the water quality index prediction mechanism model, and output the prediction results of various water quality indicators in the river basin; S4: Based on the water quality prediction results of the river basin and the dynamic changes of various water quality indicators in the river basin, threshold judgment is made, and expert emergency warning plans are called according to the judgment results; The S2 includes: S2-1: Build a deep learning model for water quality prediction based on deep learning network; S2-2: Preset a water quality prediction task, build a sliding window based on the monitoring station values ​​of the preset water quality prediction task, and fill the monitoring data and meteorological data of the monitoring station of the acquired historical water environment monitoring data into the sliding window; S2-3: Generate data samples for the sliding window with a step size of 1, and divide them into training set, evaluation set and test set in chronological order with a preset ratio; S2-4: Use the training set to train the model parameters of the water quality prediction deep learning model, use the evaluation set to optimize the hyperparameters of the water quality prediction deep learning model, use the test machine to verify the performance of the water quality prediction deep learning model, and output the trained water quality prediction deep learning model; S2-5: Use the pre-processed real-time water environment monitoring data to input the trained water quality prediction deep learning model and output the river basin water quality prediction results; S2-6: Encapsulate the deep learning model for water quality prediction; The water quality prediction deep learning model is a fusion of a water quality prediction deep learning model based on a spatiotemporal graph convolutional network, a water quality prediction deep learning model based on a multi-layer attention mechanism network, and a water quality prediction deep learning model based on a long short-term memory network; The water quality prediction deep learning model based on spatiotemporal graph convolutional network is used to construct an undirected graph of water quality monitoring network, and predict the water quality of different monitoring sections in the future region based on the temporal and spatial information in the water environment monitoring data; The water quality prediction deep learning model based on the multi-layer attention mechanism network is based on the Encoder-Decoder structure and introduces a multi-layer attention mechanism to predict water quality parameters in water environment monitoring data; The water quality prediction deep learning model based on the long short-term memory network is based on the LSTM water quality prediction AI model architecture to achieve water quality parameter prediction based on water environment monitoring data; The S3 includes: S3-1: Construct spatial database and attribute database based on the acquired water environment monitoring data, and perform ArcGIS data processing; S3-2: Construct a water quality index prediction mechanism model that integrates the water quality model and the hydrodynamic model, and conduct training and verification based on data processed by ArcGIS; The construction of the integrated water quality index prediction mechanism model includes: Step 1: Construct the control equations of the water quality index prediction mechanism model, including the two-dimensional diving control equation and the material transport equation: The vector form of the planar two-dimensional diving control equation is: Where, Represents time; 、 Represents the horizontal coordinate; is the vector of conserved variables; and is the vector of flux variables; and is the diffusion flux vector; is the vector of source terms, For water depth, 、 They are and Flow velocity in the direction, is the water level, and They are and The bottom friction term in the direction, is the diffusion coefficient of the transported substance; The mass transport equation is: represents the material transport variable; is the diffusion coefficient of the transported species, , The value is between 0.3 and 1.

0. ; is the source term of the transport variable; Step 2: Discrete processing of water flow control equations: The finite volume method is used to solve the mathematical model of water flow based on unstructured grids. The formula of the two-dimensional diving control equation is integrated on any triangular control volume to obtain: set up is the average value of the cell, stored at the center of the cell, ; Applying Green's formula to the mass transport equation, the area is decomposed into line integrals along its perimeter to obtain: In the formula For the control unit The perimeter, is the outward normal unit vector on the perimeter, is the external normal vector and The angle in the positive direction of the axis, For computing units area; The line integral of the formula obtained by integrating the two-dimensional diving control equation on any triangular control volume is discretized and sorted out to obtain: Where: is the number of edges of the control body. Since a triangle control body is used, ; is the length of each side of the unit; Indicates that through Unit No. Normal numerical flux of the edge; Step 3: Discretization of transport equations: The material transport equation is also discretized in the triangular control volume. The above equation Points are: Use Green's formula to decompose the area into a line integral along its perimeter, and discretize the line integral part and sort it out: For passing units No. The flow of the edge; For unit No. The gradient of the transport variable in the direction of the normal to the edge, where: and For unit No. The two vertices of an edge; For unit No. The transport variable values ​​of the edges are calculated by upwind reconstruction interpolation: Hypothesis Unit side The adjacent units are , and the water flow is from the unit Flow unit , then the edge Transport variables It is obtained by interpolating the transport variables of two adjacent cell centers: is the gradient of the transport variable In order to reduce numerical dissipation and improve calculation accuracy, the van Leer function is used to restrict the gradient function: For unit The flow rate flows out of the edge of the unit; For the edge The transport variable values ​​at the centers of adjacent cell grids; S3-3: Analyze the spatiotemporal distribution characteristics of water quality in the river basin water quality prediction results based on the integrated water quality index prediction mechanism model after training, and output the prediction results of various water quality indicators in the river basin; S3-4: Encapsulate and invert the water quality prediction mechanism model to generate a dynamic water quality rendering map.

2. The method for predicting and warning river basin water quality based on multi-model processing technology according to claim 1 is characterized by: Said S1 comprises: S1-1: Divide the target river basin into a grid, and obtain water environment monitoring data from monitoring stations in the target river basin; the water environment monitoring data includes historical water environment monitoring data and real-time water environment monitoring data; both the historical water environment monitoring data and the real-time water environment monitoring data include monitoring station monitoring data and meteorological data; S1-2: Preprocessing the water environment monitoring data, including missing value processing, data cleaning, normalization, filling of null value processing, spatialization processing and time series matching processing.

3. The method for predicting and warning river water quality based on multi-model processing technology according to claim 2 is characterized by: The S4 includes: S4-1: Regularly read the online water environment monitoring data from the river basin monitoring stations to build a water quality evaluation optimization model; the water quality evaluation optimization model has preset warning level and warning limit data; S4-2: Input the most recent water environment monitoring data into the water quality evaluation optimization model and output the current water quality evaluation data; S4-3: Input the read water environment monitoring data into the encapsulated water quality prediction deep learning model and water quality prediction mechanism model, output the water quality deep learning prediction results and water quality mechanism prediction results respectively, and input them into the water quality evaluation optimization model, outputting the water quality deep learning prediction evaluation data and water quality mechanism prediction evaluation data; S4-4: Compare the water quality deep learning prediction and evaluation data and the water quality mechanism prediction and evaluation data with the current water quality evaluation data, and use the model whose evaluation data error meets the preset standard in the comparison results as the water quality prediction model for the current preset period; S4-5: Compare the water quality prediction results of the selected water quality prediction model with the preset warning level and warning limit data to generate a water quality warning result, perform warning classification operations based on the water quality warning result, and output the warning classification result; S4-6: Extract the corresponding expert emergency plan based on the warning classification results and transmit it to the corresponding river basin management office.

4. A river basin water quality prediction and early warning device based on multi-model processing technology, applied to the river basin water quality prediction and early warning method based on multi-model processing technology as described in any one of claims 1 to 3 above, characterized in that: include: Data acquisition module: used to obtain water environment monitoring data from surface water quality monitoring stations in river basins, the water environment monitoring data including historical water environment monitoring data and real-time water environment monitoring data; Data preprocessing module: used to preprocess the acquired water environment monitoring data; Water quality prediction module: Builds a deep learning model for water quality prediction based on a deep learning network, uses pre-processed historical water environment monitoring data for water quality prediction analysis training, and uses pre-processed real-time water environment monitoring data to output river basin water quality prediction results; Water quality index prediction module: Build a water quality index prediction mechanism model, train the water quality index prediction mechanism model based on pre-processed historical water environment monitoring data, verify the water quality index prediction mechanism model based on pre-processed real-time water environment monitoring data, input the river basin water quality prediction results into the water quality index prediction mechanism model, and output the prediction results of various water quality indicators in the river basin; Early warning module: Make threshold judgments based on the water quality prediction results of the river basin and the dynamic changes of various water quality indicators in the river basin, and call up expert emergency early warning plans based on the judgment results.

5. An electronic device, characterized in that: It includes a processor and a memory, wherein the memory stores programs or instructions, and the processor executes the river basin water quality prediction and early warning method based on multi-model processing technology as described in any one of claims 1 to 3 by calling the programs or instructions stored in the memory.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or instruction, and the program or instruction enables a computer to execute the river basin water quality prediction and early warning method based on multi-model processing technology as described in any one of claims 1 to 3 above.

Images