Water environment monitoring and evaluation method and system based on artificial intelligence

Abstract

The invention relates to the technical field of water environment monitoring, and particularly discloses a water environment monitoring evaluation method and system based on artificial intelligence, and the method comprises the steps: integrating a multi-source data collection interface, carrying out the denoising and missing value filling processing of original data, and storing the processed data in a database; setting a dynamic weight adjustment parameter; training and optimizing the model by using historical data, and storing the trained model in a model library; performing fusion processing on the multi-source data; mining potential association among the data, and storing association rules in a knowledge base; acquiring preprocessed data from the database, and inputting the preprocessed data into the trained model for evaluation; and carrying out visualization processing on the evaluation result, carrying out early warning judgment according to an early warning threshold value, and when an early warning condition is triggered, sending out early warning information through a plurality of communication modes. According to the invention, the problems of one-sided and inaccurate traditional monitoring data are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of water environment monitoring, and in particular to a water environment monitoring and evaluation method and system based on artificial intelligence. Background Art

[0002] Traditional monitoring methods have numerous shortcomings when faced with complex and changing water environments. Manual sampling and laboratory analysis not only consume significant manpower, material resources, and time, but also have limited sampling locations, failing to fully reflect the true state of the water environment and easily overlooking localized pollution issues or short-term sudden changes in water quality.

[0003] With technological advancements, multi-source data collection methods such as satellite remote sensing, ground-based sensor networks, and drone monitoring have gradually gained popularity. Satellite remote sensing can capture macroscopic information about large water areas, such as water color and temperature distribution, enabling the timely detection of large-scale water anomalies. Ground-based sensors can monitor water quality parameters such as pH, dissolved oxygen (DO), and chemical oxygen demand (COD) at specific locations in real time, with high accuracy. Drones, with their flexibility, can provide detailed observations of specific areas, acquiring high-resolution imagery and localized water quality data. However, these multi-source data sources present significant challenges. Regarding data noise, satellite remote sensing is susceptible to atmospheric interference and cloud cover, ground-based sensors can produce errors due to environmental factors (such as electromagnetic interference and equipment aging), and drone data can be affected by flight attitude and lighting variations. Data loss is also common, such as missing data in satellite imagery due to cloud cover, sensor failure periods, or when drone signals are interrupted. Furthermore, data formats, temporal resolution, and spatial resolution vary significantly across different data sources, leading to complex data associations and difficulties in integration and utilization. This hinders accurate and comprehensive assessments of water environmental conditions and prevents timely and effective support for water resource protection and management decisions. Summary of the Invention

[0004] The purpose of the present invention is to provide a water environment monitoring and assessment method and system based on artificial intelligence to solve the above technical problems.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A water environment monitoring and assessment method based on artificial intelligence includes the following steps:

[0007] Integrate multi-source data acquisition interfaces to connect satellite remote sensing receiving equipment, ground sensors, and drone data transmission modules; implement a data cleaning algorithm based on generative adversarial networks and recurrent neural networks combined with an attention mechanism to denoise and fill missing values in raw data, and store the processed data in a database;

[0008] Build a spatiotemporal graph convolutional network and long short-term memory network fusion model based on a deep learning framework, set dynamic weight adjustment parameters; use historical data to train and optimize the model, and store the trained model in a model library;

[0009] Implement feature fusion and decision fusion algorithms to integrate multi-source data; integrate association rule mining algorithms to mine potential associations between data and store association rules in the knowledge base;

[0010] The pre-processed data is obtained from the database and input into the trained model for evaluation. The evaluation results are visualized and warning judgments are made based on the warning thresholds. When the warning conditions are triggered, warning information is issued through various communication methods.

[0011] As a further solution of the present invention: the data cleaning algorithm based on the generative adversarial network and the recurrent neural network combined with the attention mechanism specifically includes the following steps:

[0012] The generator G takes random noise z as input to generate filled data G(z). The discriminator D is used to distinguish between real data x and generated data G(z). The discriminator loss function is:

[0013]

[0014] The generator loss function is:

[0015]

[0016] The recurrent neural network combines the attention mechanism to calculate the attention weight:

[0017]

[0018] where e t =v T tanh(W e h t +b e ), the final output h t is the hidden state of the recurrent neural network.

[0019] As a further solution of the present invention: the method of building a spatiotemporal graph convolutional network and a long short-term memory network fusion model based on a deep learning framework and setting dynamic weight adjustment parameters includes:

[0020] The graph convolution operation formula in the spatiotemporal graph convolutional network is:

[0021]

[0022] Input gate i of the long short-term memory network t 、Forget Gatet , output gate o t and cell state C t Calculate according to the formula;

[0023] Dynamic weight ω=σ(W ω [h ST-GCN ;h LSTM ]+b ω ), the final fusion output h=ωh ST-GCN +(1-ω)h LSTM .

[0024] As a further solution of the present invention: the feature fusion and decision fusion algorithm is implemented to perform fusion processing on multi-source data, including:

[0025] The feature fusion formula is F fusion =W sat F sat +W sen F sen +W uav F uav ;

[0026] The decision fusion adopts the Bayesian fusion algorithm, and the decision probability after fusion is:

[0027]

[0028] As a further solution of the present invention, the pre-processed data is obtained from the database, input into the trained model for evaluation, the evaluation results are visualized, and early warning judgment is made according to the early warning threshold, including:

[0029] The mean square error is used to evaluate the model performance, and the formula is:

[0030] The basis for early warning judgment is to set thresholds T for different monitoring indicators i , when the actual monitoring value X i Greater than T i When the warning is triggered.

[0031] As a further solution of the present invention: the data types of the multi-source data include pH, dissolved oxygen, conductivity, ammonia nitrogen, water level and water flow velocity.

[0032] As a further solution of the present invention: when training the model using historical data, K-fold cross validation is used to improve the model stability and generalization ability, with K=5 or K=10;

[0033] The adaptive learning rate adjustment algorithm is used to dynamically adjust the learning rate, and L1 and L2 regularization are introduced to prevent model overfitting.

[0034] An artificial intelligence-based water environment monitoring and assessment system, comprising:

[0035] Data acquisition and preprocessing module: Integrates multi-source data acquisition interfaces, connects to satellite remote sensing receiving equipment, ground sensors, and drone data transmission modules; implements a data cleaning algorithm based on generative adversarial networks and recurrent neural networks combined with an attention mechanism, denoises the raw data, fills in missing values, and stores the processed data in a database;

[0036] Model building and training module: Build a spatiotemporal graph convolutional network and long short-term memory network fusion model based on the deep learning framework, set dynamic weight adjustment parameters; use historical data to train and optimize the model, and store the trained model in the model library;

[0037] Data fusion and analysis module: implements feature fusion and decision fusion algorithms to integrate multi-source data; integrates association rule mining algorithms to mine potential associations between data and store association rules in the knowledge base;

[0038] Evaluation and early warning module: obtains pre-processed data from the database and inputs it into the trained model for evaluation; visualizes the evaluation results and makes early warning judgments based on the early warning thresholds; when the early warning conditions are triggered, issues early warning information through various communication methods.

[0039] The beneficial effects of the present invention are as follows: the high integration of multi-source data acquisition interfaces and the application of data cleaning algorithms ensure the acquisition of comprehensive and accurate data, improve data quality, provide a solid foundation for subsequent analysis, and solve the problem of one-sided and inaccurate traditional monitoring data; the ST-GCN and LSTM fusion model combined with the dynamic weight adjustment mechanism effectively captures the spatiotemporal characteristics of data and significantly improves the accuracy of assessment. The training optimization strategy enables the model to adapt to complex water environment changes and enhances the practicality and reliability of the model; feature fusion, decision fusion algorithm and association rule mining technology fully tap the value of multi-source data, and the knowledge graph construction provides comprehensive knowledge support for the assessment, helping analysts to deeply understand the internal relationship of the water environment and improve the scientific nature of decision-making; the multi-channel early warning mechanism ensures that early warning information is conveyed to relevant personnel in a timely manner, and the visual display makes the assessment results intuitive and easy to understand, facilitates public participation and decision-making, and improves the efficiency and effectiveness of water environment management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The present invention will be further described below with reference to the accompanying drawings.

[0041] Figure 1 It is a flow chart of a water environment monitoring and assessment method based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0043] See also Figure 1 As shown, the present invention is a water environment monitoring and assessment method based on artificial intelligence, comprising the following steps:

[0044] Multi-source data collection and preprocessing:

[0045] Collection interface integration and data acquisition: Build a universal and highly compatible multi-source data acquisition interface platform. For satellite remote sensing receiving equipment, develop parsing modules that are compatible with a variety of satellite data formats, which can seamlessly connect to mainstream remote sensing satellites at home and abroad, such as the United States' Landsat series and China's Gaofen series of satellites, and receive and parse remote sensing image data containing information such as water body reflectance, temperature, and chlorophyll concentration in real time. In terms of ground sensors, standardized communication protocols and adaptive data acquisition terminals are used, which are compatible with various water quality and hydrological sensors (such as Hach and WTW brand sensors) to achieve real-time collection and transmission of multiple parameters such as pH, dissolved oxygen, conductivity, ammonia nitrogen, water level, and water flow rate. For the drone data transmission module, design a hybrid transmission solution based on wireless ad hoc networks and 5G communications to ensure that drones can stably transmit high-definition images and local water quality monitoring data in complex environments.

[0046] Data cleaning and preprocessing: Use a data cleaning algorithm based on generative adversarial networks (GANs) and long short-term memory networks (LSTMs) combined with attention mechanisms.

[0047] GAN data filling: Let the original data set be X = {x1, x2, ..., x n}, where the sample set with missing values is X miss The generator G takes random noise z as input and generates data G(z) to fill in missing values. The discriminator D receives real data x and generated data G(z). Its goal is to maximize the ability to distinguish between real data and generated data, while the generator's goal is to minimize the discriminator's ability to distinguish. The generator and discriminator are optimized through the following adversarial training process:

[0048] Discriminator loss function:

[0049]

[0050] The generator loss function is:

[0051]

[0052] LSTM combines attention mechanism to process time series: let the input time series data be x t =(x t1 ,x t2 ,...,x td ), where t=1,2,...,T represents the time step and d represents the feature dimension. The input gate i of the LSTM unit t 、Forget Gate t , output gate o t and cell state C t The calculation is as follows:

[0053] i t =σ(W ii x t +W hi h t-1 +b i );

[0054] f t =σ(W if x t +W hf h t-1 +b f );

[0055] C t =f t ⊙C t-1 +i t ⊙tanh(W ic x t +W hc h t-1 +b c );

[0056] O t =σ(W io x t +W ho h t-1 +b o );

[0057] h t =O t ⊙tanh(C t );

[0058] Among them, σ is the sigmoid function, ⊙ is the element-by-element multiplication, W is the weight matrix, b is the bias vector, h t is the hidden state. The attention mechanism calculates the attention weight α t for:

[0059]

[0060] where e t =v T tanh(W e h t +b e ), v, W e and b e is a learnable parameter. The final output

[0061] We train corresponding cleaning models based on the data characteristics of different data sources. The satellite remote sensing data cleaning model focuses on addressing noise and data gaps caused by cloud cover and atmospheric scattering; the ground sensor data cleaning model addresses outliers and missing values caused by environmental interference; and the drone data cleaning model addresses data issues caused by flight attitude changes and lighting differences. The cleaned data is standardized, with unified formats and dimensions, and stored in a distributed database for efficient subsequent access and management.

[0062] Model building and training optimization:

[0063] Fusion model construction: Based on advanced deep learning frameworks (such as TensorFlow or PyTorch), build a deep fusion model of spatiotemporal graph convolutional network (ST-GCN) and long short-term memory network (LSTM).

[0064] ST-GCN extracts spatial features: Let G = (V, E) represent the water monitoring network, where V = {v1, v2, ..., v N} is a node set (representing monitoring points), and E is an edge set (representing the spatial relationship between points). The input feature matrix of the graph convolution layer is Where N is the number of nodes and F is the feature dimension. The graph convolution operation is defined as:

[0065]

[0066] Among them H (l) is the feature matrix of the lth layer, H(0)=X, is the adjacency matrix with self-loops added, yes The degree matrix, W (l) is the learnable weight matrix and σ is the activation function.

[0067] LSTM processing of time series features: the same LSTM calculation method as in the data cleaning section.

[0068] Dynamic weight adjustment: Let the output of ST-GCN be h ST-GCN , the output of LSTM is h LSTM , the dynamic weight ω is calculated by the following formula:

[0069] ω=σ(W ω [h ST-GCN ;h LSTM ]+b ω );

[0070] where [h ST-GCN ;h LSTM ] means to concatenate the two outputs, W ω and b ω is a learnable parameter. The final fusion output h=ωh ST-GCN +(1-ω)h LSTM .

[0071] Model training and optimization: Collect massive historical water environment data, covering multi-source monitoring data in different seasons, different hydrological conditions, and different pollution levels. Carefully annotate the data, divide it into training set, validation set, and test set, and use K-fold cross-validation (K=5 or K=10) to improve model stability and generalization ability. During the training process, use an adaptive learning rate adjustment algorithm (such as AdamW optimizer) to dynamically adjust the learning rate according to the model training situation to prevent the model from falling into a local optimal solution. At the same time, introduce regularization techniques (such as L1 and L2 regularization) to prevent model overfitting and enhance the generalization ability of the model. Use the validation set to evaluate model performance in real time, and the monitoring indicators include accuracy, recall rate, mean square error (MSE), etc. The formula for calculating the mean square error is:

[0072]

[0073] where y i is the true value, is the predicted value, and n is the number of samples. Model hyperparameters, such as the number of network layers, number of nodes, and convolution kernel size, are adjusted promptly based on the evaluation results. After training is complete, the optimal performing model is stored in the model library and regularly updated to adapt to changes in the water environment.

[0074] Data fusion and association mining:

[0075] Feature fusion and decision fusion: Use multi-scale feature fusion algorithm to process multi-source data.

[0076] Feature fusion: Let the satellite remote sensing data feature be F sat , the ground sensor data feature is F sen , the drone data feature is F uav , the fused feature F fusion for:

[0077] F fusion =W sat F sat +W sen F sen +Wuav F uav ;

[0078] Where W sat 、W sen and W uav are learnable weights, determined through training.

[0079] Decision fusion: Using the Bayesian fusion algorithm, let the evaluation results of different data sources be R1, R2, ..., R m , the corresponding prior probabilities are P(R1), P(R2), ..., P(R m ), the probability of each evaluation result under the condition that evidence E appears is P(E|R1), P(E|R2),..., P(E|R m ). According to Bayes’ theorem, the decision probability after fusion P(R i |E) is:

[0080]

[0081] Association rule mining and knowledge graph construction: Integrate association rule mining algorithms such as Apriori and FP-Growth to conduct in-depth mining of multi-source data. Water quality parameters, hydrological conditions, meteorological data, etc. are used as mining objects to discover potential strong association rules. For example, the correlation between the concentration of a certain pollutant and the concentration of another pollutant under specific seasons and water level conditions is discovered. The mined association rules are stored in the knowledge base and a water environment knowledge graph is constructed. The knowledge graph uses entities (such as monitoring points, water quality parameters, pollution sources, etc.) as nodes and association rules as edges to intuitively display the complex relationships between data, providing rich knowledge support for water environment assessment, facilitating analysts to quickly obtain relevant information and assist in decision-making.

[0082] Assessment, early warning and visualization:

[0083] Evaluation and early warning: Obtain pre-processed multi-source data from the distributed database and input it into the trained fusion model for evaluation. The model output includes evaluation results such as water quality level (such as Class I-V water quality classification), degree of eutrophication of water bodies, and possible sources of pollution. According to historical data and industry standards, combined with expert experience, set multi-dimensional early warning thresholds. For example, set thresholds for different pollutant concentrations, water quality change rates, water body ecological indicators, etc. Let the threshold of the i-th monitoring indicator be Ti, and the actual monitoring value be x i , when x i >T iWhen the assessment results trigger an early warning condition, the system activates a multi-channel early warning mechanism. A text message is sent to relevant management personnel via the SMS platform, containing the location, type, and severity of the warning. A detailed early warning report with data charts and analysis results is sent to researchers via the email system. The mobile app push function promptly notifies the public of the environmental conditions of the relevant waters and encourages public participation in supervision and protection.

[0084] Visual display: Using advanced visualization technology, the assessment results are presented in an intuitive and easy-to-understand manner. Using geographic information system (GIS) technology, water quality conditions are marked in real time on electronic maps. Different water quality levels are distinguished by different colors and icons, visually displaying the spatial distribution of water pollution. Dynamic charts (such as line charts and bar charts) are used to display the changing trends of water quality parameters over time. Users can interact with the charts to view data details for specific time periods and specific points. 3D visualization technology is used to display the three-dimensional structure of water bodies and pollutant diffusion simulations, helping users to gain a deeper understanding of the complex changes in the water environment and provide an intuitive basis for decision-making.

[0085] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A water environment monitoring and assessment method based on artificial intelligence, characterized in that: The following steps are involved: Integrate multi-source data acquisition interfaces to connect satellite remote sensing receiving equipment, ground sensors, and UAV data transmission modules; Implement a data cleaning algorithm based on generative adversarial networks and recurrent neural networks combined with an attention mechanism to denoise and fill missing values in the original data, and store the processed data in a database; Build a spatiotemporal graph convolutional network and long short-term memory network fusion model based on a deep learning framework, set dynamic weight adjustment parameters; use historical data to train and optimize the model, and store the trained model in a model library; Implement feature fusion and decision fusion algorithms to integrate multi-source data; integrate association rule mining algorithms to mine potential associations between data and store association rules in the knowledge base; The pre-processed data is obtained from the database and input into the trained model for evaluation. The evaluation results are visualized and warning judgments are made based on the warning thresholds. When the warning conditions are triggered, warning information is issued through various communication methods.

2. The water environment monitoring and assessment method based on artificial intelligence according to claim 1, characterized in that: The data cleaning algorithm based on generative adversarial networks and recurrent neural networks combined with attention mechanisms specifically includes the following steps: The generator G takes random noise z as input to generate filled data G(z). The discriminator D is used to distinguish between real data x and generated data G(z). The discriminator loss function is: The generator loss function is: The recurrent neural network combines the attention mechanism to calculate the attention weight: where e t =v T tanh(W e h t +b e ), the final output h t is the hidden state of the recurrent neural network.

3. The water environment monitoring and assessment method based on artificial intelligence according to claim 1 is characterized in that: The deep learning framework is used to build a spatiotemporal graph convolutional network and long short-term memory network fusion model, and set dynamic weight adjustment parameters, including: The graph convolution operation formula in the spatiotemporal graph convolutional network is: Input gate i of the long short-term memory network t 、Forget Gate t , output gate o t and cell state C t Calculate according to the formula; Dynamic weight ω=σ(W ω [h ST-GCN ;h LSTM ]+b ω ), the final fusion output h=ωh ST-GCN +(1-ω)h LSTM .

4. The water environment monitoring and assessment method based on artificial intelligence according to claim 1, characterized in that: The feature fusion and decision fusion algorithms are implemented to fuse multi-source data, including: The feature fusion formula is F fusion =W sat F sat +W sen F sen +W uav F uav ; The decision fusion adopts the Bayesian fusion algorithm, and the decision probability after fusion is:

5. The water environment monitoring and assessment method based on artificial intelligence according to claim 1 is characterized in that: The method of obtaining pre-processed data from the database, inputting the trained model into the model for evaluation, visualizing the evaluation results, and making early warning judgments based on the early warning threshold includes: The mean square error is used to evaluate the model performance, and the formula is: The basis for early warning judgment is to set thresholds T for different monitoring indicators i , when the actual monitoring value X i Greater than T i When the warning is triggered.

6. The water environment monitoring and assessment method based on artificial intelligence according to claim 1 is characterized in that: The data types of the multi-source data include pH, dissolved oxygen, conductivity, ammonia nitrogen, water level and water flow rate.

7. The water environment monitoring and assessment method based on artificial intelligence according to claim 1 is characterized in that: When training the model using historical data, K-fold cross-validation is used to improve model stability and generalization ability, with K=5 or K=10; The adaptive learning rate adjustment algorithm is used to dynamically adjust the learning rate, and L1 and L2 regularization are introduced to prevent model overfitting.

8. A water environment monitoring and evaluation system based on artificial intelligence, characterized in that: include: Data acquisition and preprocessing module: Integrates multi-source data acquisition interfaces and connects to satellite remote sensing receiving equipment, ground sensors, and UAV data transmission modules; Implement a data cleaning algorithm based on generative adversarial networks and recurrent neural networks combined with an attention mechanism to denoise and fill missing values in the original data, and store the processed data in a database; Model building and training module: Build a spatiotemporal graph convolutional network and long short-term memory network fusion model based on the deep learning framework, set dynamic weight adjustment parameters; use historical data to train and optimize the model, and store the trained model in the model library; Data fusion and analysis module: implements feature fusion and decision fusion algorithms to integrate multi-source data; integrates association rule mining algorithms to mine potential associations between data and store association rules in the knowledge base; Evaluation and early warning module: obtains pre-processed data from the database and inputs it into the trained model for evaluation; visualizes the evaluation results and makes early warning judgments based on the early warning thresholds; when the early warning conditions are triggered, issues early warning information through various communication methods.

Images