Environment-friendly pollution monitoring method and system

Abstract

The invention discloses an environment-friendly pollution monitoring method and system. The method comprises the steps of data acquisition, data preprocessing, construction of an environment-friendly pollution monitoring model, optimization of the environment-friendly pollution monitoring model and real-time early warning. The invention belongs to the field of pollution monitoring, and particularly relates to an environmental pollution monitoring method and system, which can effectively extract space and time characteristics in environmental data by combining a graph convolutional neural network and a long-short-term memory network so as to more accurately predict a future pollution trend. Meanwhile, the optimization process is guided by establishing posterior probability distribution of an objective function, and a globally optimal solution can be found within as few evaluation times as possible. Besides, in the aspect of data preprocessing, an auto-encoder and a probability density estimation algorithm are combined, so that outliers in the environmental data can be effectively identified and eliminated, and the accuracy and efficiency of data processing are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pollution monitoring, and in particular to an environmental pollution monitoring method and system. Background Art

[0002] Current environmental pollution monitoring methods primarily include field monitoring, remote sensing technology, environmental sensors, and biological monitoring. Each of these methods has its advantages, but also has limitations. For example, the complexity and multimodal nature of environmental data make it difficult for traditional methods to process such data; traditional monitoring models perform poorly in capturing spatial and temporal pollution characteristics; and during optimization, conventional model optimization methods struggle to find the global optimal solution within a small number of evaluations.

[0003] Furthermore, current environmental pollution monitoring methods and systems present numerous challenges, including how to effectively fuse and integrate diverse multi-source monitoring data, address the accuracy and stability of sensor data, improve the generalization and prediction accuracy of monitoring models, and utilize monitoring data for accurate environmental pollution prediction and response. Therefore, further research and technological advancements are needed to address these challenges and ensure the continued development and application of environmental pollution monitoring technology. Summary of the Invention

[0004] In view of the above situation, in order to overcome the shortcomings of the existing technology, the present invention provides an environmental pollution monitoring method and system to address the limitations of the existing technology. Compared with traditional methods, the environmental pollution monitoring method and system proposed in this scheme, by combining graph convolutional neural networks and long short-term memory networks, can effectively extract spatial and temporal features from environmental data, thereby more accurately predicting future pollution trends; at the same time, by establishing the posterior probability distribution of the objective function to guide the optimization process, it is possible to find the global optimal solution within the minimum number of evaluations. In addition, in terms of data preprocessing, the combination of autoencoders and probability density estimation algorithms can effectively identify and eliminate outliers in environmental data, improving the accuracy and efficiency of data processing.

[0005] This solution can more accurately capture the complex distribution of data and effectively fuse and integrate it. It can more accurately and stably fit environmental data with multimodal characteristics obtained by multiple sensors. It has higher data processing efficiency and prediction accuracy, and can timely detect and warn of complex environmental pollution events. By comprehensively utilizing multiple data processing and model optimization methods, the system effectively solves the problems of cumbersome data processing, low model accuracy, and unstable prediction effects in traditional methods.

[0006] The technical solution adopted by the present invention is as follows: The present invention provides an environmental pollution monitoring method, which includes the following steps:

[0007] Step S1: data collection;

[0008] Step S2: Data preprocessing, combining autoencoders and probability density estimation algorithms, fitting environmental data with multimodal characteristics to identify and eliminate outliers, including basic data processing, building autoencoders, density estimation, and setting thresholds;

[0009] Step S3: Construct an environmental pollution monitoring model, combining spatial feature extraction and time series prediction, based on a weighted graph convolutional neural network and a long short-term memory network to capture spatial and temporal pollution characteristics, including spatial feature extraction and time series prediction. This constructs an efficient environmental pollution monitoring model for accurate prediction and diagnosis of environmental problems, which can not only capture the time series characteristics of various pollutants, but also predict future pollution trends.

[0010] Step S4: Optimize the environmental pollution monitoring model by selecting a priori probability distribution, collecting sample points, and updating the posterior probability distribution. This continuously strikes a balance between areas with high uncertainty and areas with good predictive performance. Using the acquisition function, the next sample point that is most likely to improve model performance under the current posterior probability distribution is found for optimization, and the dataset is updated.

[0011] Step S5: Real-time warning.

[0012] Furthermore, in step S3, the construction of the environmental pollution monitoring model combines spatial feature extraction and time series prediction, and captures spatial and temporal pollution features based on a weighted graph convolutional neural network and a long short-term memory network, specifically including the following steps:

[0013] Step S31: Spatial feature extraction: Use graph convolutional networks to mine the spatial correlation between monitoring sites, construct a weighted graph composed of monitoring points, and extract spatial features through graph convolutional networks. The algorithm formula of graph convolutional networks is as follows:

[0014]

[0015] Where H (l+1) is the node feature matrix of the l+1th layer; is the adjacency matrix of the weighted graph composed of monitoring points, including self-connections; yes The degree matrix of (l) is the training weight of the lth layer; σ is the activation function;

[0016] Step S32: Time series prediction. Based on the spatial features extracted in step S31, the long short-term memory network is used to learn the time dependency of environmental data and predict future pollution trends. The specific algorithm formula is as follows:

[0017] Ot =LSTM(H (l) ,O t-1 );

[0018] Where, O t is the output vector at time t; H (l) is the output feature matrix of the last layer; O t-1 is the output vector at time t-1; LSTM(·) is the LSTM function.

[0019] Traditional pollution monitoring methods often fail to effectively exploit the spatial correlations between monitoring sites. However, the graph convolutional network employed in step S31 can exploit the complex spatial relationships between monitoring sites and effectively capture spatial features. By constructing a weighted graph of monitoring points and applying the graph convolutional network, the relationships between different monitoring points can be more accurately described, thereby improving the model's ability to understand and express spatial features.

[0020] Environmental data exhibits significant temporal dependencies, but traditional methods often struggle to accurately capture the complex dynamic relationships within time series data. Using the long-short-term memory network (LSTM) based on spatial feature extraction in step S32, the model can better learn the temporal dependencies of environmental data and accurately predict future pollution trends. This time series prediction method can better capture the dynamic changes in pollutants, improving the model's predictive power and effectiveness.

[0021] Preferably, in step S32, formula O t =LSTM(H (l) ,O t-1 ) describes the operation of the LSTM unit at time step t, which includes the following steps:

[0022] Step S321: Forget gate, the specific formula is:

[0023] f t =σ(W f ·[h t-1 ,H (l) ]+b f ;

[0024] Where: f t Is the output of the forget gate, controlling the long-term state C of the previous time step t-1 The information discard ratio in W f 、b f are the forget gate weight matrix and the forget gate bias vector respectively; σ is the activation function, which compresses the output to between 0 and 1;

[0025] Step S322: Input gate, the specific formula is:

[0026] it =σ(W i ·H (l) ,O t-1 ]+b i );

[0027]

[0028] Where: i t It is the output of the input gate, which controls the information storage ratio of the current time step; is the candidate value generated, which combines the compressed space of the new input information; W i 、W c 、b i 、b c are the parameters related to the input gate; σ is the activation function; tanh(·) is the hyperbolic tangent activation function; O t-1 is the output vector at time t-1; H (l) is the node feature matrix of the lth layer;

[0029] Step S323: Unit status update, the specific formula is:

[0030]

[0031] Where: C t is the long-term state of the current time step, which is the state of the previous step scaled by the forget gate output, plus the new state determined by the input gate; f t Is the output of the forget gate, controlling the long-term state C of the previous time step t-1 The information discard ratio in C t-1 is the long-term state of the previous time step; i t It is the output of the input gate, which controls the information storage ratio of the current time step; is the candidate value generated, the compressed space that merges the new input information;

[0032] Step S324: Output gate, the specific formula is:

[0033] O t =σ(W o ·[H (l) ,O t-1 ]+b o );

[0034] h t =O t tanh(C t );

[0035] Where: O t Is the output of the output gate, which determines the next hidden state h t Which parts of will be output;h tis the final output value, calculated based on the activation and long-term state of the output gate; tanh(·) is the hyperbolic tangent activation function; σ is the activation function; H (l) is the output feature matrix of the last layer; b o is the bias vector of the output gate; W o is the weight matrix of the output gate;

[0036] Preferably, σ and tanh(·) are activation functions for adding nonlinear transformation.

[0037] This deep learning model, combining spatial and temporal features, effectively handles the nonlinearity and complexity of environmental monitoring data. It can effectively capture the spatial and temporal dependencies in environmental data, improving the accuracy of pollution monitoring. By extracting spatial features and combining them with time series prediction, the environmental pollution monitoring model can more accurately predict pollution trends. Furthermore, it offers real-time capabilities, enabling real-time pollution monitoring based on continuously updated data.

[0038] In this solution step, the control of the forget gate and the input gate can more effectively capture long-term dependencies and is suitable for time series data that require long-term memory; the gating mechanism in this solution step helps alleviate the gradient vanishing problem and improve the stability and effect of model training; this solution step is suitable for time series prediction and can more accurately predict environmental pollution trends, thereby improving the efficiency and accuracy of environmental pollution monitoring.

[0039] Furthermore, in step S4, the environmental pollution monitoring model optimization guides the optimization process by establishing the posterior probability distribution of the objective function to find the global optimal solution within the minimum number of evaluations, which specifically includes the following steps:

[0040] Step S41: selecting a prior probability distribution to describe the distribution of the target function;

[0041] Step S42: By collecting sample points, the posterior probability distribution is updated, and a balance is continuously achieved between areas with high uncertainty and areas with good prediction performance to guide the determination of the next sample point;

[0042] Step S43: Using the acquisition function, find the next sample point that is most likely to bring about model performance improvement under the current posterior probability distribution, perform optimization, and update the data set.

[0043] Furthermore, the formula for optimizing the environmental pollution monitoring model is as follows:

[0044] The next sample point determination formula is:

[0045] x t+1 =argmaxEI(x|Dt );

[0046] Where: x t+1 Based on the collected sample point D t and acquisition function EI, which determines the next sample point that is most likely to improve model performance under the current posterior probability distribution; D t is the collection of sample points; EI is the acquisition function; x is the sample point, that is, the input parameter or feature in the model; argmax(·) is the sum of EI(x|D t ) Perform a maximization operation, that is, find the x value that maximizes EI;

[0047] Update the dataset formula:

[0048] D t+1 =D t ∪{(x t+1 ,f(x t+1 ))};

[0049] D t+1 is f(x t+1 ) The updated dataset contains the original collected sample points D t And the newly added sample points (x t+1 ,f(x t+1 ));f(x t+1 ) is the objective function, which means that the model is t+1 The performance evaluation value under is used to evaluate the performance of the model under specific input;

[0050] f(x t+1 )=DNNx t+1 ;

[0051] Where DNN represents a deep neural network, which is used to calculate the parameter x t+1 Perform function approximation to obtain the predicted target function value.

[0052] Traditional optimization methods often require a large number of evaluations to find the global optimal solution, which is particularly difficult in high-dimensional spaces. However, the optimization method based on the posterior probability distribution employed in step S4 can find the global optimal solution with the fewest possible evaluations. By continuously updating the posterior probability distribution and using the acquisition function to determine the next sample point, the model can strike a balance between high uncertainty and good predictive performance, effectively guiding the optimization process and improving the efficiency and accuracy of finding the global optimal solution.

[0053] This method continuously improves model performance by updating the dataset and using a deep neural network to approximate the objective function. Updating the dataset and evaluating the model's performance under new sample points helps the model better adapt to data changes and the optimization process, further improving the model's accuracy and generalization capabilities.

[0054] The environmental pollution monitoring model optimization method described in step S4 significantly outperforms existing technologies in addressing difficult-to-solve technical issues such as low efficiency, slow convergence, and instability, providing an effective and reliable technical path for optimizing and improving the performance of environmental pollution monitoring models. Through the aforementioned steps and formulas, this solution makes the environmental pollution monitoring model optimization process more efficient, significantly reducing the number of evaluations, accelerating the model optimization process, and improving the model's predictive power.

[0055] Furthermore, in step S2, the data preprocessing is to combine the autoencoder and the probability density estimation algorithm to fit the environmental data with multimodal characteristics to identify and eliminate outliers. This method can more accurately capture the complex distribution of data when processing multimodal environmental data and improve the effect of outlier detection. The specific steps are as follows:

[0056] Step S21: basic data processing, data cleaning and time registration operations are performed on the collected environmental data to ensure data quality;

[0057] Step S22: construct an autoencoder, input the environmental data into the autoencoder, map the data to a low-dimensional space through the encoder, and then reconstruct the original data through the decoder to minimize the reconstruction error;

[0058] Step S23, density estimation, based on the encoding layer output of the autoencoder, estimates the probability density of the environmental data using a probability density estimation algorithm to obtain a probability density estimate for each environmental data point;

[0059] Step S24: setting a threshold value. Environmental data points below the threshold value are considered as outliers and are removed.

[0060] Preferably, in step S23, the specific formula of the probability density estimation algorithm is as follows:

[0061]

[0062] Where, P(y i ) is the environmental data point x i The probability density estimate is the probability weight π of the environmental data point for each cluster k and the sum of the Gaussian distributions belonging to the cluster; y iis an observed environmental data point, i.e., an instance in the environmental data set; k is the index of the cluster, indicating that there are k clusters; π k is the mixing ratio of the kth cluster, indicating the prior probability that any environmental data point belongs to the kth cluster and satisfies is a Gaussian distribution; μ k and ∑k are the mean vector and covariance matrix of the kth cluster, respectively, and are the parameters of the Gaussian distribution, which control the center position and shape of the cluster respectively;

[0063] in, The specific formula is as follows:

[0064]

[0065] Where, is a Gaussian distribution; y i is an observed environmental data point, i.e., an instance in the environmental dataset; μ k and ∑k are the mean vector and covariance matrix of the kth cluster respectively; d is the dimension of the data points; exp(·) is the natural exponential function; T represents the transpose of the matrix.

[0066] Step S2 of this solution combines the advantages of feature extraction and density estimation from autoencoders, better adapting to the characteristics of complex environmental data and improving the accuracy and robustness of outlier detection. Compared to traditional single Gaussian distributions or simple distance metrics, it can more accurately fit multimodal environmental data and provide greater flexibility and robustness in outlier detection in complex environmental datasets.

[0067] Preferably, in step S1, the data acquisition is to collect environmental data using intelligent sensor networks, drone remote sensing, and satellite remote sensing; the environmental data includes air quality data, water quality data, soil pollutant data, and noise data.

[0068] Furthermore, in step S5, the real-time warning is to combine the real-time collected environmental data and the analysis results of the environmental pollution monitoring model to immediately notify the monitoring personnel when pollution occurs, and dynamically determine the warning threshold. The formula for determining the warning threshold is as follows:

[0069] Threshold alert =β+κ·θ;

[0070] Where β is the historical average of pollutant concentration; θ is the standard deviation; κ is the adjustment factor, which can be adjusted according to monitoring data and environmental protection requirements; Threshold alert is the warning threshold.

[0071] Furthermore, the present invention provides an environmental pollution monitoring system, which includes a data acquisition module, a data preprocessing module, an environmental pollution monitoring model construction module, an environmental pollution monitoring model optimization module, and a real-time early warning module;

[0072] The data acquisition module is used to collect environmental data;

[0073] The data preprocessing module combines the autoencoder and probability density estimation algorithm to fit the environmental data with multimodal characteristics to identify and eliminate outliers, including basic data processing, building an autoencoder, density estimation and setting thresholds to preprocess the environmental data;

[0074] The environmental pollution monitoring model module combines spatial feature extraction and time series prediction to capture spatial and temporal pollution characteristics based on weighted graph convolutional neural networks and long short-term memory networks, including spatial feature extraction and time series prediction to build an environmental pollution monitoring model;

[0075] The environmental pollution monitoring model optimization module selects a priori probability distribution, collects sample points, and updates the posterior probability distribution, constantly striking a balance between areas of high uncertainty and areas of good predictive performance. It uses the acquisition function to find the next sample point that is most likely to improve model performance under the current posterior probability distribution, optimizes it, and updates the data set to optimize the environmental pollution monitoring model.

[0076] The real-time warning module combines the real-time collected environmental data with the analysis results of the environmental pollution monitoring model to immediately notify monitoring personnel when pollution occurs, and dynamically determine the warning threshold to achieve real-time warning.

[0077] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0078] (1) In view of the problem that the complexity and multimodal characteristics of environmental data cause traditional methods to have great difficulties in processing environmental data, this scheme uses spatial feature extraction and time series prediction. The environmental pollution monitoring model can more accurately predict the development trend of environmental pollution. The environmental pollution monitoring model has a certain real-time performance and can perform real-time environmental pollution monitoring under constantly updated data. Through the control of the forget gate and the input gate, it can more effectively capture long-term dependencies and is suitable for time series data that require long-term memory. It helps to alleviate the gradient vanishing problem, improve the stability and effect of model training, and improve the efficiency and accuracy of environmental pollution monitoring.

[0079] (2) In order to address the problem that conventional model optimization methods are difficult to find the global optimal solution within a few evaluations during the optimization process, this scheme continuously updates the posterior probability distribution and uses the acquisition function to determine the next sample point. The model can achieve a balance between high uncertainty and good prediction performance, effectively guide the optimization process, and improve the efficiency and accuracy of finding the global optimal solution.

[0080] (3) In response to the problem that traditional monitoring models perform poorly in capturing spatial and temporal pollution characteristics, this scheme can more accurately capture the complex distribution of data when processing multimodal environmental data, improve the effect of outlier detection, and more accurately fit environmental data with multimodal characteristics. It has high flexibility and robustness when performing outlier detection in complex environmental data sets.

[0081] (4) This scheme combines the advantages of feature extraction and density estimation of autoencoders, which can better adapt to the characteristics of complex environmental data and improve the accuracy and robustness of outlier detection. Compared with traditional schemes, this scheme can more accurately fit environmental data with multimodal characteristics and has higher flexibility and robustness when performing outlier detection in complex environmental datasets.

[0082] (5) Compared with traditional methods, this solution can more accurately capture the complex distribution of data and effectively fuse and integrate it. It can more accurately and stably fit the environmental data with multimodal characteristics obtained by multiple sensors, has higher data processing efficiency and prediction accuracy, and can timely detect and warn of complex environmental pollution events. By comprehensively utilizing multiple data processing and model optimization methods, the system effectively solves the problems of cumbersome data processing, low model accuracy, and unstable prediction results in traditional methods. The system and method provided by this solution can improve the accuracy and coverage of monitoring data, provide more accurate environmental pollution assessment and early warning services, and bring significant technological progress to environmental pollution monitoring and environmental protection management. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A schematic flow chart of an environmental pollution monitoring method provided by the present invention;

[0084] Figure 2 A schematic diagram of an environmental pollution monitoring system provided by the present invention;

[0085] Figure 3 Schematic diagram of the process of step S3 of the present invention.

[0086] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0087] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0088] Example 1, see Figure 1 The present invention provides an environmental pollution monitoring method, which includes the following steps:

[0089] Step S1: data collection;

[0090] Step S2: Data preprocessing, combining autoencoders and probability density estimation algorithms, fitting environmental data with multimodal characteristics to identify and eliminate outliers, including basic data processing, building autoencoders, density estimation, and setting thresholds;

[0091] Step S3: Construct an environmental pollution monitoring model, combining spatial feature extraction and time series prediction, based on a weighted graph convolutional neural network and a long short-term memory network to capture spatial and temporal pollution characteristics, including spatial feature extraction and time series prediction. This constructs an efficient environmental pollution monitoring model for accurate prediction and diagnosis of environmental problems, which can not only capture the time series characteristics of various pollutants, but also predict future pollution trends.

[0092] Step S4: Optimize the environmental pollution monitoring model by selecting a priori probability distribution, collecting sample points, and updating the posterior probability distribution. This continuously strikes a balance between areas with high uncertainty and areas with good predictive performance. Using the acquisition function, the next sample point that is most likely to improve model performance under the current posterior probability distribution is found for optimization, and the dataset is updated.

[0093] Step S5: Real-time warning.

[0094] Example 2, see Figure 3 Based on the above embodiment, this embodiment, in step S3, constructs an environmental pollution monitoring model that combines spatial feature extraction and time series prediction, and captures spatial and temporal pollution features based on a weighted graph convolutional neural network and a long short-term memory network. Specifically, the following steps are included:

[0095] Step S31: Spatial feature extraction: Use graph convolutional networks to mine the spatial correlation between monitoring sites, construct a weighted graph composed of monitoring points, and extract spatial features through graph convolutional networks. The algorithm formula of graph convolutional networks is as follows:

[0096]

[0097] Where H (l+1) is the node feature matrix of the l+1th layer; is the adjacency matrix of the weighted graph composed of monitoring points, including self-connections; yes The degree matrix of (l) is the training weight of the lth layer; σ is the activation function;

[0098] Step S32: Time series prediction. Based on the spatial features extracted in step S31, the long short-term memory network is used to learn the time dependency of environmental data and predict future pollution trends. The specific algorithm formula is as follows:

[0099] O t =LSTM(H (l) ,O t-1 );

[0100] Where, O t is the output vector at time t; H (l) is the output feature matrix of the last layer; O t-1 is the output vector at time t-1; LSTM(·) is the LSTM function.

[0101] Traditional pollution monitoring methods often fail to effectively exploit the spatial correlations between monitoring sites. However, the graph convolutional network employed in step S31 can exploit the complex spatial relationships between monitoring sites and effectively capture spatial features. By constructing a weighted graph of monitoring points and applying the graph convolutional network, the relationships between different monitoring points can be more accurately described, thereby improving the model's ability to understand and express spatial features.

[0102] Environmental data exhibits significant temporal dependencies, but traditional methods often struggle to accurately capture the complex dynamic relationships within time series data. Using the long-short-term memory network (LSTM) based on spatial feature extraction in step S32, the model can better learn the temporal dependencies of environmental data and accurately predict future pollution trends. This time series prediction method can better capture the dynamic changes in pollutants, improving the model's predictive power and effectiveness.

[0103] Example 3: This example is based on the above example. In step S32, formula O t =LSTM(H (l) ,O t-1 ) describes the operation of the LSTM unit at time step t, which includes the following steps:

[0104] Step S321: Forget gate, the specific formula is:

[0105] f t =σ(W f·[h t-1 ,H (l) ]+b f ;

[0106] Where: f t Is the output of the forget gate, controlling the long-term state C of the previous time step t-1 The information discard ratio in W f 、b f are the forget gate weight matrix and the forget gate bias vector respectively; σ is the activation function, which compresses the output to between 0 and 1;

[0107] Step S322: Input gate, the specific formula is:

[0108] i t =σ(W i ·[H (l) ,O t-1 ]+b i );

[0109]

[0110] Where: i t It is the output of the input gate, which controls the information storage ratio of the current time step; is the candidate value generated, which combines the compressed space of the new input information; W i 、W c 、b i 、b c are the parameters related to the input gate; σ is the activation function; tanh(·) is the hyperbolic tangent activation function; O t-1 is the output vector at time t-1; H (l) is the node feature matrix of the lth layer;

[0111] Step S323: Unit status update, the specific formula is:

[0112]

[0113] Where: C t is the long-term state of the current time step, which is the state of the previous step scaled by the forget gate output, plus the new state determined by the input gate; f t Is the output of the forget gate, controlling the long-term state C of the previous time step t-1 The information discard ratio in C t-1 is the long-term state of the previous time step; i t It is the output of the input gate, which controls the information storage ratio of the current time step; is the candidate value generated, the compressed space that merges the new input information;

[0114] Step S324: Output gate, the specific formula is:

[0115] O t =σ(W o ·[H (l) ,O t-1 ]+b o );

[0116] h t =O t tanh(C t );

[0117] Where: O t Is the output of the output gate, which determines the next hidden state h t Which parts of will be output;h t is the final output value, calculated based on the activation and long-term state of the output gate; tanh(·) is the hyperbolic tangent activation function; σ is the activation function; H (l) is the output feature matrix of the last layer; b o is the bias vector of the output gate; W o is the weight matrix of the output gate;

[0118] As a preferred embodiment of this embodiment, σ and tanh(·) are activation functions used to add nonlinear transformation.

[0119] The deep learning model provided in this embodiment, which combines spatial and temporal features, can effectively handle the nonlinearity and complexity of environmental monitoring data, effectively capturing the spatial correlation and temporal dependence in environmental data, and improving the accuracy of pollution monitoring. By extracting spatial features and combining time series prediction, the environmental pollution monitoring model can more accurately predict the development trends of environmental pollution. This environmental pollution monitoring model also has a certain degree of real-time performance, allowing for real-time environmental pollution monitoring using continuously updated data.

[0120] The steps of this embodiment can more effectively capture long-term dependencies through the control of the forget gate and the input gate, and are suitable for time series data that require long-term memory; the gating mechanism in the steps of this solution helps to alleviate the gradient vanishing problem and improve the stability and effect of model training; the steps of this solution are suitable for time series prediction and can more accurately predict environmental pollution trends, thereby improving the efficiency and accuracy of environmental pollution monitoring.

[0121] Example 4, based on the above example, in step S4, the environmental pollution monitoring model optimization guides the optimization process by establishing a posterior probability distribution of the objective function, and finds the global optimal solution within the minimum number of evaluations. Specifically, the steps include:

[0122] Step S41: selecting a prior probability distribution to describe the distribution of the target function;

[0123] Step S42: By collecting sample points, the posterior probability distribution is updated, and a balance is continuously achieved between areas with high uncertainty and areas with good prediction performance to guide the determination of the next sample point;

[0124] Step S43: Using the acquisition function, find the next sample point that is most likely to bring about model performance improvement under the current posterior probability distribution, perform optimization, and update the data set.

[0125] Furthermore, the formula for optimizing the environmental pollution monitoring model is as follows:

[0126] The next sample point determination formula is:

[0127] x t+1 =argmaxEI(x|D t );

[0128] Where: x t+1 Based on the collected sample point D t and acquisition function EI, which determines the next sample point that is most likely to improve model performance under the current posterior probability distribution; D t is the collection of sample points; EI is the acquisition function; x is the sample point, that is, the input parameter or feature in the model; argmax(·) is the sum of EI(x|D t ) Perform a maximization operation, that is, find the x value that maximizes EI;

[0129] Update the dataset formula:

[0130] D t+1 =D t ∪{(x t+1 ,f(x t+1 ))};

[0131] D t+1 is f(x t+1 ) The updated dataset contains the original collected sample points D t And the newly added sample points (x t+1 ,f(x t+1 ));f(x t+1 ) is the objective function, which means that the model is t+1 The performance evaluation value under is used to evaluate the performance of the model under specific input;

[0132] f(x t+1 )=DNNx t+1 ;

[0133] Where DNN represents a deep neural network, which is used to calculate the parameter x t+1 Perform function approximation to obtain the predicted target function value.

[0134] Traditional optimization methods often require a large number of evaluations to find the global optimal solution, which is particularly difficult in high-dimensional spaces. However, the optimization method based on the posterior probability distribution employed in step S4 can find the global optimal solution with the fewest possible evaluations. By continuously updating the posterior probability distribution and using the acquisition function to determine the next sample point, the model can strike a balance between high uncertainty and good predictive performance, effectively guiding the optimization process and improving the efficiency and accuracy of finding the global optimal solution.

[0135] This embodiment continuously improves the model's performance by updating the dataset and using a deep neural network to approximate the objective function. Updating the dataset and evaluating the model's performance under new sample points helps the model better adapt to data changes and the optimization process, further improving the model's accuracy and generalization capabilities.

[0136] The environmental pollution monitoring model optimization method described in step S4 significantly outperforms existing technologies in resolving difficult-to-solve technical issues such as low efficiency, slow convergence, and instability, providing an effective and reliable technical path for optimizing and improving the performance of environmental pollution monitoring models. Through the above steps and formulas, this embodiment makes the environmental pollution monitoring model optimization process more efficient, significantly reduces the number of evaluations, accelerates the model optimization process, and improves the model's predictive power.

[0137] Example 5: This example is based on the above example. In step S2, data preprocessing is to combine the autoencoder and the probability density estimation algorithm to fit the environmental data with multimodal characteristics to identify and eliminate outliers. This method can more accurately capture the complex distribution of data when processing multimodal environmental data, and improve the effect of outlier detection. The specific steps are as follows:

[0138] Step S21: basic data processing, data cleaning and time registration operations are performed on the collected environmental data to ensure data quality;

[0139] Step S22: construct an autoencoder, input the environmental data into the autoencoder, map the data to a low-dimensional space through the encoder, and then reconstruct the original data through the decoder to minimize the reconstruction error;

[0140] Step S23, density estimation, based on the encoding layer output of the autoencoder, estimates the probability density of the environmental data using a probability density estimation algorithm to obtain a probability density estimate for each environmental data point;

[0141] Step S24: setting a threshold value. Environmental data points below the threshold value are considered as outliers and are removed.

[0142] As a preferred solution of this embodiment, in step S23, the specific formula of the probability density estimation algorithm is as follows:

[0143]

[0144] Where, P(y i ) is the environmental data point x i The probability density estimate is the probability weight π of the environmental data point for each cluster k and the sum of the Gaussian distributions belonging to the cluster; y i is an observed environmental data point, i.e., an instance in the environmental data set; k is the index of the cluster, indicating that there are k clusters; π k is the mixing ratio of the kth cluster, indicating the prior probability that any environmental data point belongs to the kth cluster and satisfies is a Gaussian distribution; μ k and ∑k are the mean vector and covariance matrix of the kth cluster, respectively, and are the parameters of the Gaussian distribution, which control the center position and shape of the cluster respectively;

[0145] in, The specific formula is as follows:

[0146]

[0147] Where, is a Gaussian distribution; y i is an observed environmental data point, i.e., an instance in the environmental dataset; μ k and ∑k are the mean vector and covariance matrix of the kth cluster respectively; d is the dimension of the data points; exp(·) is the natural exponential function; T represents the transpose of the matrix.

[0148] This embodiment combines the advantages of autoencoder feature extraction and density estimation to better adapt to the characteristics of complex environmental data and improve the accuracy and robustness of outlier detection. Compared with traditional single Gaussian distribution or simple distance metrics, it can more accurately fit environmental data with multimodal characteristics and has higher flexibility and robustness when detecting outliers in complex environmental datasets.

[0149] Example 6: This example is based on the above example. In step S1, data collection uses intelligent sensor networks, drone remote sensing, and satellite remote sensing to collect environmental data; the environmental data includes air quality data, water quality data, soil pollutant data, and noise data.

[0150] Example 7, based on the above example, in step S5, real-time warning is to combine real-time collected environmental data and environmental pollution monitoring model analysis results, immediately notify monitoring personnel when pollution occurs, and dynamically determine the warning threshold. The formula for determining the warning threshold is as follows:

[0151] Threshold alert =β+κ·θ;

[0152] Where β is the historical average of pollutant concentration; θ is the standard deviation; κ is the adjustment factor, which can be adjusted according to monitoring data and environmental protection requirements; Threshold alert is the warning threshold.

[0153] Example 8, see Figure 2 This embodiment is based on the above embodiment. The present invention provides an environmental pollution monitoring system, including a data acquisition module, a data preprocessing module, an environmental pollution monitoring model construction module, an environmental pollution monitoring model optimization module, and a real-time early warning module;

[0154] The data acquisition module is used to collect environmental data;

[0155] The data preprocessing module combines autoencoders and probability density estimation algorithms to fit environmental data with multimodal characteristics to identify and eliminate outliers. This includes basic data processing, building an autoencoder, density estimation, and setting thresholds to preprocess environmental data.

[0156] Construct an environmental pollution monitoring model module that combines spatial feature extraction and time series prediction. It uses weighted graph convolutional neural networks and long short-term memory networks to capture spatial and temporal pollution characteristics. This includes spatial feature extraction and time series prediction to build an environmental pollution monitoring model.

[0157] The environmental pollution monitoring model optimization module selects a priori probability distribution, collects sample points, and updates the posterior probability distribution, constantly striking a balance between areas of high uncertainty and areas of good predictive performance. Using the acquisition function, it finds the next sample point most likely to improve model performance under the current posterior probability distribution, optimizes it, and updates the data set to optimize the environmental pollution monitoring model.

[0158] The real-time warning module combines real-time collected environmental data with the analysis results of the environmental pollution monitoring model to immediately notify monitoring personnel when pollution occurs, and dynamically determine the warning threshold to achieve real-time warning.

[0159] The technical solution provided by the present invention has obvious advantages in solving difficult problems in the field of environmental pollution monitoring, can improve the accuracy and efficiency of the monitoring model, and provide better support and guidance for environmental protection and pollution control.

[0160] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0161] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0162] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A method for monitoring environmental pollution, characterized by: The method comprises the following steps: Step S1: data collection; Step S2: Data preprocessing, combining autoencoders and probability density estimation algorithms, fitting environmental data with multimodal characteristics to identify and eliminate outliers, including basic data processing, building autoencoders, density estimation, and setting thresholds; Step S3: Construct an environmental pollution monitoring model, combining spatial feature extraction and time series prediction, based on weighted graph convolutional neural networks and long short-term memory networks to capture spatial and temporal pollution characteristics, including spatial feature extraction and time series prediction; Step S4: Optimize the environmental pollution monitoring model by selecting a priori probability distribution, collecting sample points, and updating the posterior probability distribution. This continuously strikes a balance between areas with high uncertainty and areas with good predictive performance. Using the acquisition function, the next sample point that is most likely to improve model performance under the current posterior probability distribution is found for optimization, and the dataset is updated. Step S5: Real-time warning.

2. The environmental pollution monitoring method according to claim 1, characterized in that: In step S3, the construction of the environmental pollution monitoring model combines spatial feature extraction and time series prediction, and captures spatial and temporal pollution features based on a weighted graph convolutional neural network and a long short-term memory network. Specifically, the following steps are included: Step S31: Spatial feature extraction: Use graph convolutional networks to mine the spatial correlation between monitoring sites, construct a weighted graph composed of monitoring points, and extract spatial features through graph convolutional networks. The algorithm formula of graph convolutional networks is as follows: Where H (I+1) is the node feature matrix of the l+1th layer; is the adjacency matrix of the weighted graph composed of monitoring points, including self-connections; yes The degree matrix of (l) is the training weight of the lth layer; σ is the activation function; Step S32: Time series prediction. Based on the spatial features extracted in step S31, the long short-term memory network is used to learn the time dependency of environmental data and predict future pollution trends. The specific algorithm formula is as follows: O t =LSTM(H (l), O t-1 ); Where, O t is the output vector at time t; H (l) is the output feature matrix of the last layer; O t-1 is the output vector at time t-1; LSTM(·) is the LSTM function.

3. The environmental pollution monitoring method according to claim 2, characterized in that: In step S32, formula O t =LSTM(H (l) ,O t-1 ) describes the operation of the LSTM unit at time step t, which includes the following steps: Step S321: Forget gate, the specific formula is: f t =σ(W f ·[h t-1 ,H (l) ]+b f ; Where: f t Is the output of the forget gate, controlling the long-term state C of the previous time step t-1 The information discard ratio in W f 、b f are the forget gate weight matrix and the forget gate bias vector respectively; σ is the activation function, which compresses the output to between 0 and 1; Step S322: Input gate, the specific formula is: I t =σ(W i ·[H (l) ,O t-1 ]+b i ); Where: i t It is the output of the input gate, which controls the information storage ratio of the current time step; is the candidate value generated, the compressed space that merges the new input information; W i 、W c are the weight matrices of the input gate; b i 、b c are the bias vectors of the input gate; σ is the activation function; tanh(·) is the hyperbolic tangent activation function; t-1 is the output vector at time t-1; H (l) is the node feature matrix of the lth layer; Step S323: Unit status update, the specific formula is: Where: C t is the long-term state of the current time step, which is the state of the previous step scaled by the forget gate output, plus the new state determined by the input gate; f t Is the output of the forget gate, controlling the long-term state C of the previous time step t-1 The information discard ratio in C t-1 is the long-term state of the previous time step; i t It is the output of the input gate, which controls the information storage ratio of the current time step; is the candidate value generated, the compressed space that merges the new input information; Step S324: Output gate, the specific formula is: The t =σ(W o ·[H (l) ,O t-1 ]+b o ); h t =O t ·tanh(C t ); Where: O t Is the output of the output gate, which determines the next hidden state h t Which parts of will be output;h t is the final output value, calculated based on the activation and long-term state of the output gate; tanh(·) is the hyperbolic tangent activation function; σ is the activation function; H (l) is the output feature matrix of the last layer; b o is the bias vector of the output gate; W o is the weight matrix of the output gate.

4. The environmental pollution monitoring method according to claim 1, characterized in that: In step S4, the environmental pollution monitoring model optimization guides the optimization process by establishing the posterior probability distribution of the objective function to find the global optimal solution within the minimum number of evaluations, which specifically includes the following steps: Step S41: selecting a prior probability distribution to describe the distribution of the target function; Step S42: By collecting sample points, the posterior probability distribution is updated, and a balance is continuously achieved between areas with high uncertainty and areas with good prediction performance to guide the determination of the next sample point; Step S43: Using the acquisition function, find the next sample point that is most likely to bring about model performance improvement under the current posterior probability distribution, perform optimization, and update the data set.

5. The environmental pollution monitoring method according to claim 4, characterized in that: The formula for optimizing the environmental pollution monitoring model is as follows: The next sample point determination formula is: x t+1 =argmaxEI(x|D t ); Where: x t+1 Based on the collected sample point D t and acquisition function EI, which determines the next sample point that is most likely to improve model performance under the current posterior probability distribution; D t is the collection of sample points; EI is the acquisition function; x is the sample point, that is, the input parameter or feature in the model; argmax(·) is the sum of EI(x|D t ) Perform a maximization operation, that is, find the x value that maximizes EI; Update the dataset formula: D t+1 =D t ∪{(x t+1 ,f(x t+1 ))}; D t+1 is f(x t+1 ) The updated dataset contains the original collected sample points D t And the newly added sample points (x t+1 ,f(x t+1 ));f(x t+1 ) is the objective function, which means that the model is t+1 The performance evaluation value under is used to evaluate the performance of the model under specific input; f(x t+1 )=DNNx t+1 ; Where DNN represents a deep neural network, which is used to calculate the parameter x t+1 Perform function approximation to obtain the predicted target function value.

6. The environmental pollution monitoring method according to claim 1, characterized in that: In step S2, the data preprocessing is to combine the autoencoder and the probability density estimation algorithm to fit the environmental data with multimodal characteristics to identify and eliminate outliers. The specific steps are as follows: Step S21: basic data processing, data cleaning and time registration operations are performed on the collected environmental data to ensure data quality; Step S22: construct an autoencoder, input the environmental data into the autoencoder, map the data to a low-dimensional space through the encoder, and then reconstruct the original data through the decoder to minimize the reconstruction error; Step S23, density estimation, based on the encoding layer output of the autoencoder, estimates the probability density of the environmental data using a probability density estimation algorithm to obtain a probability density estimate for each environmental data point; Step S24: setting a threshold value. Environmental data points below the threshold value are considered as outliers and are removed.

7. The environmental pollution monitoring method according to claim 6, characterized in that: In step S23, the specific formula of the probability density estimation algorithm is as follows: Where, P(y i ) is the environmental data point x i The probability density estimate is the probability weight π of the environmental data point for each cluster k and the sum of the Gaussian distributions belonging to the cluster; y i is an observed environmental data point, i.e., an instance in the environmental data set; k is the index of the cluster, indicating that there are k clusters; π k is the mixing ratio of the kth cluster, indicating the prior probability that any environmental data point belongs to the kth cluster and satisfies is a Gaussian distribution; μ k and ∑k are the mean vector and covariance matrix of the kth cluster, respectively, and are the parameters of the Gaussian distribution, which control the center position and shape of the cluster respectively; in, The specific formula is as follows: Where, is a Gaussian distribution; y i is an observed environmental data point, i.e., an instance in the environmental dataset; μ k and ∑k are the mean vector and covariance matrix of the kth cluster respectively; d is the dimension of the data points; exp(·) is the natural exponential function; T represents the transpose of the matrix.

8. The environmental pollution monitoring method according to claim 1, characterized in that: In step S1, the data collection is to collect environmental data using intelligent sensor networks, drone remote sensing, and satellite remote sensing; the environmental data includes air quality data, water quality data, soil pollutant data, and noise data.

9. The environmental pollution monitoring method according to claim 1, characterized in that: In step S5, the real-time warning is to combine the real-time collected environmental data and the analysis results of the environmental pollution monitoring model to immediately notify the monitoring personnel when pollution occurs and dynamically determine the warning threshold. The formula for determining the warning threshold is as follows: Threshold alert =β+κ·θ; Where β is the historical average of pollutant concentration; θ is the standard deviation; k is the adjustment factor, which can be adjusted according to monitoring data and environmental protection requirements; Threshold alert is the warning threshold.

10. An environmental data acquisition and monitoring control system based on the Internet of Things, used to implement an environmental pollution monitoring method according to any one of claims 1 to 9, characterized in that: It includes data acquisition module, data preprocessing module, environmental pollution monitoring model construction module, environmental pollution monitoring model optimization module, and real-time early warning module; The data acquisition module is used to collect environmental data; The data preprocessing module combines the autoencoder and probability density estimation algorithm to fit the environmental data with multimodal characteristics to identify and eliminate outliers, including basic data processing, building an autoencoder, density estimation and setting thresholds to preprocess the environmental data; The environmental pollution monitoring model module combines spatial feature extraction and time series prediction to capture spatial and temporal pollution characteristics based on weighted graph convolutional neural networks and long short-term memory networks, including spatial feature extraction and time series prediction to build an environmental pollution monitoring model; The environmental pollution monitoring model optimization module selects a priori probability distribution, collects sample points, and updates the posterior probability distribution, constantly striking a balance between areas of high uncertainty and areas of good predictive performance. It uses the acquisition function to find the next sample point that is most likely to improve model performance under the current posterior probability distribution, optimizes it, and updates the data set to optimize the environmental pollution monitoring model. The real-time warning module combines the real-time collected environmental data with the analysis results of the environmental pollution monitoring model to immediately notify monitoring personnel when pollution occurs, and dynamically determine the warning threshold to achieve real-time warning.

Images