Long-time-sequence intelligent analysis method and system for pollution risk of industrial gathering area

Abstract

The invention provides a long-time-sequence intelligent analysis method and system for the pollution risk of an industrial gathering area, belongs to the field of pollution risk management, and is used for solving the problem that the pollution risk of the industrial gathering area is difficult to dynamically and accurately analyze in the related technology. According to the method, long-time-sequence environment monitoring data and remote sensing image data can be processed and analyzed step by step based on a self-developed intelligent algorithm model, pollution risk data of the industrial gathering area is determined, dynamic and accurate monitoring of the pollution risk in the industrial gathering area is achieved, and better management of the pollution risk in the industrial gathering area is facilitated.

Description

Technical Field

[0001] The present application relates to the field of pollution risk management, and in particular to a method and system for intelligent analysis of pollution risks in industrial clusters over a long time series. Background Art

[0002] With the acceleration of global industrialization, the petrochemical industry, as an important economic pillar, has increasingly prominent environmental risks in its agglomeration areas. Petrochemical industrial agglomeration areas usually cover multiple upstream and downstream industries, involving complex production processes and a large amount of raw material storage, processing and waste disposal. These characteristics make their soil and groundwater pollution risks high, especially in the event of accidents or leaks, which pose a serious threat to the surrounding environment and human health. Therefore, timely and accurate identification and assessment of high-risk areas and formulation of corresponding emergency management measures are crucial for environmental protection and public safety.

[0003] Traditional environmental monitoring methods mostly rely on on-site sampling and laboratory analysis, which are costly, time-consuming, and difficult to fully reflect the risk status in the region. In recent years, the rapid development of remote sensing technology has provided a new means for environmental monitoring. By acquiring high-resolution remote sensing images, large areas can be monitored quickly and effectively. However, the application of existing remote sensing technology in petrochemical clusters still has certain limitations, especially in terms of the accuracy of dynamic change monitoring and risk assessment. Summary of the invention

[0004] The present application provides a long-term intelligent analysis method and system for pollution risks in industrial agglomeration areas, which can dynamically and accurately analyze the pollution risks in industrial agglomeration areas, and is conducive to better management of pollution risks in industrial agglomeration areas.

[0005] In the first aspect, the present application provides a method for intelligent analysis of pollution risks in industrial agglomeration areas over a long time series. The method comprises:

[0006] Obtain long-term environmental monitoring data and remote sensing image data for each divided area within the industrial agglomeration area;

[0007] Analyzing the remote sensing image data to obtain industrial characteristic data;

[0008] Determine the transient pollution risk of each divided area in a long time series according to the environmental monitoring data and the industrial characteristic data;

[0009] Analyze long-term environmental monitoring data, industrial characteristic data and transient pollution risks to obtain environmental change data, characteristic change data and risk change data;

[0010] fusing the environmental monitoring data and the industrial characteristic data to obtain fused characteristic data;

[0011] The environmental monitoring data, industrial characteristic data, transient pollution risk, environmental change data, characteristic change data, risk change data and fused characteristic data are input into a pre-acquired risk analysis model to obtain pollution risk data for industrial agglomeration areas.

[0012] By adopting the above-mentioned technical solution, it is possible to determine the pollution risk data of industrial agglomeration areas based on the gradual processing and analysis of long-term environmental monitoring data and remote sensing image data, thereby achieving dynamic and accurate monitoring of pollution risks in industrial agglomeration areas, which is conducive to better management of pollution risks in industrial agglomeration areas.

[0013] Furthermore, determining the transient pollution risk of each divided area in a long time series according to the environmental monitoring data and the industrial characteristic data includes:

[0014] Determine the current concentration data of each pollutant in each environmental medium based on the environmental monitoring data;

[0015] Determining industrial pollution risks and the pollution concentration impact of each pollutant in each environmental medium based on the industrial characteristic data;

[0016] Determine the associated pollution sub-risk of each pollutant in each environmental medium based on the current concentration data, the pollution concentration impact and the pollutant concentration threshold preset for each pollutant for each environmental medium;

[0017] The transient pollution risk is determined by combining the industrial pollution risk and the associated pollution sub-risks.

[0018] Further, the determining of the industrial pollution risk and the pollution concentration impact of each pollutant in each environmental medium based on the industrial characteristic data includes:

[0019] Identifying the industrial characteristics carried by the industrial characteristic data;

[0020] Determine the industrial pollution sub-risk of each industrial characteristic and the pollution concentration sub-impact of each pollutant in each environmental medium based on the pre-acquired industrial characteristic impact table, wherein the industrial characteristic impact table includes the comparison relationship between industrial characteristics and industrial pollution sub-risks and the comparison relationship between industrial characteristics, pollutants and environmental media and pollutant concentration sub-impacts;

[0021] The sum of all industrial pollution sub-risks is determined as the industrial pollution risk, and the sum of all pollution concentration sub-impacts facing pollutants in environmental media is determined as the pollution concentration impact.

[0022] Further, the determining of the associated pollution sub-risk of each pollutant in each environmental medium based on the current concentration data, the pollution concentration impact and the pollutant concentration threshold preset for each pollutant for each environmental medium includes:

[0023] Suppose the current concentration data of the i-th pollutant in the j-th environmental medium is C dij , the pollutant concentration threshold is C pij , the pollution concentration is v ij , the associated pollution sub-risk is Z ij ,but

[0024] Further, the determining the transient pollution risk by combining the industrial pollution risk and the associated pollution sub-risks includes:

[0025] Assume that the associated pollution risk of the i-th pollutant in the j-th environmental medium is Z ij , industrial pollution risk Z g , the transient contamination risk is Z s ,but

[0026] Z s =1-e -x

[0027]

[0028] In the formula, w g is the preset industrial pollution weight, w ij is the preset associated pollution weight of the i-th pollutant in the j-th environmental medium, A is a preset constant and A≥max(w g Z g , w ij Z ij ).

[0029] Furthermore, the fusing of the environmental monitoring data and the industrial characteristic data to obtain the fused characteristic data includes:

[0030] The environmental monitoring data and the industrial characteristic data are both constructed into high-dimensional vectors to obtain an environmental monitoring vector and an industrial characteristic vector;

[0031] The environmental monitoring vector and the industrial feature vector are fused to obtain the fused feature data, wherein the dimension of the fused feature data is not higher than the maximum value of the dimensions of the environmental monitoring vector and the industrial feature vector.

[0032] Furthermore, the risk analysis model includes a long short-term memory network sub-model, a time series feature analysis sub-model and a neural network sub-model;

[0033] The long short-term memory network sub-model is used to analyze long-term environmental monitoring data, industrial characteristic data, transient pollution risk, environmental change data, characteristic change data and risk change data, respectively, to obtain environmental monitoring time series characteristics, industrial characteristic time series characteristics, pollution risk time series characteristics, environmental change time series characteristics, characteristic change time series characteristics and risk change time series characteristics;

[0034] The time series feature analysis sub-model is used to analyze the time series feature prominence of environmental monitoring time series features, industrial feature time series features, pollution risk time series features, environmental change time series features, feature change time series features and risk change time series features, and divide the environmental monitoring data, industrial feature data, transient pollution risk, environmental change data, feature change data and risk change data into time series dependent data and non-time series dependent data based on the time series feature prominence;

[0035] The neural network sub-model is used to input the time series features of the time series dependent data and fuse the feature data and the non-time series dependent data to output the pollution risk data.

[0036] Furthermore, the time series feature analysis sub-model includes:

[0037] According to the ratio of the vector modulus of the time series feature to the square of the vector dimension number, the time series feature significance of the corresponding time series feature is determined as the time series feature significance. The time series feature with higher significance than at least three other time series features or higher than a preset feature significance threshold is determined as the obvious time series feature. The data with obvious time series features in the environmental monitoring data, industrial characteristic data, transient pollution risks, environmental change data, characteristic change data and risk change data are determined as time series dependent data, and the others are non-time series dependent data.

[0038] In a second aspect, the present application provides a long-term intelligent analysis system for pollution risk in industrial agglomeration areas, wherein the system applies any of the methods described in the first aspect above.

[0039] In summary, this application at least has the following beneficial effects:

[0040] A method and system for intelligent analysis of pollution risks in industrial agglomeration areas with a long time series is provided, which can gradually and intelligently analyze long-time environmental monitoring data and remote sensing image data, thereby realizing dynamic and accurate monitoring of pollution risks in industrial agglomeration areas.

[0041] It should be understood that the contents described in the Summary of the Invention are not intended to limit the key or important features of the embodiments of the present application, nor are they intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The above and other features, advantages and aspects of the embodiments of the present application will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0043] Figure 1 A flow chart of a method for intelligent analysis of pollution risks in industrial clusters over a long time period in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] In addition, the term "and / or" in this article is only a description of the association relationship between the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0046] The present application provides a long-term intelligent analysis method and system for pollution risks in industrial agglomeration areas, which can dynamically and accurately analyze the pollution risks in industrial agglomeration areas, and is conducive to better management of pollution risks in industrial agglomeration areas.

[0047] In a first aspect, the present application discloses a method for intelligently analyzing pollution risks in industrial clusters over a long time series, which can be executed by a server.

[0048] Figure 1 A flow chart of a method for intelligent analysis of pollution risks in industrial clusters over a long time period in an embodiment of the present application is shown.

[0049] Reference Figure 1 , the method specifically comprises the following steps:

[0050] S110: Obtain long-term environmental monitoring data and remote sensing image data for each divided area in the industrial agglomeration area.

[0051] Environmental monitoring data and remote sensing image data are collected by environmental sensors and remote sensing systems connected to the server respectively. Environmental monitoring data is used to reflect the real-time pollution situation in industrial agglomeration areas. Specifically, it can include soil pollutant concentrations of various pollutants with timestamps, groundwater pollutant concentrations, air pollution data, surface water quality monitoring data, etc. Environmental monitoring data can be pre-processed by data cleaning and noise reduction to ensure the quality of environmental monitoring data. Remote sensing image data is used to reflect important devices and their evolution in industrial agglomeration areas. It carries a timestamp; remote sensing image data can be pre-processed by radiation correction, atmospheric correction, geometric correction, data denoising, etc. to ensure that the information in the remote sensing image data is clear and discernible.

[0052] S120: Analyze the remote sensing image data to obtain industrial feature data.

[0053] In the method of this step, the industrial characteristic data includes industrial characteristics.

[0054] The method of this step can be implemented based on a convolutional neural network model, and specifically relies on a convolutional neural network model and a pre-trained data set to identify the types of production facilities, production link information, etc. in the remote sensing image data. The industrial features in the industrial feature data reflect the types of production facilities, production link information, etc. contained in the remote sensing image data one by one. Each different type of production facility, production link information, etc. corresponds to a special industrial feature. The industrial feature data composed of all industrial features is used to reflect the overall situation of the production facility types, production link information, etc. of the industrial agglomeration area.

[0055] In the method of this step, the industrial feature data carries a timestamp, that is, the long-time series industrial feature data is obtained based on the long-time series remote sensing image data.

[0056] S130: Determine the long-term transient pollution risk of each divided area according to the environmental monitoring data and the industrial characteristic data.

[0057] The method of this step specifically includes: determining the current concentration data of each pollutant in each environmental medium based on the environmental monitoring data; determining the industrial pollution risk and the pollution concentration impact of each pollutant in each environmental medium based on the industrial characteristic data; determining the associated pollution sub-risk of each pollutant in each environmental medium based on the current concentration data, the pollution concentration impact and the pollutant concentration threshold preset for each pollutant for each environmental medium; and determining the transient pollution risk in combination with the industrial pollution risk and the associated pollution sub-risk.

[0058] The environmental monitoring data carries a timestamp. Based on the content of the environmental monitoring data, the environmental monitoring data can directly reflect the current concentration data of each pollutant in each environmental medium corresponding to the timestamp at the current moment.

[0059] In the method of this step, the determination of the industrial pollution risk and the pollution concentration impact of each pollutant in each environmental medium based on the industrial characteristic data includes: identifying the industrial characteristics carried by the industrial characteristic data; determining the industrial pollution sub-risk of each industrial characteristic and the pollution concentration sub-impact of each pollutant in each environmental medium based on a pre-acquired industrial characteristic impact table, the industrial characteristic impact table including the comparison relationship between industrial characteristics and industrial pollution sub-risks and the comparison relationship between industrial characteristics, pollutants and environmental media and pollutant concentration sub-impacts; determining the sum of all industrial pollution sub-risks as the industrial pollution risk, and determining the sum of all pollution concentration sub-impacts of pollutants in environmental media as the pollution concentration impact.

[0060] Specifically, when the industrial characteristics contained in the industrial characteristic data are clear and the industrial characteristic impact table is clear, it can be substituted into the industrial characteristic impact table to determine the industrial pollution sub-risk, and combined with the pollutant concentration impact of each pollutant in each environmental medium, it can be substituted into the industrial characteristic impact table to determine the pollution concentration sub-impact for each pollutant in each environmental medium. The industrial pollution sub-risk and pollutant concentration sub-impact obtained here are both quantitative values. In this way, the sum of the industrial pollution sub-risks can be calculated as a quantitative industrial pollution risk, and the pollution concentration impact can also be determined for each pollutant in each environmental medium based on all the pollution concentration sub-impacts determined based on the industrial characteristic data. The pollution concentration impact is also determined independently for each pollutant in each environmental medium.

[0061] In the method of this step, the determination of the associated pollution sub-risk of each pollutant in each environmental medium based on the current concentration data, the pollution concentration impact and the pollutant concentration threshold preset for each pollutant for each environmental medium includes: assuming that the current concentration data of the i-th pollutant in the j-th environmental medium is C dij , the pollutant concentration threshold is C pij , the pollution concentration is v ij , the associated pollution sub-risk is Z ii ,but Among them, the pollutant concentration threshold is a preset constant.

[0062] In the method of this step, the combination of the industrial pollution risk and the associated pollution sub-risk to determine the transient pollution risk includes: assuming that the associated pollution sub-risk of the i-th pollutant in the j-th environmental medium is Z ij , industrial pollution risk Z g , the transient contamination risk is Z s ,but

[0063] Z s =1-e -x

[0064]

[0065] In the formula, w g is the preset industrial pollution weight, w ij is the preset associated pollution weight of the i-th pollutant in the j-th environmental medium, A is a preset constant and A≥max(w g Z g , w ij Z ij ).

[0066] S140: Analyze long-term environmental monitoring data, industrial characteristic data, and transient pollution risks to obtain environmental change data, characteristic change data, and risk change data.

[0067] In the method of this step, the environmental change data is the change amount of the environmental monitoring data of adjacent time stamps, which is generally the quantized value of the multi-dimensional environmental monitoring data in the later one-to-one correspondence minus the quantized value of the previous environmental monitoring data.

[0068] The feature change data reflects the changes in the industrial feature data of adjacent timestamps. This process requires the industrial feature data to be quantified and encoded first. For example, the type of production facility is assigned different values ​​according to its potential pollution emission intensity and other factors (such as high pollution facilities are coded as 3, medium pollution as 2, and low pollution as 1). Let the industrial feature data matrix be I, and the dimension be n t ×n i (n i is the number of industrial features), the encoded data can be used to calculate the feature change data. The algorithm of analogy environment change data can be calculated by subtracting the adjacent timestamps from the corresponding values ​​of the dimensions.

[0069] The risk change data reflects the change in the transient pollution risk of adjacent timestamps. The transient pollution risk after the adjacent timestamp is subtracted from the transient pollution risk before the adjacent timestamp.

[0070] S150: Fusing the environmental monitoring data and the industrial characteristic data to obtain fused characteristic data.

[0071] In one example, the method of this step includes: constructing the environmental monitoring data and the industrial feature data into high-dimensional vectors to obtain an environmental monitoring vector and an industrial feature vector; fusing the environmental monitoring vector with the industrial feature vector to obtain the fused feature data, wherein the dimension of the fused feature data is not higher than the maximum value of the dimensions of the environmental monitoring vector and the industrial feature vector.

[0072] For the fusion of environmental monitoring vectors and industrial feature vectors, the filling method (filling elements of the lower dimensional vector with 0 to the higher dimension, and then calculating the vector sum of the two), the repetition method (repeating the elements in the lower dimensional vector until it is extended to a higher dimension) or the projection method can be used. The specific description of the projection method includes: assuming that the dimension of vector A is m and the dimension of vector B is n (m>n). First, perform principal component analysis (PCA) on vector A, extract the first n principal components, and obtain a new vector A′, which has the same dimension as vector B. Assume that A is in matrix form X (m×p, p is the number of samples), and its covariance matrix is Calculate the eigenvalues ​​and eigenvectors of C, sort them by eigenvalue, and select the first n eigenvectors to form the projection matrix P(m×n), then A′=XP. Let the weight be w A and w B , the element calculation formula of the fused vector C is: i =w A ×a i +w B ×b i , where i = 1, 2, ..., n. For example, if w A =0.6, w B =0.4, then c1=0.6×a1+0.4×b1, c2=0.6×a2+0.4×b2, etc.

[0073] Based on the method of this step, the environmental monitoring data and the industrial characteristic data can be fused to obtain fused characteristic data, and the fused characteristic data is in the form of a multi-dimensional vector.

[0074] S160: Input the environmental monitoring data, industrial characteristic data, transient pollution risk, environmental change data, characteristic change data, risk change data and fused characteristic data into a pre-acquired risk analysis model to obtain pollution risk data for industrial agglomeration areas.

[0075] In the method of this step, the risk analysis model includes a long short-term memory network sub-model, a time series feature analysis sub-model and a neural network sub-model.

[0076] Among them, the long short-term memory network sub-model is used to analyze long-term environmental monitoring data, industrial characteristic data, transient pollution risks, environmental change data, characteristic change data and risk change data respectively, and obtain environmental monitoring time series characteristics, industrial characteristic time series characteristics, pollution risk time series characteristics, environmental change time series characteristics, characteristic change time series characteristics and risk change time series characteristics respectively.

[0077] The long short-term memory network sub-model is LSTM, which can be used to input various data with timestamps and output time series feature results. The time series feature results can be represented as the predicted value of the next timestamp, or the predicted change speed, etc. The principle of the long short-term memory network sub-model is introduced below.

[0078] LSTM specifically includes an input gate, a forget gate, an output gate, and candidate memory units. The following is the calculation principle of the input gate, the forget gate, the output gate, and the candidate memory unit (taking timestamp t as an example).

[0079] Input Gate:

[0080] i t =σ(W xi x t +W hi h t-1 +b i )

[0081] Where W xi is the input data x t The weight matrix to the input gate, dimension is n h ×n x (n h is the number of LSTM hidden units), W hi is the previous hidden state h t-1 The weight matrix to the input gate, dimension is n h ×n h , b i is the bias vector of the input gate, dimension is n h , σ is the sigmoid function.

[0082] Forget Gate:

[0083] f t =σ(W xf x t +W hf h t-1 +b f )

[0084] Where W xf , W hf and b f Similar to the corresponding parameters in the input gate, they are the weight matrix and bias vector related to the forget gate.

[0085] Output Gate:

[0086] o t =σ(W xo x t +W ho h t-1 +b o )

[0087] Where W xo , W ho and b o Parameters related to the output gate.

[0088] Candidate memory cells:

[0089]

[0090] Where W xc is the input data x t To the weight matrix of the candidate memory unit, dimension is n h ×n x , W hc is the previous hidden state H t-1 To the weight matrix of the candidate memory unit, dimension is n H ×n H ,b c is the bias vector of the candidate memory unit, with dimension n h ,tanh is the hyperbolic tangent function.

[0091] Memory cell update and hidden state update

[0092] Memory unit update:

[0093]

[0094] in represents element-wise multiplication, c t-1 is the previous memory cell state.

[0095] Hide status updates:

[0096]

[0097] Of course, the above content is only a brief introduction to the LSTM principle. The basic means is to input time series data (whether single-dimensional or multi-dimensional data) into LSTM and extract the time series features (such as predicted values, time series change trend values, etc.). This application does not improve the internal principles of the LSTM model, so it will not be elaborated.

[0098] In the method of this step, the time series feature analysis submodel is used to analyze the time series feature significance of environmental monitoring time series features, industrial feature time series features, pollution risk time series features, environmental change time series features, feature change time series features and risk change time series features, and divides the environmental monitoring data, industrial feature data, transient pollution risks, environmental change data, feature change data and risk change data into time series dependent data and non-time series dependent data based on the time series feature significance.

[0099] The time series feature analysis sub-model is an improved model of the embodiment of the present application, and the time series feature analysis sub-model includes: according to the ratio of the vector modulus value of the time series feature to the square of the vector dimension number as the time series feature significance of the corresponding time series feature, determining that the time series feature with higher significance than at least three other time series features or higher than a preset feature significance threshold is an obvious time series feature, and determining that the data with obvious time series features in the environmental monitoring data, industrial feature data, transient pollution risks, environmental change data, feature change data and risk change data is time series dependent data, and the others are non-time series dependent data.

[0100] Specifically, the time series features include environmental monitoring time series features, industrial feature time series features, pollution risk time series features, environmental change time series features, feature change time series features and risk change time series features, which correspond to six feature data: environmental monitoring data, industrial feature data, transient pollution risk, environmental change data, feature change data and risk change data, respectively. Each time series feature is a multidimensional vector. The ratio of the vector modulus value of each vector to the square of the vector dimension is calculated to obtain the time series feature significance. The time series feature significance is a quantitative value. The time series features are sorted from large to small according to the time series feature significance. The first three are taken as obvious time series features, and the last three that are higher than the preset feature significance threshold are determined to be obvious time series features. In this way, the six feature data can be divided into two categories according to the time series features: time series dependent data and non-time series dependent data. That is, the feature data with obvious time series features are time series dependent data, and the others are non-time series dependent data.

[0101] In the method of this step, the neural network sub-model is used to input the time series features of the time series dependent data and the fused feature data and the non-time series dependent data to output the pollution risk data. The specific principle here is that first, the fused feature data is input, and for the six feature data, the non-time series dependent data itself is input, and the time series features of the time series dependent data itself are not input, and the final output result is the quantified pollution risk data.

[0102] In an embodiment of the present application, the neural network sub-model includes an input layer, multiple hidden layers and an output layer, wherein the input layer includes multiple input nodes, which are used to input timing features of timing-dependent data and fusion feature data and non-timing-dependent data respectively, and the output layer includes an output node for outputting quantified pollution risk data.

[0103] Regarding the specific calculation principle of the neural network sub-model, no improvement is made in the embodiments of the present application, so only the following exemplary introduction is made. The following is a detailed construction and calculation formula of a neural network with multiple input nodes, one output node, multiple hidden layers, and each hidden layer contains multiple neurons.

[0104] Input layer: Assume that the input layer has n nodes and the input vector is represented as x=[x1,x2,…,x n ].

[0105] Hidden layer: Assume there are L hidden layers, and the lth hidden layer has m l neurons, where l = 1, 2, …, L.

[0106] Output layer: has 1 output node.

[0107] Input layer to the first hidden layer: Let the weight matrix from the input layer to the first hidden layer be W1, with a dimension of m1×n, and the bias vector be b1, with a dimension of m1. For the input vector x, the input z1 of the first hidden layer is calculated as:

[0108] z1=W1x+b1

[0109] The activation function of the first hidden layer uses the ReLU function (other activation functions, such as Tanh or Sigmoid, can also be used) to obtain the output a1 of the first hidden layer:

[0110] a1=Re LU(z1)=max(0,z1)

[0111] Hidden layer to hidden layer: For the lth hidden layer (l = 2, 3, ..., L), let the weight matrix from the l-1th hidden layer to the lth hidden layer be W l , dimension is m l ×m l-1 , the bias vector is b l , dimension is m l The input z of the lth hidden layer is l Calculated as:

[0112] z l =W l a l-1 +b l

[0113] The output of the lth hidden layer is a l Processed by activation function:

[0114] a l =ReLU(z l )=max(0,z l )

[0115] From the last hidden layer to the output layer: Let the weight matrix from the last hidden layer (the Lth hidden layer) to the output layer be W L+1 , with dimensions of 1×m L , the bias vector is b L+1 , with a dimension of 1. The input z of the output layer L+1 Calculated as:

[0116] z L+1 =W L+1 a L +b L+1

[0117] The output y of the output layer uses the Sigmoid function (if it is a regression task, you can not use the activation function, if it is a multi-classification task, you can use the Softmax function):

[0118]

[0119] Regarding the loss function, for the binary classification task: use the cross entropy loss function. Let y be the predicted output of the network, y true is the true label (0 or 1), and the loss function L is:

[0120] L=-[y true log(y)+(1-y true )log(1-y)]

[0121] The regression task is: use the mean square error loss function. Let y be the predicted output of the network, y true is the true value, and the loss function L is:

[0122]

[0123] Regarding back propagation, the gradient calculation of the output layer includes using the binary classification task: according to the chain rule, the cross entropy loss function L with respect to z L+1 Derivative:

[0124]

[0125] Using regression tasks: the mean square error loss function L about z L+1 Derivative:

[0126]

[0127] And to W L+1 and b L+1 Gradient calculation of:

[0128]

[0129] Gradient calculation of hidden layer (from back to front): For the Lth hidden layer, calculate the gradient of a L The gradient is:

[0130]

[0131] Right L The gradient ofL =ReLU(z L ), when z L >0, When z L When ≤0,

[0132] so:

[0133]

[0134] To W L and b L The gradient calculation of

[0135]

[0136] For the lth hidden layer (l=L-1,L-2,…,1), and so on, calculate the pair a l The gradient is:

[0137]

[0138] Right l The gradient is:

[0139] To W l and b l Gradient calculation of :

[0140]

[0141] Regarding parameter update, we can use the gradient descent algorithm to update the parameters. Let the learning rate be α and update W. L+1 :

[0142]

[0143] Update b L+1 :

[0144]

[0145] Update W l (l=1,2,…,L):

[0146]

[0147] Update b l (l=1,2,…,L):

[0148]

[0149] Based on the above content, a neural network sub-model can be constructed to determine the pollution risk data of industrial agglomeration areas based on all input data.

[0150] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0151] In a second aspect, the present application discloses a long-term intelligent analysis system for pollution risk in industrial agglomeration areas, wherein the system applies any method disclosed in the first aspect above.

[0152] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0153] In summary, this application at least has the following beneficial effects:

[0154] A method and system for intelligent analysis of pollution risks in industrial agglomeration areas with a long time series is provided, which can gradually and intelligently analyze long-time environmental monitoring data and remote sensing image data, thereby realizing dynamic and accurate monitoring of pollution risks in industrial agglomeration areas.

[0155] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.

Claims

1. A long-term intelligent analysis method for pollution risk in industrial agglomeration areas, characterized in that: include: Obtain long-term environmental monitoring data and remote sensing image data for each divided area within the industrial agglomeration area; Analyzing the remote sensing image data to obtain industrial characteristic data; Determine the transient pollution risk of each divided area in a long time series according to the environmental monitoring data and the industrial characteristic data; Analyze long-term environmental monitoring data, industrial characteristic data and transient pollution risks to obtain environmental change data, characteristic change data and risk change data; fusing the environmental monitoring data and the industrial characteristic data to obtain fused characteristic data; The environmental monitoring data, industrial characteristic data, transient pollution risk, environmental change data, characteristic change data, risk change data and fused characteristic data are input into a pre-acquired risk analysis model to obtain pollution risk data for industrial agglomeration areas.

2. The method according to claim 1, characterized in that Determining the transient pollution risk of each divided area in a long time series according to the environmental monitoring data and the industrial characteristic data includes: Determine the current concentration data of each pollutant in each environmental medium based on the environmental monitoring data; Determining industrial pollution risks and the pollution concentration impact of each pollutant in each environmental medium based on the industrial characteristic data; Determine the associated pollution sub-risk of each pollutant in each environmental medium based on the current concentration data, the pollution concentration impact and the pollutant concentration threshold preset for each pollutant for each environmental medium; The transient pollution risk is determined by combining the industrial pollution risk and the associated pollution sub-risks.

3. The method according to claim 2, characterized in that Determining the industrial pollution risk and the pollution concentration impact of each pollutant in each environmental medium based on the industrial characteristic data includes: Identifying the industrial characteristics carried by the industrial characteristic data; Determine the industrial pollution sub-risk of each industrial characteristic and the pollution concentration sub-impact of each pollutant in each environmental medium based on the pre-acquired industrial characteristic impact table, wherein the industrial characteristic impact table includes the comparison relationship between industrial characteristics and industrial pollution sub-risks and the comparison relationship between industrial characteristics, pollutants and environmental media and pollutant concentration sub-impacts; The sum of all industrial pollution sub-risks is determined as the industrial pollution risk, and the sum of all pollution concentration sub-impacts facing pollutants in environmental media is determined as the pollution concentration impact.

4. The method according to claim 3, characterized in that Determining the associated pollution sub-risk of each pollutant in each environmental medium based on the current concentration data, the pollution concentration impact and the pollutant concentration threshold preset for each pollutant for each environmental medium includes: Suppose the current concentration data of the i-th pollutant in the j-th environmental medium is C dij , the pollutant concentration threshold is C pij , the pollution concentration is v ij , the associated pollution sub-risk is Z ij ,but 5. The method according to claim 2, characterized in that: The determining the transient pollution risk by combining the industrial pollution risk and the associated pollution sub-risks comprises: Assume that the associated pollution risk of the i-th pollutant in the j-th environmental medium is Z ij , industrial pollution risk Z g , the transient contamination risk is Z s ,but With s =1-e -x In the formula, w g is the preset industrial pollution weight, w ij is the preset associated pollution weight of the i-th pollutant in the j-th environmental medium, A is a preset constant and A≥max(w g Z g ,w ij Z ij ).

6. The method according to claim 1, characterized in that The fusing of the environmental monitoring data and the industrial characteristic data to obtain the fused characteristic data includes: The environmental monitoring data and the industrial characteristic data are both constructed into high-dimensional vectors to obtain an environmental monitoring vector and an industrial characteristic vector; The environmental monitoring vector and the industrial feature vector are fused to obtain the fused feature data, wherein the dimension of the fused feature data is not higher than the maximum value of the dimensions of the environmental monitoring vector and the industrial feature vector.

7. The method according to claim 1, characterized in that The risk analysis model includes a long short-term memory network sub-model, a time series feature analysis sub-model and a neural network sub-model; The long short-term memory network sub-model is used to analyze long-term environmental monitoring data, industrial characteristic data, transient pollution risk, environmental change data, characteristic change data and risk change data, respectively, to obtain environmental monitoring time series characteristics, industrial characteristic time series characteristics, pollution risk time series characteristics, environmental change time series characteristics, characteristic change time series characteristics and risk change time series characteristics; The time series feature analysis sub-model is used to analyze the time series feature prominence of environmental monitoring time series features, industrial feature time series features, pollution risk time series features, environmental change time series features, feature change time series features and risk change time series features, and divide the environmental monitoring data, industrial feature data, transient pollution risk, environmental change data, feature change data and risk change data into time series dependent data and non-time series dependent data based on the time series feature prominence; The neural network sub-model is used to input the time series features of the time series dependent data and fuse the feature data and the non-time series dependent data to output the pollution risk data.

8. The method according to claim 7, characterized in that The timing feature analysis sub-model includes: According to the ratio of the vector modulus of the time series feature to the square of the vector dimension number, the time series feature significance of the corresponding time series feature is determined as the time series feature significance. The time series feature with higher significance than at least three other time series features or higher than a preset feature significance threshold is determined as the obvious time series feature. The data with obvious time series features in the environmental monitoring data, industrial characteristic data, transient pollution risks, environmental change data, characteristic change data and risk change data are determined as time series dependent data, and the others are non-time series dependent data.

9. A long-term intelligent analysis system for pollution risk in industrial agglomeration areas, characterized in that: Application of the method as claimed in any one of claims 1 to 8.

Images