A chemical circulation risk supervision system in chemical parks based on industrial Internet

Abstract

The present invention discloses a chemical circulation risk supervision system for chemical parks based on the industrial Internet, including a data acquisition module, a chemical circulation risk assessment model establishment module and a chemical circulation risk supervision module for chemical parks. The present invention belongs to the field of data processing, and specifically refers to a chemical circulation risk supervision system for chemical parks based on the industrial Internet. This scheme designs an adaptive SE block based on a risk intensity adjustment item, dynamically adapts to different risk environments, introduces a risk factor bias item to strengthen key time dynamics, introduces a trajectory anomaly point factor, and improves the accuracy of chemical circulation risk supervision; constructs a risk point positioning loss item based on dynamic weight adjustment; focuses on optimizing high-offset boundary points through a weighted combination of offset distances; prioritizes optimizing the overlap of high-risk areas based on the dynamic weights of intersections and unions, and finally obtains a loss function; thereby improving the effect of chemical circulation risk supervision.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and specifically refers to a chemical park chemical circulation risk supervision system based on industrial Internet. Background Art

[0002] The chemical circulation risk supervision system is a comprehensive management system based on data collection, artificial intelligence and risk assessment. It is used to monitor, analyze and predict potential risks in chemical parks or chemical circulation processes in real time, so as to ensure the safety of chemicals in storage, transportation and use. However, the general chemical circulation risk supervision system lacks dynamic environmental adaptability and cannot extract the deep correlation between complex chemical circulation data, which leads to poor accurate perception of chemical status and risks and low supervision accuracy. The general chemical circulation risk supervision system has improper consideration of risk point positioning, regional range matching and data range differences, and ignores multi-level feature matching, which leads to poor circulation risk supervision effect. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a chemical park chemical circulation risk supervision system based on the industrial Internet. The general chemical circulation risk supervision system lacks dynamic environmental adaptability and cannot extract the deep correlation between complex chemical circulation data, which leads to poor accurate perception of chemical status and risks and low supervision accuracy. This scheme designs an adaptive SE block based on the risk intensity adjustment item, dynamically adapts to different risk environments, improves the ability to identify high-risk areas, introduces risk factor bias items to strengthen key time dynamics, introduces trajectory anomaly factors, strengthens the feature extraction of logistics paths and tank locations, and thus improves adaptability to complex circulation environments. The response capacity is improved to improve the accuracy of chemical circulation risk supervision; the general chemical circulation risk supervision system has improper consideration of risk point positioning, regional range matching and data range differences, and ignores multi-level feature matching, which leads to poor circulation risk supervision effect. This solution is based on dynamically adjusting the error focus points according to the horizontal and vertical weights, adapting to the layout of chemical parks, and constructing risk point positioning loss items based on dynamic weight adjustment; through the weighted combination of offset distance and gas concentration, focus on optimizing high-offset boundary points to reduce large deviations in range prediction; according to the dynamic weights of intersection and union, prioritize optimizing the overlap of high-risk areas, and finally obtain the loss function; thereby improving the effect of chemical circulation risk supervision.

[0004] The technical solution adopted by the present invention is as follows: A chemical park chemical circulation risk supervision system based on the industrial Internet provided by the present invention includes a data acquisition module, a chemical circulation risk assessment model establishment module and a chemical park chemical circulation risk supervision module;

[0005] The data collection module collects historical chemical park chemical circulation data based on the industrial Internet;

[0006] The chemical circulation risk assessment model establishment module realizes the establishment of the chemical circulation risk assessment model through adaptive SE blocks, time and channel attention mechanisms, coordinate attention mechanisms, normalized attention mechanisms and similarity attention mechanisms, combined with dynamic weight adjustment and loss optimization;

[0007] The chemical park chemical circulation risk supervision performs circulation risk supervision on real-time chemical park chemical circulation data based on an established chemical circulation risk assessment model.

[0008] Furthermore, in the data acquisition module, the historical chemical park chemical circulation data includes temperature sensor data, gas concentration sensor data, pressure sensor data, flow sensor data, sampling time, logistics data, transportation equipment data and circulation risk assessment results; the circulation risk assessment results are used as data labels; the collected data are converted and standardized; the circulation risk assessment results include normal, low risk, medium risk and high risk; when collecting historical chemical park chemical circulation data, the sampling frequency and sampling range of the sensor are dynamically adjusted according to environmental changes, which is expressed as: ; ; Multi-sensor data fusion is performed, expressed as: ; and perform data enhancement; enhance the training data by simulating chemical leakage and equipment failure scenarios to make the model more adaptable to different risk environments; the generation of enhanced data is expressed as: ;in, and are the sampling frequency and sampling range of the sensor respectively; , and are the readings of the temperature sensor, gas sensor, and pressure sensor at time t; , and is the adjustment factor of the sampling frequency; , and is the adjustment factor of the sampling range; is the fused sensor data value; is the sensor weight; is the raw reading of the sensor; nc is the sensor number, ic is the sensor index; is the enhanced sensor data; SF is the simulation scenario factor, which indicates the simulation intensity of faults and leakage; Ls is the simulation scenario duration.

[0009] Furthermore, the chemical circulation risk assessment model establishment module specifically includes the following contents:

[0010] Model building unit; Based on the deep learning model, the established chemical circulation risk assessment model includes an input layer, an attention mechanism integration layer and an output layer; the input layer receives the data set processed by the data acquisition module; the attention mechanism integration layer is used to enhance the feature extraction capability and adapt to the complex circulation environment of dense tank areas and hazardous chemical gathering areas; including: the global feature extraction stage, which is used to extract global key features from the historical chemical park chemical circulation data and dynamically adapt to environmental changes; the local feature enhancement stage, through the joint modeling of time and space dimensions, further extracts the local risk features in the circulation path and equipment monitoring; the feature aggregation stage, combining global and local features, analyzes the similarities between the historical chemical park chemical circulation data, and extracts the subtle differences in chemical status and risks; in the global feature extraction stage, adaptive SE blocks and normalized attention mechanisms are used; in the local feature enhancement stage, time and channel attention and coordinate attention mechanisms are used; in the feature aggregation stage, similarity attention mechanisms are used; the output layer generates the circulation risk assessment results corresponding to the data;

[0011] The attention mechanism integration layer building unit; specifically includes:

[0012] Adaptive SE block design; Based on the adaptive SE block, the focus of the model is automatically adjusted according to the changes in the environment; the adaptive SE block is expressed as: ; Risk intensity adjustment It is expressed as: ;in, is the adaptive attention weight; X is the input; and is the adaptive SE block weight matrix; is the ReLU activation function; , and is the weighting factor of risk intensity; , and They are the risk intensity of temperature, gas and pressure; is the Sigmoid activation function;

[0013] Aiming at the dynamic changes in multivariate data of chemical parks, time and channel attention are combined; risk factor bias term is introduced, and the time attention weight calculation is expressed as: ; ; The channel attention weight calculation is expressed as: ; Output features It is expressed as: ;in, and are the attention weights of the time dimension and channel dimension respectively; is the risk factor bias term; It is a one-dimensional convolution operation; and They are average pooling and maximum pooling respectively; and is the risk weight factor; It is a multi-layer perceptron; It is a data normalization operation; is the input feature matrix of time and channel attention;

[0014] The coordinate attention mechanism is introduced; the trajectory abnormal point factor is introduced, and the coordinate information extraction is expressed as: ; ; ; ; The weight calculation is expressed as: ; ; The output features are expressed as: ;in, and are the feature values ​​extracted in the spatial dimension and the temporal dimension respectively; is the data at input matrix position i3 and time t; and are the trajectory anomaly factors in the spatial dimension and the temporal dimension respectively; is the trajectory deviation of position x; is the proportion of retention time exceeding the expected value; and is the spatial anomaly weight; , and is the temporal anomaly weight; , and are the rate of change of temperature, gas concentration and pressure at time t respectively; and are the attention weights of the spatial dimension and the temporal dimension respectively; and is the spatial dimension weight; and is the time dimension weight; and XZ are the output feature matrix and input feature matrix of the coordinate attention mechanism respectively; N3 is the total number of positions; T is the total number of times;

[0015] Introduce the normalized attention mechanism; the normalization operation is expressed as: ; The weight calculation is expressed as: ; The output features are expressed as: ; is the normalized eigenvalue; and are the mean and standard deviation of the input data respectively; is a smoothing term; is the normalized channel attention weight; is the normalized attention weight matrix; and They are the output features and input features of the normalized attention mechanism; is the variance; is the normalized weight;

[0016] Introduce the similarity attention mechanism; the similarity calculation is expressed as: ; The output features are expressed as: ;in, is the similarity attention weight; d is the dimension index; and They are the output features and input features of the dth dimension of the similarity attention mechanism; is the feature mean; and is the similarity weight; Corr(·) is the correlation; Max(·) is the maximum value;

[0017] Loss function design unit; specifically includes:

[0018] Construct risk point positioning loss items , quantify the error between the predicted risk point and the actual risk point, and use it to mark the specific high-risk location, expressed as: ; ; ;in, and It is the horizontal and vertical coordinates of the predicted risk point; and It is the horizontal and vertical coordinates of the real risk point; is the positioning loss weight; is the dynamic weight of the importance of the risk point; d1 is the direction identifier; and is the influence weight in the horizontal and vertical coordinate directions; is the horizontal axis direction, used to mark the error of the optimized horizontal position; is the ordinate direction, used to mark the error of the optimized height position;

[0019] Introducing data range deviation loss term , quantifies the difference between the predicted data range and the true data range, and is used to evaluate the prediction accuracy of the high-risk range; it is expressed as:

[0020] ; ;in, is the offset distance between the predicted boundary point and the actual boundary point; is the normalization factor of the risk area; n is the number of boundary points in the risk area; i is the boundary point index; and is the horizontal and vertical coordinates of the i-th predicted boundary point; and is the horizontal and vertical coordinates of the ith real boundary point; max(·) is the maximum value;

[0021] Constructing overlapping loss items in insurance coverage , by calculating the overlap of the risk area, the matching degree between the predicted range and the actual range is measured, which is expressed as: ; ; ;in, and They are the predicted risk range and the actual risk range; and are the dynamic weights of the intersection and union, respectively; is the number of monitoring points; i2 is the monitoring point index;

[0022] Final loss function It is expressed as: ; ;in, is the dynamic weighting coefficient; is the minimum enclosing rectangle area of ​​the predicted range and the true range;

[0023] Model judgment unit; update model parameters based on gradient descent algorithm; divide the data set processed by the data acquisition module into test set and training set in advance, and set the accuracy threshold; train the chemical circulation risk assessment model based on the training set; when the prediction accuracy of the chemical circulation risk assessment model for the test set is higher than the accuracy threshold, the chemical circulation risk assessment model is established.

[0024] Furthermore, the chemical circulation risk assessment model establishment module is based on the established chemical circulation risk assessment model, collects chemical park chemical circulation data in real time, and inputs it into the chemical circulation risk assessment model after preprocessing. When the circulation risk assessment result output by the model is medium risk, the sampling frequency is increased, and when the output is high risk, early warning processing is issued to management personnel.

[0025] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0026] (1) In view of the fact that the general chemical circulation risk supervision system lacks dynamic environmental adaptability and is unable to extract the deep correlation between complex chemical circulation data, which leads to poor accurate perception of chemical status and risks and low supervision accuracy, this scheme designs an adaptive SE block based on risk intensity adjustment items to dynamically adapt to different risk environments and improve the ability to identify high-risk areas. It introduces risk factor bias items to strengthen the key time dynamics and introduces trajectory anomaly factors to strengthen the feature extraction of logistics paths and tank locations, thereby improving the adaptability to complex circulation environments and improving the accuracy of chemical circulation risk supervision.

[0027] (2) In view of the problems that the general chemical circulation risk supervision system has inappropriately considered the risk point positioning, regional range matching and data range differences, and ignored the multi-level feature matching, which led to poor circulation risk supervision effects. This scheme is based on dynamically adjusting the error focus points according to the horizontal and vertical weights, adapting to the layout of the chemical park, and constructing the risk point positioning loss item based on dynamic weight adjustment; through the weighted combination of offset distance and gas concentration, the boundary points with high offset are optimized to reduce the large deviation in range prediction; according to the dynamic weights of intersection and union, the overlap of high-risk areas is optimized first, and finally the loss function is obtained; thereby improving the effect of chemical circulation risk supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic diagram of a process flow of a chemical park chemical circulation risk supervision system based on the industrial Internet provided by the present invention;

[0029] Figure 2 Schematic diagram of the process for building modules for the chemical distribution risk assessment model.

[0030] 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

[0031] 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.

[0032] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0033] Example 1, see Figure 1 , the present invention provides a chemical park chemical circulation risk supervision system based on industrial Internet, including a data acquisition module, a chemical circulation risk assessment model establishment module and a chemical park chemical circulation risk supervision module;

[0034] The data collection module collects historical chemical park chemical circulation data based on the industrial Internet; and sends the data to the chemical circulation risk assessment model establishment module;

[0035] The chemical circulation risk assessment model establishment module realizes the establishment of the chemical circulation risk assessment model through adaptive SE blocks, time and channel attention mechanisms, coordinate attention mechanisms, normalized attention mechanisms and similarity attention mechanisms, combined with dynamic weight adjustment and loss optimization; and sends the data to the chemical circulation risk supervision module of the chemical park;

[0036] The chemical park chemical circulation risk supervision performs circulation risk supervision on real-time chemical park chemical circulation data based on an established chemical circulation risk assessment model.

[0037] Example 2, see Figure 1 , this embodiment is based on the above embodiment. In the data acquisition module, the historical chemical park chemical circulation data includes temperature sensor data, gas concentration sensor data, pressure sensor data, flow sensor data, sampling time, logistics data, transportation equipment data and circulation risk assessment results; the circulation risk assessment results are used as data labels; the collected data are converted and standardized; the circulation risk assessment results include normal, low risk, medium risk and high risk; the temperature sensor data includes ambient temperature and medium temperature; the gas concentration sensor data includes volatile organic compound concentration, toxic gas concentration, combustible gas concentration and concentration gradient of gas leakage points; the pressure sensor data includes pressure in the storage tank, fluid pressure in the pipeline and pressure fluctuation of key equipment; the flow sensor data includes flow monitoring and flow abnormal points in chemical circulation; the logistics data includes transportation trajectory data, residence time and deviation; when collecting historical chemical park chemical circulation data, the sampling frequency and sampling range of the sensor are dynamically adjusted according to environmental changes, which is expressed as: ; ; Multi-sensor data fusion is performed, expressed as: ; and perform data enhancement; enhance the training data by simulating chemical leakage and equipment failure scenarios to make the model more adaptable to different risk environments; the generation of enhanced data is expressed as: ;in, and are the sampling frequency and sampling range of the sensor respectively; , and are the readings of the temperature sensor, gas sensor, and pressure sensor at time t; , and is the adjustment factor of the sampling frequency; , and is the adjustment factor of the sampling range; is the fused sensor data value; is the sensor weight; is the raw reading of the sensor; nc is the sensor number, ic is the sensor index; is the enhanced sensor data; SF is the simulation scenario factor, which indicates the simulation intensity of faults and leakage; Ls is the simulation scenario duration.

[0038] Example 3, see Figure 1 and Figure 2 Based on the above embodiment, this embodiment includes the following contents:

[0039] Model building unit; Based on the deep learning model, the established chemical circulation risk assessment model includes an input layer, an attention mechanism integration layer and an output layer; the input layer receives the data set processed by the data acquisition module; the attention mechanism integration layer is used to enhance the feature extraction capability and adapt to the complex circulation environment of dense tank areas and hazardous chemical gathering areas; including: the global feature extraction stage, which is used to extract global key features from the historical chemical park chemical circulation data and dynamically adapt to environmental changes; the local feature enhancement stage, through the joint modeling of time and space dimensions, further extracts the local risk features in the circulation path and equipment monitoring; the feature aggregation stage, combining global and local features, analyzes the similarities between the historical chemical park chemical circulation data, and extracts the subtle differences in chemical status and risks; in the global feature extraction stage, adaptive SE blocks and normalized attention mechanisms are used; in the local feature enhancement stage, time and channel attention and coordinate attention mechanisms are used; in the feature aggregation stage, similarity attention mechanisms are used; the output layer generates the circulation risk assessment results corresponding to the data;

[0040] The attention mechanism integration layer building unit; specifically includes:

[0041] Adaptive SE block design; Based on the adaptive SE block, the focus of the model is automatically adjusted according to the changes in the environment. If the gas leak is the current high-risk area, the weight of the gas sensor data is adaptively increased; the adaptive SE block is expressed as: ; Risk intensity adjustment It is expressed as: ;in, is the adaptive attention weight; X is the input; and is the adaptive SE block weight matrix; is the ReLU activation function; , and is the weighting factor of risk intensity; , and They are the risk intensity of temperature, gas and pressure; is the Sigmoid activation function;

[0042] Aiming at the dynamic changes in multivariate data of chemical parks, time and channel attention are combined; risk factor bias term is introduced, and the time attention weight calculation is expressed as: ; ; The channel attention weight calculation is expressed as: ; Output features It is expressed as: ;in, and are the attention weights of the time dimension and channel dimension respectively; is the risk factor bias term; It is a one-dimensional convolution operation; and They are average pooling and maximum pooling respectively; and is the risk weight factor; It is a multi-layer perceptron; It is a data normalization operation; is the input feature matrix of time and channel attention;

[0043] The coordinate attention mechanism is introduced to enhance the feature extraction of location information in the logistics path, especially the detection of outliers in the transportation trajectory. The trajectory outlier factor is introduced, and the coordinate information extraction is expressed as: ; ; ; ; The weight calculation is expressed as: ; ; The output features are expressed as: ;in, and are the feature values ​​extracted in the spatial dimension and the temporal dimension respectively; is the data at input matrix position i3 and time t; and are the trajectory anomaly factors in the spatial dimension and the temporal dimension respectively; is the trajectory deviation of position x; is the proportion of retention time exceeding the expected value; and is the spatial anomaly weight; , and is the temporal anomaly weight; , and are the rate of change of temperature, gas concentration and pressure at time t respectively; and are the attention weights of the spatial dimension and the temporal dimension respectively; and is the spatial dimension weight; and is the time dimension weight; and XZ are the output feature matrix and input feature matrix of the coordinate attention mechanism respectively; N3 is the total number of positions; T is the total number of times;

[0044] The normalized attention mechanism is introduced to adapt to the dynamics of temperature fluctuations and logistics timeliness of chemical circulation data, and to enhance the detection of outliers through normalization; the normalization operation is expressed as: ; The weight calculation is expressed as: ; The output features are expressed as: ; is the normalized eigenvalue; and are the mean and standard deviation of the input data respectively; is a smoothing term; is the normalized channel attention weight; is the normalized attention weight matrix; and They are the output features and input features of the normalized attention mechanism; is the variance; is the normalized weight;

[0045] The similarity attention mechanism is introduced to analyze the similarity between multi-dimensional features and extract the subtle differences between chemical status and risks; the similarity calculation is expressed as: ; The output features are expressed as: ;in, is the similarity attention weight; d is the dimension index; and They are the output features and input features of the dth dimension of the similarity attention mechanism; is the feature mean; and is the similarity weight; Corr(·) is the correlation; Max(·) is the maximum value;

[0046] Loss function design unit;

[0047] Model judgment unit; update model parameters based on gradient descent algorithm; divide the data set processed by the data acquisition module into test set and training set in advance, and set the accuracy threshold; train the chemical circulation risk assessment model based on the training set; when the prediction accuracy of the chemical circulation risk assessment model for the test set is higher than the accuracy threshold, the chemical circulation risk assessment model is established.

[0048] By performing the above operations, the general chemical circulation risk supervision system lacks dynamic environmental adaptability and cannot extract deep correlations between complex chemical circulation data, which leads to poor accurate perception of chemical status and risks and low supervision accuracy. This solution designs an adaptive SE block based on risk intensity adjustment items to dynamically adapt to different risk environments, improve the ability to identify high-risk areas, introduce risk factor bias items to strengthen key time dynamics, introduce trajectory anomaly factors, and strengthen the feature extraction of logistics paths and tank locations, thereby improving the adaptability to complex circulation environments and improving the accuracy of chemical circulation risk supervision.

[0049] Example 4, see Figure 1 This embodiment is based on the above embodiment. In the chemical circulation risk assessment model establishment module, the loss function design unit specifically includes:

[0050] Construct risk point positioning loss items , quantify the error between the predicted risk point and the actual risk point, and use it to mark the specific high-risk location, expressed as: ; ; ;in, and It is the horizontal and vertical coordinates of the predicted risk point; and It is the horizontal and vertical coordinates of the real risk point; is the positioning loss weight; is the dynamic weight of the importance of the risk point; d1 is the direction identifier; and is the influence weight in the horizontal and vertical coordinate directions; is the horizontal axis direction, used to mark the error of the optimized horizontal position; is the ordinate direction, used to mark the error of the optimized height position;

[0051] Introducing data range deviation loss term , quantifies the difference between the predicted data range and the true data range, and is used to evaluate the prediction accuracy of the high-risk range; it is expressed as:

[0052] ; ;in, is the offset distance between the predicted boundary point and the actual boundary point; is the normalization factor of the risk area; n is the number of boundary points in the risk area; i is the boundary point index; and is the horizontal and vertical coordinates of the i-th predicted boundary point; and is the horizontal and vertical coordinates of the ith real boundary point; max(·) is the maximum value;

[0053] Constructing overlapping loss items in insurance coverage , by calculating the overlap of the risk area, the matching degree between the predicted range and the actual range is measured, which is expressed as: ; ; ;in, and They are the predicted risk range and the actual risk range; and are the dynamic weights of the intersection and union, respectively; is the number of monitoring points; i2 is the monitoring point index;

[0054] Final loss function It is expressed as: ; ;in, is the dynamic weighting coefficient; is the minimum enclosing rectangular area of ​​the predicted range and the true range.

[0055] By performing the above operations, the general chemical circulation risk supervision system has problems such as improper consideration of risk point positioning, regional range matching and data range differences, ignoring multi-level feature matching, and thus leading to poor circulation risk supervision effects. This solution is based on dynamically adjusting the error focus points according to the horizontal and vertical weights, adapting to the layout of chemical parks, and constructing risk point positioning loss items based on dynamic weight adjustments; through the weighted combination of offset distance and gas concentration, focusing on optimizing high-offset boundary points to reduce large deviations in range prediction; according to the dynamic weights of intersection and union, prioritizing the optimization of the overlap of high-risk areas, and finally obtaining the loss function; thereby improving the effect of chemical circulation risk supervision.

[0056] Example 5, see Figure 1 This embodiment is based on the above embodiment. The chemical circulation risk assessment model establishment module is based on the established chemical circulation risk assessment model. The chemical circulation data of the chemical park is collected in real time and input into the chemical circulation risk assessment model after preprocessing. When the circulation risk assessment result output by the model is medium risk, the sampling frequency is increased. When the output is high risk, the management personnel are warned.

[0057] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0058] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0059] The present invention and its embodiments are described above, and such 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 ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A chemical park chemical circulation risk supervision system based on the industrial Internet, characterized by: The system includes a data collection module, a chemical circulation risk assessment model building module, and a chemical park chemical circulation risk supervision module; The data collection module collects historical chemical park chemical circulation data based on the industrial Internet; and sends the data to the chemical circulation risk assessment model establishment module; The chemical circulation risk assessment model establishment module realizes the establishment of the chemical circulation risk assessment model through adaptive SE blocks, time and channel attention mechanisms, coordinate attention mechanisms, normalized attention mechanisms and similarity attention mechanisms, combined with dynamic weight adjustment and loss optimization; and sends the data to the chemical circulation risk supervision module of the chemical park; The chemical park chemical circulation risk supervision is based on the established chemical circulation risk assessment model to conduct circulation risk supervision on the real-time chemical park chemical circulation data; The chemical circulation risk assessment model establishment module includes the following contents: adaptive SE block design; Based on the adaptive SE block, the focus of the model is automatically adjusted according to the changes in the environment. If gas leakage is the current high-risk area, the weight of the gas sensor data is adaptively increased; The adaptive SE block is expressed as: ; Risk intensity adjustment It is expressed as: ;in, is the adaptive attention weight; X is the input; and is the adaptive SE block weight matrix; is the ReLU activation function; , and is the weighting factor of risk intensity; is the Sigmoid activation function; , and are the risk intensities of temperature, gas and pressure respectively.

2. According to claim 1, a chemical park chemical circulation risk supervision system based on industrial Internet is characterized in that: The chemical circulation risk assessment model establishment module specifically includes the following contents: Model building unit; Based on the deep learning model, the established chemical circulation risk assessment model includes an input layer, an attention mechanism integration layer and an output layer; the input layer receives the data set processed by the data acquisition module; The attention mechanism integration layer is used to enhance the feature extraction capability and adapt to the complex circulation environment of dense tank areas and hazardous chemical gathering areas; it includes: the global feature extraction stage, which is used to extract global key features from the historical chemical park chemical circulation data and dynamically adapt to environmental changes; the local feature enhancement stage, through the joint modeling of time and space dimensions, further extracts the local risk features in the logistics path and equipment monitoring; the feature aggregation stage, combining global and local features, analyzes the similarities between the historical chemical park chemical circulation data, and extracts the subtle differences in chemical status and risk; in the global feature extraction stage, the adaptive SE block and normalized attention mechanism are used; in the local feature enhancement stage, the time and channel attention and coordinate attention mechanisms are used; in the feature aggregation stage, the similarity attention mechanism is used; the output layer generates the circulation risk assessment results corresponding to the data; Attention mechanism integration layer building unit; Loss function design unit; Model judgment unit; update model parameters based on gradient descent algorithm; divide the data set processed by the data acquisition module into test set and training set in advance, and set the accuracy threshold; train the chemical circulation risk assessment model based on the training set; when the prediction accuracy of the chemical circulation risk assessment model for the test set is higher than the accuracy threshold, the chemical circulation risk assessment model is established.

3. According to claim 2, a chemical park chemical circulation risk supervision system based on industrial Internet is characterized in that: The attention mechanism integration layer construction unit specifically includes: Adaptive SE block design; Aiming at the dynamic changes in multivariate data of chemical parks, time and channel attention are combined; risk factor bias term is introduced, and the time attention weight calculation is expressed as: ; ; The channel attention weight calculation is expressed as: ; Output features It is expressed as: ;in, and are the attention weights of the time dimension and channel dimension respectively; is the risk factor bias term; It is a one-dimensional convolution operation; and They are average pooling and maximum pooling respectively; and is the risk weight factor; It is a multi-layer perceptron; It is a data normalization operation; is the input feature matrix of time and channel attention; The coordinate attention mechanism is introduced; the trajectory abnormal point factor is introduced, and the coordinate information extraction is expressed as: ; ; ; ; The weight calculation is expressed as: ; ; The output features are expressed as: ;in, and are the feature values ​​extracted in the spatial dimension and the temporal dimension respectively; is the data at input matrix position i3 and time t; and are the trajectory anomaly factors in the spatial dimension and the temporal dimension respectively; is the trajectory deviation of position x; is the proportion of retention time exceeding the expected value; and is the spatial anomaly weight; , and is the temporal anomaly weight; , and are the rate of change of temperature, gas concentration and pressure at time t respectively; and are the attention weights of the spatial dimension and the temporal dimension respectively; and is the spatial dimension weight; and is the time dimension weight; and XZ are the output feature matrix and input feature matrix of the coordinate attention mechanism respectively; N3 is the total number of positions; T is the total number of times; Introduce the normalized attention mechanism; the normalization operation is expressed as: ; The weight calculation is expressed as: ; The output features are expressed as: ; is the normalized eigenvalue; and are the mean and standard deviation of the input data respectively; is a smoothing term; is the normalized channel attention weight; is the normalized attention weight matrix; and They are the output features and input features of the normalized attention mechanism; is the variance; is the normalized weight; Introduce the similarity attention mechanism; the similarity calculation is expressed as: ; The output features are expressed as: ;in, is the similarity attention weight; d is the dimension index; and They are the output features and input features of the dth dimension of the similarity attention mechanism; is the feature mean; and is the similarity weight; Corr(·) is the correlation; Max(·) is the maximum value.

4. According to claim 3, a chemical park chemical circulation risk supervision system based on industrial Internet is characterized in that: In the chemical circulation risk assessment model establishment module, the loss function design unit specifically includes: Construct risk point positioning loss items , quantify the error between the predicted risk point and the actual risk point, and use it to mark the specific high-risk location, expressed as: ; ; ;in, and It is the horizontal and vertical coordinates of the predicted risk point; and It is the horizontal and vertical coordinates of the real risk point; is the positioning loss weight; is the dynamic weight of the importance of the risk point; d1 is the direction identifier; and is the influence weight in the horizontal and vertical coordinate directions; is the horizontal axis direction, used to mark the error of the optimized horizontal position; is the ordinate direction, used to mark the error of the optimized height position; Introducing data range deviation loss term , quantifies the difference between the predicted data range and the true data range, and is used to evaluate the prediction accuracy of the high-risk range; it is expressed as: ; ;in, is the offset distance between the predicted boundary point and the actual boundary point; is the normalization factor of the risk area; n is the number of boundary points in the risk area; i is the boundary point index; and is the horizontal and vertical coordinates of the i-th predicted boundary point; and is the horizontal and vertical coordinates of the ith real boundary point; max(·) is the maximum value; Constructing risk range overlapping loss terms , by calculating the overlap of the risk area, the matching degree between the predicted range and the actual range is measured, which is expressed as: ; ; ;in, and They are the predicted risk range and the actual risk range; and are the dynamic weights of the intersection and union, respectively; is the number of monitoring points; i2 is the monitoring point index; Final loss function It is expressed as: ; ;in, is the dynamic weighting coefficient; is the minimum enclosing rectangular area of ​​the predicted range and the true range.

5. According to claim 4, a chemical park chemical circulation risk supervision system based on industrial Internet is characterized in that: In the data collection module, the historical chemical park chemical circulation data includes temperature sensor data, gas concentration sensor data, pressure sensor data, flow sensor data, sampling time, logistics data, transportation equipment data and circulation risk assessment results; the circulation risk assessment results are used as data labels; Convert and standardize the collected data; The circulation risk assessment results include normal, low risk, medium risk and high risk; when collecting historical chemical park chemical circulation data, the sampling frequency and sampling range of the sensor are dynamically adjusted according to environmental changes, which is expressed as: ; ; Multi-sensor data fusion is performed, which is expressed as: ; and perform data enhancement; enhance the training data by simulating chemical leakage and equipment failure scenarios to make the model more adaptable to different risk environments; the generation of enhanced data is expressed as: ;in, and are the sampling frequency and sampling range of the sensor respectively; , and are the readings of the temperature sensor, gas sensor, and pressure sensor at time t; , and is the adjustment factor of the sampling frequency; , and is the adjustment factor of the sampling range; is the fused sensor data value; is the sensor weight; is the raw reading of the sensor; nc is the sensor number, ic is the sensor index; is the enhanced sensor data; SF is the simulation scenario factor, which indicates the simulation intensity of faults and leakage; Ls is the simulation scenario duration.

6. According to claim 5, a chemical park chemical circulation risk supervision system based on industrial Internet is characterized in that: The chemical circulation risk assessment model establishment module is based on the established chemical circulation risk assessment model, collects chemical circulation data in the chemical park in real time, and inputs it into the chemical circulation risk assessment model after preprocessing. When the circulation risk assessment result output by the model is medium risk, the sampling frequency is increased, and when the output is high risk, early warning processing is carried out for management personnel.

Images