Chemical production area environment safety early warning system based on big data analysis

Abstract

The invention discloses a chemical production area environment safety early warning system based on big data analysis, and relates to the technical field of intelligent prediction, the system comprises an acquisition module, an analysis module and an early warning module, an LSTM model is trained, a prediction model is obtained, a predicted expected output matrix is obtained, the prediction model is updated, and current monitoring parameters and current prediction nodes are obtained. And calculating a second prediction value of the current prediction node, obtaining a standard monitoring parameter range, carrying out safety level division on the second prediction value according to the standard monitoring parameter range to obtain a safety level, and sending an early warning signal according to the safety level. According to the invention, a more accurate basis is provided for accident prediction and prevention through a big data analysis technology, equipment abnormity, production line fluctuation and emergencies are identified by adopting a deep learning technology, the risk of potential accidents is reduced, sudden environmental events occurring in a chemical industry park can be quickly responded, invalid alarms are greatly reduced, and the safety of the chemical industry park is improved. And the accuracy of risk hierarchical management and control is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent prediction, and in particular to an environmental safety early warning system for chemical production areas based on big data analysis. Background Art

[0002] In recent years, the environmental safety early warning technology for chemical production areas has been in a stage of rapid development, with strong policy support, continuous in-depth research and application of technologies. The application of big data and intelligent technologies provides new solutions for chemical safety early warning. With the rapid development of big data, through data sharing, the fuzzy calculation theory has been widely applied in technology. The current research on the safety early warning system for chemical enterprises is still in the initial stage, and its theoretical basis still comes from other industries, mainly the application of mathematical model theory.

[0003] Currently, in the Chinese invention patent with the publication number CN 116453292 B, a method for classifying and early warning the production safety risk level of chemical enterprises is disclosed. This method obtains the operation risk coefficient through the operation data of chemical production equipment, obtains the equipment risk coefficient based on the operation risk coefficient and the proportion of fire hydrants with abnormal water pressure, constructs an evaluation model based on the personnel risk coefficient, equipment risk coefficient, and risk source risk coefficient to obtain the safety risk coefficient, and divides the production safety risk level based on the safety risk coefficient, and determines whether to give an early warning according to the production safety risk level. However, in the related technology, the safety level is not classified according to the historically monitored parameters, lacking the scientificity and applicability of data analysis, and the prediction model is not compensated and updated through historical monitoring data, which is likely to cause prediction errors and is not conducive to the accuracy of prediction. Summary of the Invention

[0004] The technical problem solved by the present invention is that in the related technology, the safety level is not classified according to the historically monitored parameters, lacking the scientificity and applicability of data analysis, and the prediction model is not compensated and updated through historical monitoring data, which is likely to cause prediction errors and is not conducive to the accuracy of prediction.

[0005] To solve the above technical problem, the present invention provides the following technical solution: An environmental safety early warning system for chemical production areas based on big data analysis, including a collection module, an analysis module, and an early warning module;

[0006] The collection module is used to obtain the historical monitoring parameters for the first time period;

[0007] The analysis module divides the historical monitoring parameters into input samples and output samples, converts the input samples and output samples into an input matrix and an expected output matrix, trains an LSTM model based on the input matrix and the expected output matrix to obtain a prediction model, acquires any historical monitoring parameter in the second time period, sets a prediction node, predicts the monitoring data of the prediction node according to the prediction model to obtain a predicted expected output matrix, and updates the prediction model according to the predicted expected output matrix and the corresponding historical monitoring parameter;

[0008] The warning module acquires the current monitoring parameter and the current prediction node, calculates a second predicted value of the current prediction node according to the updated prediction model, acquires the standard monitoring parameter range, divides the second predicted value into safety levels according to the standard monitoring parameter range to obtain a safety level, and sends a warning signal according to the safety level.

[0009] As a preferred solution of the chemical production area environmental safety warning system based on big data analysis according to the present invention, wherein: the acquisition module automatically inputs the account number and password of the safety officer in the production area to be detected, logs in to the production monitoring software, and retrieves the corresponding production monitoring log, sets the first time period as the first acquisition time period, sets the second time period as the second acquisition time period, respectively acquires the historical monitoring logs in the first time period and the historical monitoring logs in the second time period, sets the bed temperature, inlet pressure, separator liquid level, hydrogen-nitrogen ratio, and coolant flow rate as keywords, automatically extracts the monitoring information corresponding to the keywords in the historical monitoring logs in the first time period and the historical monitoring logs in the second time period, and deletes the remaining monitoring information, and sets the extracted monitoring information corresponding to the keywords as historical monitoring parameters;

[0010] The historical monitoring parameters include historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate.

[0011] As a preferred solution of the chemical production area environmental safety warning system based on big data analysis according to the present invention, wherein: the division logic of the input samples and output samples includes:

[0012] Acquire the first time period, set a first time ratio, divide the first time period into a first sub-time period and a second sub-time period according to the first time ratio, wherein the first sub-time period and the second sub-time period are distributed in chronological order, and the first time ratio is set according to the conventional setting values of the training set and the test set;

[0013] Set the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate in the first sub-time period as input samples, and set the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate in the second sub-time period as output samples.

[0014] As a preferred solution of the environmental safety early warning system for chemical production areas based on big data analysis according to the present invention, wherein: the historical monitoring parameters are set as x i , wherein the historical bed temperature is set as x1, the historical inlet pressure is set as x2, the historical separator liquid level x3, the historical hydrogen-nitrogen ratio x4 and the historical coolant flow rate x5;

[0015] Set the time point numbers of the historical monitoring parameters within the first time period, and the setting logic of the time point numbers of the historical monitoring parameters within the first time period includes:

[0016] In chronological order, sort and number the time points corresponding to the data collected at the same time point. The numbers of the time points are natural numbers and are distributed from 1 to t. Among them, the time points within the first sub-time period are distributed from 1 to m, and the time points within the second sub-time period are distributed from m + 1 to t.

[0017] As a preferred solution of the environmental safety early warning system for chemical production areas based on big data analysis according to the present invention, wherein: convert the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio and historical coolant flow rate within the first sub-time period into an input matrix, and the input matrix is expressed as:

[0018]

[0019] Wherein, X is the input matrix, and the first 5 columns of elements from left to right represent the monitored historical production parameters, and the sixth column of elements represents the serial number of the time point corresponding to the elements in the corresponding row;

[0020] The expected output matrix is expressed as:

[0021] Y = [x m+1 x m+2 ...x t i] T

[0022] Wherein, Y is the expected output matrix, that is, the predicted value. The first t - m columns of elements from left to right represent the output values of the historical monitoring parameter numbered i at any time point within the second sub-time period, and the (t - m + 1)-th column of elements represents the number of the historical monitoring parameter, and T represents the transpose flag.

[0023] As a preferred solution of the environmental safety early warning system for chemical production areas based on big data analysis according to the present invention, wherein: obtain the monitoring period Q, and the expected output matrix represents any predicted value of the historical monitoring parameter numbered i within the time period of mQ to tQ, and any predicted value represents the predicted value at an integer multiple of any of the monitoring periods;

[0024] Any element in the input matrix and its corresponding element in the expected output matrix are combined to form a training sample, and the expression of the training sample is:

[0025] P j =(x j , Y j )

[0026] where P j is the training sample, x j is an element in the input matrix with any number j, and Y j is the expected output matrix corresponding to the element in the input matrix with the number j.

[0027] As a preferred solution of the chemical production area environmental safety early warning system based on big data analysis according to the present invention, wherein: an activation function, input weights, and the number of hidden layers are selected, the training sample is transmitted to the LSTM model for training, and the output weights are calculated according to the inverse matrix method to obtain a prediction model. The prediction model is the LSTM model trained by the training sample and can predict monitoring parameters according to time series, that is, by inputting the serial number of the time point corresponding to the element in the input matrix with any number j into the prediction model, the corresponding expected output matrix can be obtained.

[0028] As a preferred solution of the chemical production area environmental safety early warning system based on big data analysis according to the present invention, wherein: an actual output matrix is obtained according to the prediction model, and the acquisition logic of the actual output matrix includes:

[0029] Set the minimum output weight of the training sample, and match the first output matrix according to the minimum output weight. The calculation expression of the first output matrix is:

[0030]

[0031] Y0 and H0 are different expression forms of the first output matrix, g is the activation function, W L is the random input weight corresponding to the Lth historical monitoring parameter with any number, b L is the Lth hidden layer unit, and y L is the output of the Lth hidden layer unit.

[0032] As a preferred solution of the chemical production area environmental safety early warning system based on big data analysis according to the present invention, wherein: according to the inverse matrix method, the first output weight is calculated, and the calculation expression of the first output weight is:

[0033]

[0034] Obtain the predicted expected output matrix of the monitoring parameter according to the prediction model;

[0035] Obtain any historical monitoring parameter in the second time period, and set prediction nodes. The setting logic of the prediction nodes includes:

[0036] Obtain the monitoring period and the historical monitoring time points corresponding to the historical monitoring parameters, perform weighted calculation on the monitoring period or multiples of the monitoring period and the historical monitoring time points to obtain each prediction node, and obtain the time point numbers of the prediction nodes. The setting logic of the time point numbers of the prediction nodes is the same as that of the time point numbers of the historical monitoring parameters in the first time period;

[0037] Obtain each historical monitoring data and its corresponding time point number after obtaining the expected output matrix to be predicted, obtain the time point numbers corresponding to each element in the expected output matrix to be predicted, compare the historical monitoring data with the same time number with the elements in the expected output matrix to be predicted, and set the first difference threshold as the maximum allowable amount of error.

[0038] When the differences between each historical monitoring data with the same time number and the elements in the expected output matrix to be predicted are all less than the first difference threshold, stop updating the prediction model. Otherwise, set the second value as the change amount of the first output weight, continuously increase or decrease the first output weight, and obtain the differences between each historical monitoring data with the same time number and the elements in the expected output matrix to be predicted after regulation. When the differences between each historical monitoring data with the same time number and the elements in the expected output matrix to be predicted are all less than the first difference threshold, stop updating the prediction model.

[0039] As a preferred solution of the chemical production area environmental safety early warning system based on big data analysis according to the present invention, wherein: the early warning module obtains the current monitoring parameter and obtains the current prediction node;

[0040] Input the current monitoring parameter into the updated prediction model, calculate the expected output matrix to be predicted currently according to the updated prediction model, and obtain the current monitoring period and the time point corresponding to the current monitoring parameter;

[0041] When the difference between the current prediction node and the time point corresponding to the current monitoring parameter is an integer multiple of the current monitoring period, obtain the element in the expected output matrix to be predicted corresponding to the current prediction node, denoted as the second predicted value;

[0042] When the difference between the current prediction node and the time point corresponding to the current monitoring parameter is not an integer multiple of the current monitoring period, set the time point corresponding to the difference between the current prediction node and the time point corresponding to the current monitoring parameter being an integer multiple of the current monitoring period as the candidate time point, and obtain the candidate time points adjacent to the current prediction node. The adjacent candidate time points are represented as the candidate time points with the smallest absolute value of the difference from the time point corresponding to the current prediction node. Obtain the elements in the current expected output matrix corresponding to the current prediction node, denoted as the second prediction value;

[0043] Retrieve the production database and the name of the current monitoring parameter, input the name of the current monitoring parameter into the production database, obtain the corresponding standard monitoring parameter range, and perform a safety level classification on the second prediction value according to the standard monitoring parameter range to obtain the safety level. The classification logic of the safety level includes:

[0044] Obtain the upper limit value and the lower limit value of the standard monitoring parameter range, compare the upper limit value and the lower limit value with the elements in the current expected output matrix. When the element in the current expected output matrix is greater than or equal to the upper limit value, set the safety level to the first level. When the element in the current expected output matrix is less than or equal to the lower limit value, set the safety level to the second level. When the element in the current expected output matrix is less than the upper limit value and greater than the lower limit value, set the safety level to the third level, where the safety levels of the first level, the second level, and the third level increase gradually;

[0045] Send a warning signal according to the safety level. When the safety level is the first level, send the first warning signal. When the safety level is the second level, send the second warning signal. When the safety level is the third level, send the third warning signal.

[0046] The beneficial effects of the present invention: Through big data analysis technology, the system can deeply explore the potential patterns and laws of chemical enterprise accidents, providing a more comprehensive perspective for the enterprise, enabling it to understand the underlying mechanism of accident occurrence more deeply, providing a more accurate basis for accident prediction and prevention. By adopting deep learning technology, real-time intelligent monitoring of the production process is realized, which can identify equipment abnormalities, production line fluctuations, and emergencies, discover problems in a timely manner and issue warnings, reducing the risk of potential accidents, being able to quickly respond to sudden environmental events occurring in chemical industrial parks, significantly reducing ineffective alarms, and effectively improving the accuracy of risk classification and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the basic process of the environmental safety warning system for chemical production areas based on big data analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0048] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments.

[0049] Example, referring to Figure 1 , which is an embodiment of the present invention, provides an environmental safety early warning system for chemical production areas based on big data analysis, including a collection module, an analysis module, and an early warning module;

[0050] The collection module is used to obtain historical monitoring parameters for the first time period;

[0051] The analysis module divides the historical monitoring parameters into input samples and output samples, converts the input samples and output samples into an input matrix and an expected output matrix, trains an LSTM model based on the input matrix and the expected output matrix to obtain a prediction model, obtains any historical monitoring parameter for the second time period, sets a prediction node, predicts the monitoring data of the prediction node according to the prediction model to obtain a predicted expected output matrix, and updates the prediction model according to the predicted expected output matrix and the corresponding historical monitoring parameter;

[0052] The early warning module obtains the current monitoring parameter and the current prediction node, calculates the second prediction value of the current prediction node according to the updated prediction model, obtains the standard monitoring parameter range, divides the second prediction value into safety levels according to the standard monitoring parameter range to obtain a safety level, and sends an early warning signal according to the safety level.

[0053] Through big data analysis technology, the system of the present invention can deeply explore the potential patterns and rules of accidents in chemical enterprises, providing a more comprehensive perspective for enterprises, enabling them to understand the underlying mechanisms behind accidents more deeply, providing a more accurate basis for accident prediction and prevention. By adopting deep learning technology, it realizes real-time intelligent monitoring of the production process, can identify equipment abnormalities, production line fluctuations, and emergencies, discovers problems in a timely manner and issues early warnings, reducing the risk of potential accidents, can quickly respond to sudden environmental events occurring in chemical industrial parks, significantly reducing ineffective alarms, and effectively improving the accuracy of risk classification and control.

[0054] The acquisition module automatically inputs the account number and password of the safety officer in the production area to be detected, logs in to the production monitoring software, retrieves the corresponding production monitoring logs, sets the first time period as the first acquisition time period, sets the second time period as the second acquisition time period, respectively obtains the historical monitoring logs of the first time period and the historical monitoring logs of the second time period, sets the bed temperature, inlet pressure, separator liquid level, hydrogen-nitrogen ratio, and coolant flow rate as keywords, automatically extracts the monitoring information corresponding to the keywords in the historical monitoring logs of the first time period and the historical monitoring logs of the second time period, deletes the remaining monitoring information, and sets the extracted monitoring information corresponding to the keywords as historical monitoring parameters;

[0055] The historical monitoring parameters include historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate.

[0056] In specific implementation, logging in to the monitoring software by automatically inputting the account number and password reduces the complexity of manual operations, improves the efficiency and accuracy of data acquisition. By setting specific time periods, the system can collect historical monitoring logs in a targeted manner, which helps to more accurately analyze the production status within a specific time period, thereby improving the timeliness and accuracy of the early warning system. By setting keywords (bed temperature, inlet pressure, separator liquid level, hydrogen-nitrogen ratio, and coolant flow rate), the system can automatically extract the monitoring information related to these parameters and delete the irrelevant data, thus reducing the complexity of data processing and improving the pertinence and efficiency of data processing.

[0057] The division logic of the input sample and the output sample includes:

[0058] Obtain the first time period, set the first time ratio, and divide the first time period into a first sub-time period and a second sub-time period according to the first time ratio. Among them, the first sub-time period and the second sub-time period are distributed in chronological order, and the first time ratio is set by the conventional setting values of the training set and the test set;

[0059] Set the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate within the first sub-time period as the input sample, and set the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate within the second sub-time period as the output sample.

[0060] In specific implementation, by dividing the first time period into a first sub - time period and a second sub - time period, and setting them as input samples and output samples respectively, the changing trend of parameters in the chemical production process can be captured more precisely, thereby improving the accuracy of the prediction model. Reasonable sample division helps train a more effective LSTM model, enabling the model to better learn and predict the complex dynamic process in chemical production. By dynamically adjusting the first time ratio, the system can adapt to different production conditions and environmental changes, improving the flexibility and adaptability of the early - warning system.

[0061] Set the historical monitoring parameters as x i , where the historical bed - layer temperature is set as x1, the historical inlet pressure is set as x2, the historical separator liquid level is x3, the historical hydrogen - nitrogen ratio is x4, and the historical coolant flow rate is x5;

[0062] Set the time - point numbers of the historical monitoring parameters within the first time period. The setting logic of the time - point numbers of the historical monitoring parameters within the first time period includes:

[0063] In chronological order, sort and number the time points corresponding to the data collected at the same time point. The numbers of the time points are natural numbers and are distributed from 1 to t. Among them, the time points within the first sub - time period are distributed from 1 to m, and the time points within the second sub - time period are distributed from m + 1 to t.

[0064] In specific implementation, through automated time - point numbering and data division, the system can quickly process a large amount of historical monitoring data, improving the efficiency of data processing. Through accurate time - point numbering, the system can real - time monitor and predict key parameters in the production process, promptly detect abnormal situations, and improve the timeliness of the early - warning system. The automated process reduces human operation errors and improves the accuracy and reliability of data division.

[0065] Convert the historical bed - layer temperature, historical inlet pressure, historical separator liquid level, historical hydrogen - nitrogen ratio, and historical coolant flow rate within the first sub - time period into an input matrix. The input matrix is expressed as:

[0066]

[0067] where X is the input matrix. The first 5 columns of elements from left to right represent the monitored historical production parameters, and the sixth - column element represents the serial number of the time point corresponding to the element in the corresponding row;

[0068] The expected output matrix is expressed as:

[0069] Y = [x m+1 x m+2 ...x t i] T

[0070] Among them, Y is the expected output matrix, that is, the predicted value. The elements of the first t - m columns from left to right represent the output values of the historical monitoring parameter numbered i at any time point in the second sub - time period, and the element of the (t - m + 1)-th column represents the number of the historical monitoring parameter. T represents the transpose flag.

[0071] In specific implementation, the clear division of the input matrix and the output matrix helps to train machine learning models, especially time - series prediction models such as LSTM, to optimize parameters and train the model more efficiently. By incorporating the time - point sequence number into the input matrix, the model can capture the trends and patterns in time - series data, thereby improving the accuracy of prediction. After converting the monitoring parameters into matrix form, real - time data updating is facilitated, enabling the real - time monitoring system to capture abnormal patterns and issue early warnings in a timely manner.

[0072] Obtain the monitoring period Q. The expected output matrix represents any predicted value of the historical monitoring parameter numbered i within the time period from mQ to tQ, and any predicted value represents the predicted value at an integer multiple of any monitoring period.

[0073] Form a training sample by combining any element in the input matrix and its corresponding element in the expected output matrix. The expression of the training sample is:

[0074] P j =(x j , Y j )

[0075] Among them, P j is the training sample, x j is an element in the input matrix numbered j, and Y j is the expected output matrix corresponding to the element in the input matrix numbered j.

[0076] In specific implementation, by converting historical monitoring parameters into input matrices and expected output matrices and forming training samples, the time - series prediction model can be trained more precisely, thereby improving the prediction accuracy. By setting predicted values within different monitoring periods, the system can adapt to different production conditions and environmental changes, improving the flexibility and adaptability of the early - warning system. Obtaining the monitoring period Q, the expected output matrix represents any predicted value of the historical monitoring parameter numbered i within the time period from mQ to tQ, which helps the model to make predictions within a specific time period.

[0077] Select an activation function, input weights, and the number of hidden layers, transmit the training samples to the LSTM model for training, calculate the output weights according to the inverse matrix method to obtain a prediction model. The prediction model is an LSTM model trained with training samples and can predict monitoring parameters according to time series, that is, by inputting the sequence number of the time point corresponding to the elements in any input matrix numbered j into the prediction model, the corresponding expected output matrix can be obtained.

[0078] In specific implementation, by carefully selecting the activation function and adjusting the input weights, the LSTM model can better capture complex patterns and long-term dependencies in time series data, thereby improving the accuracy of prediction. A reasonable number of hidden layers and neurons enhances the generalization ability of the model, enabling the model to also exhibit good prediction performance on unseen data. Using the inverse matrix method to calculate the output weights optimizes the performance of the model, especially in dealing with linear relationships and simplifying the model structure. Selecting appropriate activation functions and weight initialization methods speeds up the convergence rate of the model and improves the training efficiency. The parameter settings of the LSTM include: the number of features is 5, the hidden layer size is 32, the output size is 100, the number of LSTM layers is 4, and the activation function is GRU.

[0079] Obtain the actual output matrix according to the prediction model. The acquisition logic of the actual output matrix includes:

[0080] Set the minimum output weight of the training samples, and match the first output matrix. The calculation expression of the first output matrix is:

[0081]

[0082] Y0 and H0 are different expression forms of the first output matrix, g is the activation function, W L is the random input weight corresponding to the Lth historical monitoring parameter of any number, b L is the Lth hidden layer unit, y L is the output of the Lth hidden layer unit.

[0083] In specific implementation, by setting the minimum output weight, the model's dependence on unimportant features is reduced, thereby improving the accuracy of prediction. Matching the first output matrix helps the model better learn the patterns and trends in the data and optimize the performance of the model. By setting the minimum output weight, the structure of the model is simplified, unnecessary parameters are reduced, and the efficiency of the model is improved.

[0084] According to the inverse matrix method, calculate the first output weight. The calculation expression of the first output weight is:

[0085]

[0086] Obtain the predicted expected output matrix of the monitoring parameters according to the prediction model;

[0087] Obtain any historical monitoring parameter in the second time period and set prediction nodes. The setting logic of the prediction nodes includes:

[0088] Obtain the monitoring period and the historical monitoring time point corresponding to the historical monitoring parameter, perform weighted calculation on the monitoring period or a multiple of the monitoring period and the historical monitoring time point to obtain each prediction node, obtain the time point number of the prediction node, and the setting logic of the time point number of the prediction node is the same as that of the time point number of the historical monitoring parameter in the first time period;

[0089] Obtain each historical monitoring data and its corresponding time point number after obtaining the predicted expected output matrix, obtain the time point number corresponding to each element in the predicted expected output matrix, compare the historical monitoring data with the same time number with the elements in the predicted expected output matrix, and set the first difference threshold as the maximum allowable amount of error;

[0090] When the difference between each historical monitoring data with the same time number and the element in the predicted expected output matrix is less than the first difference threshold, stop updating the prediction model; otherwise, set the second value as the change amount of the first output weight, continuously increase or decrease the first output weight, and obtain the difference between each historical monitoring data with the same time number and the element in the predicted expected output matrix after regulation. When the difference between each historical monitoring data with the same time number and the element in the predicted expected output matrix is less than the first difference threshold, stop updating the prediction model.

[0091] In specific implementation, the inverse matrix method provides a method for directly calculating the output weight, simplifies the calculation process of weight estimation, and improves the generalization ability of the model to new data by accurately calculating the output weight.

[0092] The early warning module obtains the current monitoring parameter and obtains the current prediction node;

[0093] Input the current monitoring parameter into the updated prediction model, calculate the predicted current expected output matrix according to the updated prediction model, and obtain the current monitoring period and the time point corresponding to the current monitoring parameter;

[0094] When the difference between the current prediction node and the time point corresponding to the current monitoring parameter is an integer multiple of the current monitoring period, obtain the element in the current expected output matrix corresponding to the current prediction node, denoted as the second predicted value;

[0095] When the difference between the current prediction node and the time point corresponding to the current monitoring parameter is not an integer multiple of the current monitoring period, set the time point corresponding to the difference between the current prediction node and the time point corresponding to the current monitoring parameter being an integer multiple of the current monitoring period as the candidate time point, and obtain the candidate time points adjacent to the current prediction node. The adjacent candidate time points are represented as the candidate time points with the smallest absolute value of the difference from the time point corresponding to the current prediction node. Obtain the elements in the current expected output matrix corresponding to the current prediction node, denoted as the second prediction value;

[0096] Retrieve the production database and the name of the current monitoring parameter, input the name of the current monitoring parameter into the production database, obtain the corresponding standard monitoring parameter range, and perform a safety level classification on the second prediction value according to the standard monitoring parameter range to obtain the safety level. The classification logic of the safety level includes:

[0097] Obtain the upper limit value and the lower limit value of the standard monitoring parameter range, compare the upper limit value and the lower limit value with the elements in the current expected output matrix. When the element in the current expected output matrix is greater than or equal to the upper limit value, set the safety level to the first level; when the element in the current expected output matrix is less than or equal to the lower limit value, set the safety level to the second level; when the element in the current expected output matrix is less than the upper limit value and greater than the lower limit value, set the safety level to the third level, where the safety degrees of the first level, the second level, and the third level increase gradually;

[0098] Send a warning signal according to the safety level. When the safety level is the first level, send the first warning signal; when the safety level is the second level, send the second warning signal; when the safety level is the third level, send the third warning signal.

[0099] In specific implementation, by obtaining the current monitoring parameter and the prediction node in real time, the system can respond to the changes in the production process in a timely manner, quickly send a warning signal, and reduce potential safety risks. By ensuring that the difference between the prediction node and the time point corresponding to the monitoring parameter is an integer multiple of the monitoring period, the system can accurately synchronize the time points and improve the accuracy of prediction. When the difference between the prediction node and the time point of the monitoring parameter is not an integer multiple of the monitoring period, the system can flexibly process the candidate time points and select the candidate time point with the smallest absolute value of the time difference from the current prediction node to ensure the timeliness and accuracy of the warning. By comparing with the standard monitoring parameter range in the production database, the system can perform a standardized safety level classification on the prediction value, making the warning signal more scientific and standardized.

[0100] Through big data analysis technology, the system can deeply explore the potential patterns and rules of chemical enterprise accidents, providing a more comprehensive perspective for the enterprise, enabling it to more deeply understand the underlying mechanism of accident occurrence, providing a more accurate basis for accident prediction and prevention. By adopting deep learning technology, real-time intelligent monitoring of the production process is achieved, which can identify equipment abnormalities, production line fluctuations, and emergencies, discover problems in a timely manner and issue early warnings, reducing the risk of potential accidents. It can quickly respond to sudden environmental events occurring in chemical industrial parks, significantly reducing false alarms and effectively improving the accuracy of risk classification and control.

[0101] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. Among them, the storage medium is implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM for short), electrically erasable programmable read-only memory (EEPROM for short), erasable programmable read-only memory (EPROM for short), programmable read-only memory (PROM for short), read-only memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc. These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one process or multiple processes and / or Figure 1 boxes or multiple boxes.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that modifications or equivalent replacements to the technical solutions of the present invention, without departing from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A chemical production area environmental safety early warning system based on big data analysis, characterized in that, It includes a collection module, an analysis module, and a warning module; The collection module is used to obtain historical monitoring parameters for the first time period; The analysis module divides the historical monitoring parameters into input samples and output samples, converts the input samples and output samples into an input matrix and an expected output matrix, trains an LSTM model based on the input matrix and the expected output matrix to obtain a prediction model, obtains any historical monitoring parameter for the second time period, sets a prediction node, predicts the monitoring data of the prediction node according to the prediction model to obtain a predicted expected output matrix, and updates the prediction model according to the predicted expected output matrix and the corresponding historical monitoring parameter; The warning module obtains the current monitoring parameter and the current prediction node, calculates a second predicted value of the current prediction node according to the updated prediction model, obtains a standard monitoring parameter range, divides the second predicted value into safety levels according to the standard monitoring parameter range to obtain a safety level, and sends a warning signal according to the safety level.

2. The environmental safety early warning system for chemical production areas based on big data analysis according to claim 1, wherein: The collection module automatically inputs the account number and password of the safety officer in the production area to be detected, logs in to the production monitoring software, retrieves the corresponding production monitoring log, sets the first time period as the first collection time period, sets the second time period as the second collection time period, respectively obtains the historical monitoring log for the first time period and the historical monitoring log for the second time period, sets the bed temperature, inlet pressure, separator liquid level, hydrogen-nitrogen ratio, and coolant flow rate as keywords, automatically extracts the monitoring information corresponding to the keywords in the historical monitoring log for the first time period and the historical monitoring log for the second time period, deletes the remaining monitoring information, and sets the extracted monitoring information corresponding to the keywords as historical monitoring parameters; The historical monitoring parameters include historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate.

3. The chemical production area environmental safety early warning system based on big data analysis according to claim 1, wherein: The division logic of the input samples and output samples includes: Obtain the first time period, set a first time ratio, divide the first time period into a first sub-time period and a second sub-time period according to the first time ratio, where the first sub-time period and the second sub-time period are distributed in chronological order, and the first time ratio is set according to the conventional setting values of the training set and the test set; Set the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate within the first sub-time period as input samples, and set the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate within the second sub-time period as output samples.

4. The chemical production area environmental safety early warning system based on big data analysis according to claim 1, characterized in that: Set the historical monitoring parameters to x i , where the historical bed temperature is set to x1, the historical inlet pressure is set to x2, the historical separator liquid level is x3, the historical hydrogen-nitrogen ratio is x4, and the historical coolant flow rate is x5; Set the time point numbers of the historical monitoring parameters within the first time period, and the setting logic of the time point numbers of the historical monitoring parameters within the first time period includes: Sort and number the time points corresponding to the data collected at the same time point in chronological order. The numbers of the time points are natural numbers and are distributed from 1 to t. Among them, the time points within the first sub-time period are distributed from 1 to m, and the time points within the second sub-time period are distributed from m + 1 to t.

5. The chemical production area environmental safety early warning system based on big data analysis according to claim 1, characterized in that: Convert the historical bed temperature, historical inlet pressure, historical separator liquid level, historical hydrogen-nitrogen ratio, and historical coolant flow rate within the first sub-time period into an input matrix, and the input matrix is expressed as: where X is the input matrix, and the first 5 columns of elements from left to right represent the monitored historical production parameters, and the sixth column of elements represents the serial number of the time point corresponding to the elements in the corresponding row; The expected output matrix is expressed as: Y = [x m+1 x m+2 ...x t i] T where Y is the expected output matrix, that is, the predicted value. The first t - m columns of elements from left to right represent the output values of the historical monitoring parameter numbered i at any time point within the second sub-time period, the (t - m + 1)-th column of elements represents the number of the historical monitoring parameter, and T represents the transpose flag.

6. The chemical production area environmental safety early warning system based on big data analysis according to claim 1, characterized in that: Obtain the monitoring period Q. The expected output matrix represents any predicted value of the historical monitoring parameter numbered i within the time period from mQ to tQ, and any predicted value represents the predicted value at an integer multiple of any of the monitoring periods; Form a training sample from any element in the input matrix and its corresponding element in the expected output matrix, and the expression of the training sample is: P j = (x j , Y j ) Among them, P j is the training sample, x j is an element in the input matrix numbered j, Y j is the expected output matrix corresponding to the element in the input matrix numbered j.

7. The environmental safety early warning system for chemical production areas based on big data analysis according to claim 1, characterized in that: Select an activation function, input weights, and the number of hidden layers, transmit the training sample to the LSTM model for training, calculate the output weights according to the inverse matrix method, and obtain a prediction model. The prediction model is an LSTM model trained by the training sample and can predict the monitoring parameters according to the time series, that is, by inputting the serial number of the time point corresponding to the element in the input matrix numbered j into the prediction model, the corresponding expected output matrix can be obtained.

8. The chemical production area environmental safety early warning system based on big data analysis according to claim 1, wherein: Obtain the actual output matrix according to the prediction model, and the acquisition logic of the actual output matrix includes: Set the minimum output weight of the training sample, and match the first output matrix according to the minimum output weight. The calculation expression of the first output matrix is: Y0 and H0 are different expression forms of the first output matrix, g is the activation function, W L is the random input weight corresponding to the L-th historical monitoring parameter with any number, b L is the L-th hidden layer unit, y L is the output of the L-th hidden layer unit.

9. The chemical production area environmental safety early warning system based on big data analysis according to claim 1, characterized in that: Calculate the first output weight according to the inverse matrix method. The calculation expression of the first output weight is: Obtain the predicted expected output matrix of the monitoring parameter according to the prediction model; Obtain any historical monitoring parameter in the second time period, and set prediction nodes. The setting logic of the prediction nodes includes: Obtain the monitoring period and the historical monitoring time point corresponding to the historical monitoring parameter, perform weighted calculation on the monitoring period or a multiple of the monitoring period and the historical monitoring time point to obtain each prediction node, obtain the time point number of the prediction node, and the setting logic of the time point number of the prediction node is the same as that of the time point number of the historical monitoring parameter within the first time period; Obtain each historical monitoring data and its corresponding time point number after obtaining the predicted expected output matrix, obtain the time point number corresponding to each element in the predicted expected output matrix, compare the historical monitoring data with the same time number with the elements in the predicted expected output matrix, and set the first difference threshold as the maximum allowable amount of error; When the differences between the historical monitoring data with the same time number and the elements in the predicted expected output matrix are all less than the first difference threshold, stop updating the prediction model; otherwise, set the second value as the change amount of the first output weight, continuously increase or decrease the first output weight, and obtain the differences between the historical monitoring data with the same time number after regulation and the elements in the predicted expected output matrix. When the differences between the historical monitoring data with the same time number and the elements in the predicted expected output matrix are all less than the first difference threshold, stop updating the prediction model.

10. The chemical production area environmental safety early warning system based on big data analysis according to claim 1, characterized in that: The warning module obtains the current monitoring parameters and the current prediction node. Input the current monitoring parameters into the updated prediction model, calculate the predicted current expected output matrix according to the updated prediction model, and obtain the current monitoring cycle and the time point corresponding to the current monitoring parameters. When the difference between the current prediction node and the time point corresponding to the current monitoring parameters is an integer multiple of the current monitoring cycle, obtain the element in the current expected output matrix corresponding to the current prediction node, denoted as the second predicted value. When the difference between the current prediction node and the time point corresponding to the current monitoring parameters is not an integer multiple of the current monitoring cycle, set the time point corresponding to the difference between the current prediction node and the time point corresponding to the current monitoring parameters being an integer multiple of the current monitoring cycle as the candidate time point, obtain the candidate time points adjacent to the current prediction node, where the adjacent candidate time points are represented as the candidate time points with the smallest absolute value of the difference from the time point corresponding to the current prediction node, and obtain the element in the current expected output matrix corresponding to the current prediction node, denoted as the second predicted value. Retrieve the production database and the name of the current monitoring parameters, input the name of the current monitoring parameters into the production database, obtain the corresponding standard monitoring parameter range, and perform a safety level classification on the second predicted value according to the standard monitoring parameter range to obtain the safety level. The classification logic of the safety level includes: Obtain the upper limit value and the lower limit value of the standard monitoring parameter range, compare the upper limit value and the lower limit value with the element in the current expected output matrix. When the element in the current expected output matrix is greater than or equal to the upper limit value, set the safety level to the first level; when the element in the current expected output matrix is less than or equal to the lower limit value, set the safety level to the second level; when the element in the current expected output matrix is less than the upper limit value and greater than the lower limit value, set the safety level to the third level, where the safety degrees of the first level, the second level, and the third level increase gradually. Send a warning signal according to the safety level. When the safety level is the first level, send the first warning signal; when the safety level is the second level, send the second warning signal; when the safety level is the third level, send the third warning signal.

Images