Chemical tower equipment online HAZOP quantitative analysis and risk prediction method

Abstract

The application discloses a kind of chemical tower equipment online HAZOP quantification analysis and risk prediction method, comprising: collecting chemical tower equipment online data, the online data is trained based on neural network prediction model, obtains trained neural network prediction model, and obtains the predicted physical quantity of target variable by the trained neural network prediction model;Determine the standard comparison value of target variable, based on standard comparison value, process simulation or twin digital model analysis is carried out, and the theoretical optimal target value under the standard comparison value corresponding working condition is obtained;Compare predicted physical quantity with the theoretical optimal target value, obtain the deviation of predicted value and HAZOP prompt, based on the deviation of predicted value and HAZOP prompt, adjust chemical tower equipment, and generate risk analysis form.The application quantifies HAZOP deviation and timely analyzes to accurately and directionally obtain information leading to device safety problem.

Description

Technical Field

[0001] This invention relates to the field of chemical technology, and in particular to a method for online HAZOP quantitative analysis and risk prediction of chemical tower equipment. Background Technology

[0002] In chemical separation processes, tower equipment is a key component. Taking distillation towers as an example, they involve numerous controlled and manipulated variables, a wide variety and quantity of equipment, and the composition of the resulting products often requires precise control. Furthermore, tower equipment is often a pressure vessel (positive or negative pressure), and utilities involve high-temperature hot media, low-temperature cold media, and pressure-controlled gases. Chemical products themselves, due to their physical properties, may also possess certain risk characteristics. Therefore, hazard analysis and monitoring at each stage and node in chemical separation processes are particularly important. HAZOP analysis is one of the main methods for analyzing safety issues in modern production and life. It can be used to analyze potential hazards in chemical production. However, HAZOP is a qualitative analysis, relying on experienced professionals in related fields for analysis and discussion to identify potential biases and consequences, leading to reflection and improvement. The analysis results suffer from ambiguity and poor accuracy. In recent years, some scholars and developers have conducted quantitative research on HAZOP, but online quantitative analysis and prediction of HAZOP at various nodes of chemical process tower equipment during operation is still imperfect.

[0003] Therefore, there is a need for an HZAOP analysis method that can acquire and quantitatively analyze data online during the operation of a chemical process, providing quantitative information on deviations at each node and offering prompts and adjustments. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes an online HAZOP quantitative analysis and risk prediction method for chemical tower equipment.

[0005] A method for online HAZOP quantitative analysis and risk prediction of chemical tower equipment, including:

[0006] Online data from the chemical tower equipment is collected, and the online data is used to train a neural network prediction model to obtain a trained neural network prediction model. The trained neural network prediction model is then used to obtain the predicted physical quantity of the target variable. The target variable is the reference value that needs to be quantified during the HZAOP quantitative analysis of the process, and the deviation between the quantified and measured parameters and the theoretical value under the corresponding operating conditions.

[0007] Determine the standard comparison value of the target variable, and perform process simulation or twin digital model analysis based on the standard comparison value of the target variable to obtain the theoretical optimal target value under the working condition corresponding to the standard comparison value;

[0008] By comparing the predicted physical quantity with the theoretical optimal target value, the deviation of the predicted value and HAZOP hints are obtained. Based on the deviation of the predicted value and HAZOP hints, the chemical tower equipment is adjusted, and a risk analysis table is generated.

[0009] Preferably, the online data of the chemical tower equipment includes feed data, tower top data, tower bottom data, and setpoint parameters of all automatic control instruments;

[0010] The feed data includes: feed flow rate, feed temperature, feed pressure, and feed valve opening.

[0011] The data at the top of the tower includes: top output, top pressure, top temperature, top condenser cold medium flow rate, top condenser cold medium temperature, top condenser cold medium valve opening, number or power of air-cooled units in operation, vent valve opening, reflux tank level, composition of the extracted material, and pressure control gas valve opening.

[0012] The bottom data includes: bottom output, bottom pressure, bottom temperature, reboiler heat medium flow rate, reboiler heat medium temperature, top condenser valve opening, composition of the output material, and bottom liquid level.

[0013] Preferably, obtaining a trained neural network prediction model includes:

[0014] The online data from the chemical tower equipment is input into the neural network prediction model and normalized to obtain X = (x1, x2, x3, ... x...). n ) T ;

[0015] Add several neurons to the hidden layer of the neural network prediction model. The output target of each neuron includes pressure P = (P1, P2). T Temperature T = (T D T W ) T Composition X = (X D X W ) T The number of input parameters corresponding to the output target of the neuron is obtained, and the weight of each input parameter in the neuron connection weight is w. ni (n = 1, 2, ..., n) T The input parameters are any parameters from the online data of the chemical tower equipment.

[0016] Based on the weight values, the initial number of nodes in the hidden layer is obtained, and the trained neural network prediction model is obtained.

[0017] The threshold of the neuron is bi, and the activation value is ui.

[0018] Preferably, the activation value is the weighted sum and difference, and the output value is a function of the input value.

[0019] The method for calculating the activation value is as follows:

[0020] u i =∑(w ni *n i )-b i

[0021] Y i =f(u i )

[0022] Among them, u i w is the activation value. ni b is the connection weight for the input parameters. i Y is the threshold of the neuron. i The output is f(), which is the activation function.

[0023] Preferably, the neural network prediction model determines the initial number of nodes j of the hidden layer using empirical formulas or cascade correlation methods, wherein the empirical formulas include, but are not limited to:

[0024]

[0025] j = log₂n1

[0026]

[0027] Where n1 and n2 are the number of input layer nodes and the number of output layer nodes, respectively, and m is the number of samples.

[0028] Preferably, the method for determining the standard comparison value of the target variable includes determining it by establishing a calculation model of the theoretical process, or obtaining it by obtaining the instrument setting value during device operation.

[0029] Preferably, obtaining the theoretically optimal target value under the working condition corresponding to the standard comparison value includes:

[0030] By establishing a model or using commercial process simulation tools or software, the theoretical optimal target value of the target variable under the corresponding working conditions is obtained, and all relevant set values ​​are taken and correlated to obtain the pressure P' = (P1, P2). T Temperature T' = (T D T W ) T Composition X' = (X D X W ) T Liquid level L' = (L R LW ) T .

[0031] Preferably, comparing the predicted physical quantity with the theoretically optimal target value includes:

[0032] The difference between the predicted physical quantity and the theoretical optimal target value is calculated to obtain the corresponding difference values ​​for different data, and the corresponding difference values ​​are compared with the allowable error respectively.

[0033] Compared with the prior art, the present invention has the following advantages and technical effects:

[0034] This invention utilizes online data acquisition and data model building to perform parameter analysis under operating conditions and compares the results with those under standard conditions. It quantifies and analyzes HAZOP deviations in a timely manner, thereby accurately and specifically obtaining information that leads to device safety issues. Attached Figure Description

[0035] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0036] Figure 1 This is a schematic diagram of an online HAZOP quantitative analysis and risk prediction method for chemical tower equipment in an embodiment of the present invention;

[0037] Figure 2 This is a flowchart illustrating the online HAZOP quantitative analysis and risk prediction method for chemical tower equipment in this embodiment of the invention.

[0038] Figure 3 This is a schematic diagram of the neural network prediction model structure in an embodiment of the present invention. Detailed Implementation

[0039] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0040] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0041] This invention proposes an online HAZOP quantitative analysis and risk prediction method for chemical tower equipment, such as... Figures 1-2 Specifically, it includes:

[0042] (1) Collect online information of chemical tower equipment, train and learn the neural network prediction model, and after the training is qualified (judged by the error of the training sample and the inspection sample), predict the operating conditions through online parameter processing, and obtain the predicted operating conditions and the target physical quantities that need to be accurately analyzed under the operating conditions.

[0043] The collected online data from the chemical tower equipment includes:

[0044] Feed: Feed flow rate (F), feed temperature (T) F ), feed pressure (P) F ), feed valve opening (V) F );

[0045] Top of the tower: Top output (D), top pressure (P1), top temperature (T) D ), Cooling medium flow rate (C) of the top condenser, Cooling medium temperature (T) of the top condenser C ), Tower top condensate valve opening (V) D ), number of air-cooled units in operation (N) or power (P), vent valve opening (V), return tank level (L) R ), composition of extracted materials (X) D ), control gas valve opening (V) C );

[0046] Bottom of the column: Bottom output (W), bottom pressure (P2), bottom temperature (T) W ), reboiler heat medium flow rate (S), reboiler heat medium temperature (T) S ), Tower top condensate valve opening (V) W ), composition of extracted materials (X) W ), bottom liquid level (L) W );

[0047] In addition, it includes the setpoint parameters of all automatic control instruments.

[0048] The model used for prediction can be a machine learning neural network model, including an input layer, hidden layers, and an output layer;

[0049] The input signal of the neural network is X; w is the connection weight between the input variable and the neuron; b is the neuron threshold, u is the activation value, f(u) is the activation function, and Y is the output signal. Its structure is as follows: Figure 3 As shown, when the weighting of the neuron's input signal exceeds b, the neuron is in an excited state.

[0050] When using a neural network model, its input is some or all of the obtained information, represented as X = (F, T) F ,P FNormalization yields X = (x1, x2, x3, ..., x...). n ) T ;

[0051] The normalization formula is:

[0052]

[0053] There are i neurons, added in the hidden layer between the input and output layers. The output target of each neuron is the operating parameters of the tower corresponding to the input variables, including pressure P = (P1, P2). T Temperature T = (T D T W ) T Composition X = (X D X W ) T L = (L R L W ) T The weight of each parameter of the selected tower device in the neuron connection weights is w, corresponding to the number of input values ​​obtained. ni (n = 1, 2, ..., n) T The threshold for each neuron is b. i The activation value is u i Y is input through the excitation function f() i .

[0054] In the model, the activation values ​​are the weighted sum and difference of the input values, and the output value is a function of the input values.

[0055] u i =∑(w ni *n i )-b i

[0056] Y i =f(u i )

[0057] Based on the monitoring intent, a portion of the samples within a certain period are used as training samples, and another portion is used as test samples for model training and learning.

[0058] The neural network model is a backpropagation (BP) neural network. The initial number of nodes j in its hidden layers is determined using empirical formulas or cascade correlation methods. The empirical formulas include, but are not limited to, the following:

[0059]

[0060] j = log₂n1

[0061]

[0062] n1 and n2 are the number of nodes in the input layer and the number of nodes in the output layer, respectively, and m is the number of samples.

[0063] Through training and learning of the neural network, an acceptable goodness of fit and mean squared error are achieved. Online data is collected, predictive analysis is performed, and the predicted pressure P = (P1, P2) is obtained through the activation function. T Temperature T = (T D T W ) T Composition X = (X D X W ) T Liquid level L = (L R L W ) T .

[0064] The activation functions of neural network models include those commonly used in BP neural networks, primarily linear activation functions, sigmoid activation functions, hyperbolic tangent activation functions, threshold or step activation functions, and Gaussian activation functions, including but not limited to the logsig function, tansig function, satlin function, tribas function, radbas function, purelin function, or combinations thereof. These can be used in conjunction with genetic algorithms, particle swarm optimization, and cuckoo search algorithms.

[0065] Through neural network training and learning, acceptable fit and mean squared error are achieved. Online data is collected, predictive analysis is performed, and predicted values ​​of target variables are obtained. The target variables are the reference values ​​that need to be quantified during the HZAOP quantitative analysis of the process. The deviation between the quantified and measured parameters and the theoretical values ​​under corresponding operating conditions (including but not limited to different loads, different feed compositions, different utility operating conditions, different separation requirements, and different valve equipment operating conditions) can characterize the probability or severity of risk occurrence to a certain extent.

[0066] The acquired parameters and trained model include, but are not limited to, the pressure P = (P1, P2). T Temperature T = (T D T W ) T Composition X = (X D X W ) T Liquid level L = (L R L W ) T Predictive analysis of the model is performed using variables such as [variable name].

[0067] (2) The standard comparison value of the target variable can be determined by establishing a theoretical process calculation model or by setting the instrument value during device operation.

[0068] During equipment operation, several important parameters and variables are monitored and automated controls are established. Correspondingly, there are setpoints. HAZOP analysis requires a wide range of parameters. In addition to the pre-set values ​​for equipment operation, it uses some or all of the obtained tower equipment operating parameters as initial values ​​to calculate the variable values ​​under theoretical operating conditions. This process can be achieved by building a model or using commercial process simulation tools or software. The goal is to obtain the theoretical values ​​of the target variables under corresponding operating conditions during normal operation.

[0069] The variable values ​​for the theoretical operating condition are calculated, and all relevant setpoints are taken and correlated to obtain the pressure P' = (P1, P2). T Temperature T' = (T D T W ) T Composition X' = (X D X W ) T Liquid level L' = (L R L W ) T .

[0070] (3) Compare the predicted value with the theoretical target value to obtain the deviation of the predicted value;

[0071] The predicted values ​​and setpoint / model theoretical values ​​are compared, and the differences between the two sets of temperature, pressure, composition, and liquid level values ​​are calculated. The corresponding differences for each set of data include: temperature difference σ. T Pressure difference σ P The difference σ X Liquid level difference σ L Compare with the allowable error.

[0072] (4) The valves of the chemical tower equipment are used as actuators for regulation, and operation prompts are given for valve nodes without automatic control. A risk analysis table is generated.

[0073] The analysis of temperature, pressure, composition, and liquid level mainly considers the potential risks caused by these variables falling outside the error range, such as: abnormal over-temperature / under-temperature operation of the tower, abnormally high / low pressure in the chemical tower equipment, unqualified composition and concentration of materials at the top and bottom of the tower and the resulting changes in physical properties, and risks such as excessively high liquid level / low liquid level and dry burning. After identification, automatic adjustments are implemented. Variables for automatic adjustment include, but are not limited to, feed rate, feed temperature, and the consumption of various utilities. The execution target is the valve opening controlling each parameter, up to σ. T σ Pσ X σ L Within permissible limits, operational prompts and warnings may be provided, and reports may be submitted.

[0074] This method is applicable to nodes of chemical tower equipment, including feeding, tower separation (adsorption, stripping, extraction, absorption, evaporation, distillation), reflux, tower-related heating / condensation, and material delivery.

[0075] The above method will be explained below with specific examples.

[0076] Taking the pre-separation distillation column of a 480,000-ton / year methanol distillation unit in a certain industrial park as an example, the pressure at the top of the column is controlled by the inlet water flow of the water cooler. The actuator controls the opening of the water-cooled valve. Under this load, the control pressure at the top of the column is 20 kPG, the water-cooled valve opening is 80%, the flow rate is 1500 t / h, and the temperature is approximately 75.5℃; the feed flow rate is 120 m³ / h. 3 / h, 60℃, flow rate controlled by a simple loop; reboiler temperature 77℃, liquid level 60%, level controlled by a reboiler heat source. The subsequent unit capacity is expanded to 550,000 tons / year, increasing the feed flow rate to 138m³ / h. 3 / h, taking the operating data of the past 300 days, a neural network model was established and trained. When the feed increases, the opening of the water cooling valve changes by 80% to 100%, indicating that the pressure control is 25kPaG. At the same time, the process simulation was performed based on the design parameters of the tower and the flow rate after capacity expansion. The pressure control within 40kPaG (>25kPaG) will not cause damage or overpressure risk to the tower equipment and reflux tank equipment, while ensuring that the product quality is qualified and the quantity is sufficient.

[0077] After a period of monitoring, the scheme in the above embodiments is operating well, with no high-risk or large deviation conditions appearing. Even if there are some pressure fluctuations, the risk level is indicated to the operator as Level I, and normal operation is possible.

[0078] The solutions in the above embodiments of the present invention are established through data modeling, and through a neural network model of real-time data acquisition and analysis, as well as process simulation and verification, the predicted value and the set value / theoretical value calculated by the process are compared to inform the HAZOP deviation quantification status, and the tower pressure and equipment pressure conditions are monitored.

[0079] Existing HAZOP analysis models mainly involve empirical discussions of the process flow at different nodes by professional teams, without quantitative analysis. During production operation, it is impossible to obtain HAZOP deviation information in a timely manner to predict possible consequences.

[0080] The invention utilizes online data acquisition and data model building to perform parameter analysis under operating conditions and compare it with values ​​under standard conditions. It quantifies and analyzes HAZOP deviations in a timely manner, thereby accurately and specifically obtaining information that leads to device safety issues.

[0081] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An online HAZOP quantitative analysis and risk prediction method for chemical tower equipment, characterized in that, The method comprises the following steps: Collecting online data of a chemical tower device, training a neural network prediction model by using the online data, obtaining a trained neural network prediction model, and obtaining a predicted physical quantity of a target variable by using the trained neural network prediction model; The target variable is a reference value to be quantified in the HAZOP quantitative analysis process of a process, and the deviation between the measured parameter to be quantified and the theoretical value under the corresponding working condition; Determining a standard comparison value of the target variable, performing process simulation or twin digital model analysis based on the standard comparison value of the target variable, and obtaining a theoretical optimal target value under the standard comparison value of the corresponding working condition; Comparing the predicted physical quantity with the theoretical optimal target value to obtain a deviation of the predicted value and a HAZOP prompt, adjusting the chemical tower device based on the deviation of the predicted value and the HAZOP prompt, and generating a risk analysis table; Obtaining the trained neural network prediction model comprises the following steps: inputting the chemical tower equipment online data into the neural network prediction model to obtain X=(x1, x2, x3, …x n ) T ; Adding several neurons to the hidden layer in the neural network prediction model, the output targets of the neurons include pressure P=(P1, P2) T , temperature T=(T D , T W ) T , composition X=(X D , X W ) T ; based on the number of input parameters corresponding to the output targets of the neurons, the weight of the neuron connection weight is w ni , n=1, 2…n; the input parameters are any parameters in the online data of the chemical tower equipment; Obtaining the initial node number of the hidden layer based on the weight value, and obtaining the trained neural network prediction model; Wherein, the threshold value of the neuron is b i , and the activation value is u i ; The activation value is obtained by adding and weighting the input value and then subtracting the threshold value of the corresponding neuron, and the output value is a function of the activation value: The method for calculating the activation value is: wherein, is an activation value, is a connection weight for an input parameter, is a threshold value for a neuron, is an output, and f() is an activation function; The neural network prediction model determines the initial node number j of the hidden layer by an empirical formula or a cascade correlation method, wherein the empirical formula includes but is not limited to: Wherein, n1 and n2 are the node numbers of the input layer and the output layer respectively, and m is the sample number; The method for determining the standard comparison value of the target variable comprises determining by establishing a calculation model of a theoretical process or obtaining by an instrument setting value during device operation; Obtaining the theoretical optimal target value under the standard comparison value of the corresponding working condition comprises the following steps: The theoretical optimal target values of the target variables under the corresponding working conditions are obtained by establishing a model or using a commercial process simulation tool language or software, and all the set values are obtained and associated to obtain pressure P'= (P1, P2) T , temperature T'= (T D , T W ) T , composition X'= (X D , X W ) T , liquid level L'= (L R , L W ) T ; Comparing the predicted physical quantity with the theoretical optimal target value comprises the following steps: Subtracting the predicted physical quantity from the theoretical optimal target value to obtain a corresponding difference value of different data, and comparing the corresponding difference value with an allowable error; The online data of the chemical tower device includes feed data, tower top data and tower bottom data, and setting value parameters of all automatic control instruments; The feed data includes feed flow, feed temperature, feed pressure and feed valve opening degree; The tower top data includes tower top production amount, tower top pressure, tower top temperature, tower top condenser cold medium flow, tower top condenser cold medium temperature, tower top condenser valve opening degree, air cooling unit running number or power, vent valve opening degree, reflux tank liquid level, produced material composition and pressure control gas valve opening degree; The tower bottom data includes tower bottom production amount, tower bottom pressure, tower bottom temperature, tower kettle reboiler heat medium flow, tower kettle reboiler heat medium temperature, tower top condenser valve opening degree, produced material composition and tower bottom liquid level.

Images