Multi-fluid easily-crystallized component salt deposition characteristic risk assessment method based on data driving

Abstract

The invention discloses a data-driven risk assessment method for salt deposition characteristics of easily-crystallized components of a multi-fluid. The method comprises the following steps: firstly, collecting salt deposition related characteristic data of easily-crystallized components of the multi-fluid to construct a sample database, preprocessing the data to form a preprocessing database, and dividing the preprocessing database into a test set and a verification set according to a certain proportion; then constructing a particle swarm optimization algorithm to optimize the radial basis function neural network, and carrying out prediction analysis based on the particle swarm optimization algorithm to optimize the radial basis function neural network; comparing and analyzing a result obtained by prediction and analysis with actual data, and verifying the prediction precision and reliability of the established radial basis function neural network model through indexes such as a root mean square error and the like; and finally, evaluating the salt deposition risk of the easily-crystallized component of the multi-element fluid by using the radial basis function neural network model. According to the method, the salt deposition risk in the complex multi-fluid environment can be rapidly and quantitatively predicted, the equipment failure detection rate is increased, and safe, stable and long-period operation of a multi-fluid salt deposition high-risk equipment system is promoted.

Description

Technical Field

[0001] The present invention belongs to the field of salt corrosion failure and protection of multi-component fluids, and specifically relates to a data-driven risk assessment method for salting characteristics of easily crystallized components of multi-component fluids. Background Art

[0002] At present, the petroleum and petrochemical industry has a very important strategic position in my country's industrial system. At the same time, China is the world's largest oil importer. Heavy, inferior and high-sulfur crude oil are the main trends in the quality change of global crude oil. Processing heavy crude oil has become the main direction of the global refining industry. In the future, the demand for heavy crude oil in refineries will also increase. With the continuous increase in the processing volume of heavy crude oil, the problem of salt corrosion failure of refining equipment caused by high-sulfur, high-nitrogen and high-chlorine crude oil has become increasingly prominent.

[0003] In the petrochemical industry, failure and leakage of key equipment caused by the deposition of crystallizable phase particles is very common. The failure mechanism is complex, the failure forms are diverse, and the failure forms are obviously sudden and harmful, which seriously affects the industrial production efficiency and the long-term operation of equipment. Existing studies have shown that sulfur-containing, nitrogen-containing and chlorine-containing compounds in crude oil will react with hydrogen to produce corrosive gases NH3, HCl, and H2S. Gases such as NH3, HCl, and H2S are prone to crystallization reactions in the equipment to produce salts such as NH4Cl and NH4HS and deposit on the pipe wall, causing blockage, corrosion and other problems. The current research focus in China is mainly on the simulation evaluation and risk assessment of salting characteristics through numerical simulation, but the current research results still find it difficult to predict the salting characteristics and flow deposition laws of the crystallizable components of multi-element fluids.

[0004] The commonly used method in current research is numerical simulation, which only needs to be operated on a computer. Compared with the experimental method, the simulation method is powerful, the manpower, material and time costs are greatly reduced, and the operation difficulty is low. However, there are still limitations such as low model accuracy, parameter uncertainty, poor algorithm, and difficult model verification. In recent years, various prediction technologies based on artificial neural networks have been promoted and applied. By adopting appropriate machine learning algorithms and artificial neural networks, based on the input raw data, the relationship between the simulated output and the various influencing factors is trained, thereby achieving accurate prediction of the output. Neural networks have powerful nonlinear fitting capabilities, parallel computing capabilities and good generalization capabilities. Therefore, it is urgent to build a data-driven neural network model to effectively solve the current problem of difficult prediction and poor accuracy of the salt-forming properties of the crystallizable components of multi-element fluids. Summary of the invention

[0005] In order to solve the problems existing in the background technology, the present invention provides a data-driven risk assessment method for the salt-forming properties of easily crystallized components of multi-fluids, which can achieve rapid and accurate prediction of the salt deposition temperature of easily crystallized components in multi-fluids and perform risk assessment.

[0006] The technical solution adopted by the present invention is as follows, comprising the following steps:

[0007] Step S1: First, collect the working condition characteristic parameters of the multi-component fluid in the pipeline during the operation of the oil refining equipment, and use the working condition characteristic parameters collected in real time to construct the working condition characteristic vector X of the multi-component fluid. i Specifically, a set of working condition characteristic parameters can be obtained each time the data is collected, and a working condition characteristic vector is constructed using the set of working condition characteristic parameters collected this time;

[0008] Step S2: the processor preprocesses all the operating condition feature vectors to remove abnormal data, and uses the preprocessed operating condition feature vectors to establish an operating condition feature training data set;

[0009] Step S3, constructing a radial basis function neural network model for predicting the salt formation temperature of the easily crystallized component of a multi-element fluid, wherein the radial basis function neural network model is optimized using a particle swarm algorithm, and the radial basis function neural network model is trained using the operating condition characteristic training data set in step S2, and after the training is completed, the trained radial basis function neural network model is obtained;

[0010] Step S4, inputting the working condition characteristic parameters of the multi-component fluid to be detected into the trained radial basis function neural network model to obtain the salt formation temperature prediction result of the easy crystallization component of the multi-component fluid;

[0011] Step S5, the processor evaluates the salt formation risk of the easily crystallized components of the multi-fluid according to the salt formation temperature prediction result of the easily crystallized components of the multi-fluid, and adjusts the water injection amount in the pipeline of the oil refining equipment according to the salt formation risk assessment result.

[0012] In step S1, the working condition characteristic parameters of the multi-component fluid include the processing amount x i1 , sulfur content x i2 , chlorine content x i3 , nitrogen content x i4 , flow rate x i5 , process temperature x i6 and operating pressure x i7 There are seven operating characteristic parameters in total, and the operating characteristic vector X of the multi-element fluid i The expression is as follows:

[0013] X i =[x i1 ,···,x in ,···,x i7 ]

[0014] Among them, X i represents the characteristic vector of the i-th working condition; x in It represents the nth operating condition characteristic parameter of the multi-element fluid collected for the i-th time; i represents the ordinal number of the collected operating condition characteristic parameter.

[0015] The step S2 is specifically as follows:

[0016] Step S2.1: First, use the working condition characteristic vector X i Construct the working condition characteristic data set D:

[0017] D=[X1,···,X i ,,···,X M ]

[0018] Among them, X i represents the i-th operating condition feature vector; M represents the total number of operating condition feature vectors;

[0019] Step S2.2: Then, the anomaly detection value z(x) of all data in the working condition feature data set D is obtained by using the isolation forest algorithm. in ):

[0020]

[0021] Among them, z(x in ) represents the working condition characteristic parameter x in The corresponding abnormal detection value; l(x in ) represents the operating data x obtained by using the isolation forest algorithm in The corresponding path length; E(l(x in )) represents the path length l(x in ) is the expectation of all the working condition data x in the i-th working condition feature vector. in The average value of the corresponding path length;

[0022] Step S2.3: The abnormal detection value z(x in )∈(0.8,1) corresponds to the working condition characteristic parameter x in As abnormal data, delete the operating condition feature vector X with abnormal data. i , and use the remaining operating condition feature vector X i Establish a working condition characteristic training data set.

[0023] Specifically, all remaining operating condition feature vectors X i The data are combined into a working condition characteristic data set, and then divided into a training set and a validation set according to a preset ratio. The training set is used to train the radial basis function neural network model, and the validation set is used to verify the accuracy of the radial basis function neural network model.

[0024] The step S3 is specifically as follows:

[0025] The input of the radial basis neural network model is the operating condition feature vector X i , the output is the salt formation temperature y of the easily crystallized component of the multi-element fluid i , the salt formation temperature y of the easily crystallized component of the multi-element fluid i The expression is as follows:

[0026] y i =W i Z i

[0027] Among them, y i Represents the operating condition characteristic vector X i The corresponding multi-fluid easily crystallized component predicts the salt formation temperature y i ; Z i Represents the operating condition characteristic vector X i The corresponding hidden layer activation function; W i Represents the operating condition characteristic vector X i The corresponding connection weight of the i-th radial basis function;

[0028] In the above formula, the hidden layer activation function Z i According to the following formula:

[0029]

[0030] Among them, C i represents the central value of the i-th radial basis function; σ i represents the width of the i-th radial basis function; ‖‖ represents the norm, and exp() represents the exponential function with the natural number e as the base;

[0031] The parameters of the radial basis function include the center value C of the radial basis function i , width σ i and the connection weight W i , the parameters of the radial basis function are obtained by iterative optimization using the particle swarm algorithm. The specific process is as follows:

[0032] Step S3.1, using the parameters required to form the radial basis function to form the particles in the particle swarm algorithm, the number of particles is consistent with the number of radial basis functions (that is, the number of working condition characteristic vectors = M), randomly initialize the population, initialize the speed and position of each particle in the population, set the number of iterations, the population size to M, the inertia weight and the learning factor;

[0033] Step S3.2: Update the optimal position of each particle separately and the optimal location of the population

[0034] Step S3.3, update the speed and position of each particle according to the speed update formula and the position update formula respectively, and calculate the fitness value after the update;

[0035] Step S3.4, determine whether the number of iterations has been reached or whether the current fitness value has reached the preset value. If not, return to step S3.2. Otherwise, take the population obtained by the current iteration as the optimal population, and use the parameters of each particle in the optimal population as the parameters of the radial basis function.

[0036] The step S5 is specifically as follows:

[0037] Step S5.1: First, according to the output y of the radial basis neural network model during training i The root square error RMSE of the radial basis neural network model is obtained according to the following formula:

[0038]

[0039] Among them, h(X i ) represents the working condition feature vector X in the working condition feature training data set i The actual salt formation temperature; i Represents the working condition feature vector X in the working condition feature training data set i The corresponding multi-fluid easy-to-crystallize component predicts the salt formation temperature; M represents the total number of operating condition characteristic vectors;

[0040] Step S5.2: Based on the process temperature range (T1, T2) of the multi-component fluid, combined with the salt formation temperature y of the multi-component fluid crystallization component predicted in step S4, i , the salt formation risk of the easily crystallized components of multi-fluids is divided into three risk levels: high risk, medium risk and low risk:

[0041] If the crystallization temperature of the multi-component fluid is y i ∈(T1,T2), the salt formation risk of the easily crystallized components of the multi-component fluid under the current working conditions is considered to be high risk;

[0042] If the crystallization temperature of the multi-component fluid is y i ∈(T1-T1×RMSE,T2+T2×RMSE), and Therefore, the risk of salt formation of the crystallizable components of the multi-element fluid under the current working conditions is considered to be medium risk;

[0043] If the crystallization temperature of the multi-component fluid is Therefore, the risk of salt formation of the easily crystallized components of the multi-component fluid under the current operating conditions is considered to be low risk.

[0044] Step S5.3: Adjust the operating characteristic parameters of the refinery equipment pipeline according to the salt risk assessment result:

[0045] If the salt formation risk of the easily crystallized component of the multi-component fluid is determined to be high, increase the process temperature x i6 And increase the water injection volume in the pipeline;

[0046] If the salt formation risk of the easily crystallized component of the multi-component fluid is determined to be medium risk, the water injection volume in the pipeline is increased;

[0047] If the salt risk of the easily crystallized component of the multi-fluid is determined to be low, the current process temperature of the pipeline is maintained at x i6 and the current water volume in the pipe.

[0048] A working condition characteristic vector X i Corresponding to a current working condition.

[0049] The speed update formula and position update formula in the particle swarm algorithm are as follows:

[0050] Speed ​​update formula:

[0051]

[0052] Position update formula:

[0053]

[0054] in, represents the velocity vector of the dth dimension of the i-th particle in the t-th iteration; i represents the ordinal number of the particle; d represents the ordinal number of the particle dimension; ω represents the inertia weight; c1 represents the individual learning factor; c2 represents the group learning factor; r1 and r2 both represent a random number generated from [0,1]. represents the optimal position value of particle i in the dth dimension in the cumulative t iterations; represents the position vector of particle i in the dth dimension in the tth iteration; It represents the optimal position of the dth dimension of the group in the cumulative t iterations.

[0055] The neural network model of the present invention adopts a particle swarm algorithm to optimize the radial basis neural network neural network. The operating condition data set after data preprocessing is used as a database, combined with the machine learning and training of the neural network, to achieve the purpose of predicting the salt-forming characteristics of the easily crystallized components of the multi-fluid according to the actual operating conditions of the field equipment. The present invention can make full use of the actual operating condition data of multi-fluid equipment that has been accumulated in large quantities, and establish a corresponding equipment failure prediction method in combination with the neural network, which is of great significance for improving the accuracy of equipment failure detection.

[0056] Unlike traditional prediction methods based on experience or physical models, the RBF neural network optimized by the particle swarm algorithm is a data-driven model that can learn and discover potential patterns and relationships from historical data, so that it can not only perform well on training data, but also provide accurate predictions on new data that has not been seen. Compared with the BP neural network and the ordinary RBF neural network, the RBF neural network optimized by the particle swarm algorithm shows higher accuracy on the same task. This shows that this method has significant advantages in dealing with the prediction and risk assessment of the salt-forming properties of the crystallizable components of multi-fluids. The RBF neural network optimized by the particle swarm algorithm can comprehensively consider a variety of fluid properties and environmental factors, provide comprehensive salt-forming property prediction and risk assessment, and provide high-precision prediction results in a relatively short time, which is crucial for real-time monitoring and early warning systems. At the same time, it provides a quantitative evaluation standard for salt-forming risk, which helps decision makers take corresponding preventive measures according to the risk level and provides real-time support and decision-making assistance for field operations.

[0057] The beneficial effects of the present invention are:

[0058] 1. The prediction model based on neural network in the present invention can discover hidden patterns and relationships between data, identify data trends that may be missed by humans, and process various data formats, especially in dynamic, large-capacity and complex data environments. At the same time, it can reduce human intervention, optimize processes, improve efficiency and accuracy. Combined with neural networks, a corresponding risk assessment method for the salt-forming properties of easily crystallized components of multi-component fluids is established, which is of great significance for improving the accuracy of risk assessment of the salt-forming properties of easily crystallized components of multi-component fluids.

[0059] 2. Aiming at the problem of salt formation of easily crystallized components of multi-fluids in petrochemical process-type industrial equipment, the present invention establishes a method for predicting the salt formation characteristics and risk assessment of easily crystallized components of multi-fluids based on the particle swarm algorithm optimized radial basis neural network model. The method can quickly and quantitatively predict the salt formation risk in a complex multi-fluid environment, and can realize direct prediction of salt formation risk from online data of the industrial site, effectively improve the equipment failure detection rate, and promote the safe, stable, and long-term operation of multi-fluid equipment systems with high risk of salt deposition. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is the overall flow chart of the present invention.

[0061] Figure 2 is a diagram of the neural network structure used in the embodiment of the present invention.

[0062] Figure 3 This is a comparison chart of the prediction of the properties of the easily crystallized component aggregated salt according to the embodiment of the present invention.

[0063] Figure 4This is a comparison chart of the prediction of the properties of the agglomerated salt tested for the easily crystallized component in the embodiment of the present invention. DETAILED DESCRIPTION

[0064] The present invention is described in detail below in conjunction with specific implementation cases. The following implementation cases will help those skilled in the art to further understand the present invention, but will not limit the present invention in any form.

[0065] Embodiments of the present invention are as follows:

[0066] like Figure 1 The figure shows a logic schematic diagram of the method of the present invention. First, the data on the salt formation characteristics of the easily crystallized components of the multi-element fluid are collected to construct a sample database, and the preprocessed database is formed by preprocessing the data, and is divided into a test set and a validation set at a ratio of 8 to 2; then, a particle swarm algorithm is constructed to optimize the radial basis function neural network, and a prediction analysis based on the particle swarm algorithm to optimize the radial basis function neural network is carried out; further, the results obtained by the prediction analysis are compared with the actual data, and the root mean square error (RMSE) and the determination coefficient R are used to determine the optimal solution. 2 The prediction accuracy and reliability of the established particle swarm algorithm-optimized radial basis function neural network model were verified by using indicators such as swarm optimization. Finally, the salt formation risk of the easily crystallized components of the multi-fluid was evaluated.

[0067] For this embodiment, the established multi-fluid easy crystallization component salt risk characteristic database has a total of 256 data (collected from a heat exchanger of a hydrogenation reaction effluent system of a petroleum refining enterprise), including 205 data in the training set and 51 data in the validation set. i1 (kg / h), sulfur content x i2 (mg / kg), chlorine content x i3 (mg / kg), nitrogen content x i4 (mg / kg), flow rate x i5 (m / s), process temperature x i6 (℃), operating pressure x i7 (MPa) as input.

[0068] Step S1: construct the operating characteristic vector X of the multi-element fluid using the operating characteristic parameters i Specifically, each acquisition can obtain a set of operating condition characteristic parameters, and a set of operating condition characteristic parameters collected this time is used to construct an operating condition characteristic vector; the operating condition characteristic vector X of the multivariate fluid i The expression is as follows:

[0069] X i =[x i1 ,···,x in ,···,x i7 ]

[0070] Among them, Xi represents the characteristic vector of the i-th working condition; x in It represents the nth operating condition characteristic parameter of the multi-element fluid collected for the i-th time; i represents the ordinal number of the collected operating condition characteristic parameter.

[0071] Step S2: the processor preprocesses all the operating condition feature vectors to remove abnormal data, and uses the preprocessed operating condition feature vectors to establish an operating condition feature training data set;

[0072] The step S2 is specifically as follows:

[0073] Step S2.1: First, use the working condition characteristic vector X i Construct the working condition characteristic data set D:

[0074] D=[X1,···,X i ,,···,X M ]

[0075] Among them, X i represents the i-th operating condition feature vector; M represents the total number of operating condition feature vectors;

[0076] Step S2.2: Then, the anomaly detection value z(x) of all data in the working condition feature data set D is obtained by using the isolation forest algorithm. in ):

[0077]

[0078] Among them, z(x in ) represents the working condition characteristic parameter x in The corresponding abnormal detection value; l(x in ) represents the operating data x obtained by using the isolation forest algorithm in The corresponding path length; E(l(x in )) represents the path length l(x in ) is the expectation of all the working condition data x in the i-th working condition feature vector. in The average value of the corresponding path length;

[0079] In the specific implementation, the working condition feature data set D is input into the isolation forest algorithm to obtain the path length l(x in ), path length l(x in ) and calculate the anomaly detection value z of each data.

[0080] Step S2.3: The abnormal detection value z(x in )∈(0.8,1) corresponds to the working condition characteristic parameter x in As abnormal data, delete the operating condition feature vector X with abnormal data. i, and use the remaining operating condition feature vector X i Establish a working condition characteristic training data set.

[0081] Specifically, all remaining operating condition feature vectors X i The data are combined into a working condition characteristic data set, and then divided into a training set and a validation set in a ratio of 8:2. The radial basis function neural network model is trained using the training set, and the accuracy of the radial basis function neural network model is verified using the validation set.

[0082] Step S3, constructing a radial basis function neural network model for predicting the salt formation temperature of the easily crystallized component of the multi-element fluid, wherein the radial basis function neural network model is optimized by using a particle swarm algorithm, and the radial basis function neural network model is trained using the working condition characteristic training data set in step S2, and after the training is completed, the trained radial basis function neural network model is obtained;

[0083] In step S3, the operating condition characteristic training data set is input into the radial basis neural network model, and the mean square error (MSE) is used as the loss function to train the radial basis neural network model until the error of the radial basis neural network model reaches the minimum value and converges, thereby obtaining the trained radial basis neural network model. Figure 2 As shown, the radial basis neural network model includes an input layer, an output layer and a hidden layer; the input of the input layer is the operating condition feature vector X i The output of the output layer is the salt formation temperature y of the easily crystallized component of the multi-element fluid i The input and output ends of the hidden layer are connected to the input layer and the output layer respectively. Multiple independent parallel activation functions are set in the hidden layer, and the number of activation functions is equal to the working condition feature vector X i The number of activation functions makes each activation function associated with a working condition feature vector X i One-to-one correspondence, the i-th activation function Z i According to the i-th radial basis function, the i-th radial basis function and the i-th operating condition characteristic vector X i One to one correspondence.

[0084] Z i =R i (X i -C i )

[0085] Among them, R i () is the i-th radial basis function.

[0086] The input of the radial basis neural network model is the operating condition feature vector X i , the output is the salt formation temperature y of the easily crystallized component of the multi-element fluid i , the salt formation temperature y of the easily crystallized component of the multi-element fluid i The expression is as follows:

[0087] y i =W i Z i

[0088] Among them, y i Represents the operating condition characteristic vector X i The corresponding multi-fluid easily crystallized component predicts the salt formation temperature y i ; Z i Represents the operating condition characteristic vector X i The corresponding hidden layer activation function; W i Represents the operating condition characteristic vector X i The corresponding connection weight of the i-th radial basis function;

[0089] In the above formula, the hidden layer activation function Z i According to the following formula:

[0090]

[0091] Among them, C i represents the central value of the i-th radial basis function; σ i represents the width of the i-th radial basis function; ‖‖ represents the norm, and exp() represents the exponential function with the natural number e as the base;

[0092] The parameters of the radial basis function in step S3 include the center value C of the radial basis function i , width σ i and the connection weight W i , the parameters of the radial basis function are obtained by iterative optimization using the particle swarm algorithm. The specific process is as follows:

[0093] Step S3.1, using the parameters required to form the radial basis function to form the particles in the particle swarm algorithm, the number of particles is consistent with the number of radial basis functions (that is, the number of working condition characteristic vectors = M), randomly initialize the population, initialize the speed and position of each particle in the population, set the number of iterations, the population size to M, the inertia weight and the learning factor;

[0094] Step S3.2: Update the optimal position of each particle separately and the optimal location of the population

[0095] Step S3.3, update the speed and position of each particle according to the speed update formula and the position update formula respectively, and calculate the fitness value after the update;

[0096] Among them, the speed update formula and position update formula in the particle swarm algorithm are as follows:

[0097] Speed ​​update formula:

[0098]

[0099] Position update formula:

[0100]

[0101] in, represents the velocity vector of the dth dimension of the i-th particle in the t-th iteration; i represents the ordinal number of the particle; d represents the ordinal number of the particle dimension; ω represents the inertia weight; c1 represents the individual learning factor; c2 represents the group learning factor; r1 and r2 both represent a random number generated from [0,1]. represents the optimal position value of particle i in the dth dimension in the cumulative t iterations; represents the position vector of particle i in the dth dimension in the tth iteration; It represents the optimal position of the dth dimension of the group in the cumulative t iterations.

[0102] The fitness function formula used by the particle swarm algorithm in this embodiment adopts a conventional fitness function.

[0103] Step S3.4, determine whether the number of iterations has been reached or whether the current fitness value has reached the preset value. If not, return to step S3.2. Otherwise, take the population obtained by the current iteration as the optimal population, and use the parameters of each particle in the optimal population as the parameters of the radial basis function.

[0104] Here are three items from the total sample as examples:

[0105]

[0106] The parameters of the three radial basis functions are as follows:

[0107] Center value: [0.0394, 0.4867, 1.7790]; Width: [0.5428, 1.9967, 1.4999]; Connection weight: [0.6060, 0.8047, 0.7173]

[0108] In the specific implementation, the output of the trained radial basis neural network model is compared and analyzed with the actual data to obtain the root mean square error RMSE and determination coefficient R of the radial basis neural network model. 2 :

[0109]

[0110] Among them, h(X i ) represents the working condition feature vector X in the working condition feature training data set i The actual salt formation temperature; yi represents the operating condition feature vector X in the operating condition feature training data set iThe corresponding multi-fluid easy-to-crystallize component predicts the salt formation temperature; M represents the total number of operating condition characteristic vectors; Represents the working condition feature vector X in the working condition feature training data set i The actual average temperature.

[0111] The predicted values ​​of the training set and test set of the embodiment are compared and analyzed with the actual data samples, and the verification results are as follows Figure 3 and Figure 4 As shown, it can be seen that the predicted data of the present invention has a good consistency with the actual data samples, and the established neural network model has good prediction accuracy.

[0112] Step S4, inputting the working condition characteristic parameters of the multi-component fluid to be detected into the trained radial basis function neural network model to obtain the salt formation temperature prediction result of the easy crystallization component of the multi-component fluid;

[0113] Step S5, the processor evaluates the salt formation risk of the easily crystallized components of the multi-fluid according to the salt formation temperature prediction result of the easily crystallized components of the multi-fluid, and adjusts the water injection amount in the pipeline of the oil refining equipment according to the salt formation risk assessment result.

[0114] Based on the process temperature range (T1, T2) of the multi-component fluid, combined with the salt formation temperature y of the multi-component fluid crystallization component predicted in step S4 i , the salt formation risk of the easily crystallized components of multi-fluids is divided into three risk levels: high risk, medium risk and low risk:

[0115] If the crystallization temperature of the multi-component fluid is y i ∈(T1,T2), the salt formation risk of the easily crystallized components of the multi-component fluid under the current working conditions is considered to be high risk;

[0116] If the crystallization temperature of the multi-component fluid is y i ∈(T1-T1×RMSE,T2+T2×RMSE), and Therefore, the risk of salt formation of the crystallizable components of the multi-element fluid under the current working conditions is considered to be medium risk;

[0117] If the crystallization temperature of the multi-component fluid is Therefore, the risk of salt formation of the easily crystallized components of the multi-component fluid under the current operating conditions is considered to be low risk.

[0118] Finally, the operating characteristic parameters of the refinery equipment pipeline are adjusted according to the salt risk assessment results:

[0119] If the salt formation risk of the easily crystallized component of the multi-component fluid is determined to be high risk, the process temperature x should be adjusted before the water injection point. i6 Increase, and the process temperature x i6It should be greater than the larger value of the salt formation temperature and the water dew point temperature, and the water injection volume in the pipeline should be increased to make the effluent gas phase water saturated to avoid blockage and corrosion caused by salt deposition;

[0120] If the salt risk of the easily crystallized component of the multi-component fluid is determined to be medium risk, the water injection volume in the pipeline is increased to dissolve the possible salt in the water to avoid the deposition of salt;

[0121] If the salt risk of the easily crystallized component of the multi-fluid is determined to be low, the current process temperature of the pipeline is maintained at x i6 and the current water injection volume in the pipeline, and continuous attention should be paid to the process temperature and water injection volume, and salt risk assessment should be maintained at a certain frequency.

[0122] For this embodiment, the process temperature range is (180° C., 250° C.), and the predicted temperatures of the three examples are 197° C., 223° C., and 204° C. Therefore, the risk level of salt formation in the heat exchanger of the hydrogenation reaction effluent system of the three examples is high risk.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data-driven risk assessment method for salt formation characteristics of easily crystallized components of multi-element fluids, characterized in that: The following steps are involved: Step S1: First, collect the working condition characteristic parameters of the multi-component fluid in the pipeline during the operation of the oil refining equipment, and use the working condition characteristic parameters collected in real time to construct the working condition characteristic vector X of the multi-component fluid. i ; Step S2: the processor preprocesses all the operating condition feature vectors to remove abnormal data, and uses the preprocessed operating condition feature vectors to establish an operating condition feature training data set; Step S3, constructing a radial basis function neural network model for predicting the salt formation temperature of the easily crystallized component of a multi-element fluid, wherein the radial basis function neural network model is optimized using a particle swarm algorithm, and the radial basis function neural network model is trained using the operating condition characteristic training data set in step S2, and after the training is completed, the trained radial basis function neural network model is obtained; Step S4, inputting the working condition characteristic parameters of the multi-component fluid to be detected into the trained radial basis function neural network model to obtain the salt formation temperature prediction result of the easy crystallization component of the multi-component fluid; Step S5, the processor evaluates the salt formation risk of the easily crystallized components of the multi-fluid according to the salt formation temperature prediction result of the easily crystallized components of the multi-fluid, and adjusts the process temperature and water injection volume in the pipeline of the refinery equipment according to the salt formation risk assessment result.

2. A data-driven risk assessment method for salt formation characteristics of easily crystallized components of multi-element fluids according to claim 1, characterized in that: In step S1, the working condition characteristic parameters of the multi-component fluid include the processing amount x i1 , sulfur content X i2 , chlorine content x i3 , nitrogen content x i4 , flow rate x i5 , process temperature x i6 and operating pressure x i7 , the operating characteristic vector X of the multi-element fluid i The expression is as follows: X i =[x i1 ,…,x in ,…,x i7 ] Among them, X i represents the characteristic vector of the i-th working condition; x in It represents the nth operating condition characteristic parameter of the multi-element fluid collected for the i-th time; i represents the ordinal number of the collected operating condition characteristic parameter.

3. The data-driven risk assessment method for salt formation characteristics of easily crystallized components of multi-element fluids according to claim 1, characterized in that: The step S2 is specifically as follows: Step S2.1: First, use the working condition characteristic vector X i Construct the working condition characteristic data set D: d=[X1,…,X i ,,…,X M ] Among them, X i represents the i-th operating condition feature vector; M represents the total number of operating condition feature vectors; Step S2.2: Then, the abnormal detection value z(x) of all data in the working condition feature data set D is obtained by using the isolation forest algorithm. in ): Among them, z(x in ) represents the working condition characteristic parameter x in The corresponding abnormal detection value; l(x in ) represents the working condition data x in The corresponding path length; E(l(x in )) represents the path length l(x in ) is the expectation of all the working condition data x in the i-th working condition feature vector. in The average value of the corresponding path length; Step S2.3: The abnormal detection value z(x in )∈(0.8,1) corresponds to the working condition characteristic parameter x in As abnormal data, delete the operating condition feature vector X with abnormal data. i , and use the remaining operating condition feature vector X i Establish a working condition characteristic training data set.

4. The data-driven risk assessment method for salt formation characteristics of easily crystallized components of multi-element fluids according to claim 1, characterized in that: The step S3 is specifically as follows: The input of the radial basis neural network model is the operating condition feature vector X i , the output is the salt formation temperature y of the easily crystallized component of the multi-element fluid i , the salt formation temperature y of the easily crystallized component of the multi-element fluid i The expression is as follows: y i =W i Z i Among them, y i Represents the operating condition characteristic vector X i The corresponding multi-fluid easily crystallized component predicts the salt formation temperature y i ; Z i Represents the operating condition characteristic vector X i The corresponding hidden layer activation function; W i Represents the operating condition characteristic vector X i The corresponding connection weight of the i-th radial basis function; In the above formula, the hidden layer activation function Z i According to the following formula: Among them, C i represents the central value of the i-th radial basis function; σ i represents the width of the i-th radial basis function; || || represents the norm.

5. The data-driven risk assessment method for salt formation characteristics of easily crystallized components of multi-element fluids according to claim 1, characterized in that: The parameters of the radial basis function include the center value C of the radial basis function i , width σ i and connection weight W i , the parameters of the radial basis function are obtained by iterative optimization using the particle swarm algorithm. The specific process is as follows: Step S3.1, using the parameters required to form the radial basis function to form the particles in the particle swarm algorithm, randomly initialize the population, initialize the speed and position of each particle in the population, set the number of iterations, population size, inertia weight and learning factor; Step S3.2: Update the optimal position of each particle separately and the optimal location of the population Step S3.3, update the speed and position of each particle according to the speed update formula and the position update formula respectively, and calculate the fitness value after the update; Step S3.4, determine whether the number of iterations has been reached or whether the current fitness value has reached the preset value. If not, return to step S3.

2. Otherwise, take the population obtained by the current iteration as the optimal population, and use the parameters of each particle in the optimal population as the parameters of the radial basis function.

6. The data-driven risk assessment method for salt formation characteristics of easily crystallized components of multi-element fluids according to claim 1, characterized in that: The step S5 is specifically as follows: Step S5.1: First, according to the output y of the radial basis neural network model during training i The root square error RMSE of the radial basis neural network model is obtained according to the following formula: Among them, h(X i ) represents the working condition feature vector X in the working condition feature training data set i The actual salt formation temperature; i Represents the working condition feature vector X in the working condition feature training data set i The corresponding multi-fluid easy-to-crystallize component predicts the salt formation temperature; M represents the total number of operating condition characteristic vectors; Step S5.2: Based on the process temperature range (T1, T2) of the multi-component fluid, combined with the salt formation temperature y of the multi-component fluid crystallization component predicted in step S4, i , the salt formation risk of the easily crystallized components of multi-fluids is divided into three risk levels: high risk, medium risk and low risk: If the crystallization temperature of the multi-component fluid is y i ∈(T1, T2), the salt formation risk of the easily crystallized components of the multi-component fluid under the current working conditions is considered to be high risk; If the crystallization temperature of the multi-component fluid is y i ∈(T1-T1×RMSE,T2+T2×RMSE), and Therefore, the risk of salt formation of the crystallizable components of the multi-element fluid under the current working conditions is considered to be medium risk; If the crystallization temperature of the multi-component fluid is The salt formation risk of the easily crystallized components of the multi-component fluid under the current working conditions is considered to be low risk; Step S5.3: Adjust the operating characteristic parameters of the refinery equipment pipeline according to the salt risk assessment result: If the salt formation risk of the easily crystallized component of the multi-component fluid is determined to be high, increase the process temperature x i6 And increase the water injection volume in the pipeline; If the salt formation risk of the easily crystallized component of the multi-component fluid is determined to be medium risk, the water injection volume in the pipeline is increased; If the salt risk of the easily crystallized component of the multi-fluid is determined to be low, the current process temperature of the pipeline is maintained at x i6 and the current water volume in the pipe.

7. The data-driven risk assessment method for salt formation characteristics of easily crystallized components of multi-element fluids according to claim 5, characterized in that: The speed update formula and position update formula in the particle swarm algorithm are as follows: Speed ​​update formula: Position update formula: in, represents the velocity vector of the dth dimension of the i-th particle in the t-th iteration; i represents the ordinal number of the particle; d represents the ordinal number of the particle dimension; ω represents the inertia weight; c1 represents the individual learning factor; c2 represents the group learning factor; r1 and r2 both represent a random number, represents the optimal position value of particle i in the dth dimension in the cumulative t iterations; represents the position vector of particle i in the dth dimension in the tth iteration; It represents the optimal position of the dth dimension of the group in the cumulative t iterations.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images